EstablishedConfidence 99%
As of August 31, 2026, Google says the Search Console Generative AI performance report is rolled out worldwide and includes impressions from AI Overviews and AI Mode.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
EstablishedConfidence 99%
Q2D-Web contains 190 million web documents and 69,721 agent-reformulated queries in 10 languages, sampled from nine months of PII-free production search traffic.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
EstablishedConfidence 99%
Google Merchant Center AI performance insights compare a retailer’s share of voice with other brands across AI Mode and AI Overviews.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
EstablishedConfidence 99%
Perplexity Research describes CobbleDB as a decoupled architecture for AI-search storage that separates durable state, batched updates, and query-time retrieval to target lower latency and cost.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
EstablishedConfidence 98%
Google states there are no additional technical requirements to appear as a supporting link in AI Overviews or AI Mode beyond being indexed and eligible for a Search snippet.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
EstablishedConfidence 98%
Cloudflare announced personal-assistant traffic analytics and per-supported-agent Allow, Block on pages with ads, and Block controls on November 4, 2026.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
EstablishedConfidence 98%
Lighthouse v13.5.0 includes an Agent Resource Discovery schema conformance audit in its agentic-browsing tooling.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
EstablishedConfidence 98%
Bing states that content marked with data-nosnippet remains indexed and ranking eligible but is excluded from Bing Search snippets and Bing-powered AI-generated answer previews.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
EstablishedConfidence 98%
Bing states accurate sitemap lastmod timestamps help prioritize recrawling and reindexing, while IndexNow complements sitemaps with URL-level notifications; neither guarantees AI answer inclusion.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
EstablishedConfidence 98%
Cloudflare introduced an Agent Readiness score to assess website capabilities relevant to AI-agent interaction, including authentication, content access, agent-consumable formats, and payments.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
EstablishedConfidence 98%
In AI Mode and AI Overviews, content from a site selected by a user as a Preferred Source can be highlighted with a preferred badge for that user.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
EstablishedConfidence 98%
Google AI Mode and AI Overviews may use query fan-out to issue multiple related searches across subtopics and data sources.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Strong EvidenceConfidence 98%
In a preregistered field experiment with 1,100 participants, removing Google AI Overviews and AI Mode increased click-through rates to publishers, while an AI Mode-only experience reduced publisher click-through rates.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
EstablishedConfidence 96%
Cloudflare Disallow AI Training is designed to preserve search crawling by accountable mixed-use crawlers while expressing or enforcing a no-training preference, separating Search from Training controls.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Strong EvidenceConfidence 95%
GMMM defines a GEO media input as an expected count of generated answers in which a specified property occurs and is noticed, combining answer occurrence probabilities with relevant-question counts, generative-system shares, and notice probabilities.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Strong EvidenceConfidence 95%
In WANDR, the strongest of six evaluated production research systems at high effort reached 0.363 soft F1 and 0.133 hard F1, with incomplete discovery, missing enrichment, and incomplete evidence construction identified as major bottlenecks.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
EstablishedConfidence 95%
llms.txt is a proposal/convention, not an established web standard with proven universal ranking or citation effects.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
EstablishedConfidence 95%
Cloudflare Bot Preference Sync reflects configured Search, Agent, and Training bot preferences into a site’s robots.txt so declared preferences can stay aligned with Cloudflare bot-control settings.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
EstablishedConfidence 95%
Cloudflare announced a waitlisted Monetization Gateway designed to let automated clients pay for protected web pages, datasets, APIs and MCP tools through x402, with optional Web Bot Auth requirements.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
EstablishedConfidence 95%
OpenAI documents an in-product ChatGPT app directory for browsing and searching published apps and says it is experimenting with surfacing relevant apps directly in conversations using contextual, usage and preference signals.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
EstablishedConfidence 95%
OpenAI states that ChatGPT product discovery can receive structured product feeds and promotions through the Agentic Commerce Protocol and integrated catalog providers.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Strong EvidenceConfidence 94%
Google confirmed an early-stage pilot that pays participating publishers when their web content significantly contributes to AI answers in Gemini, AI Overviews, and AI Mode; post-generation links alone do not qualify.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Strong EvidenceConfidence 94%
Google Search VP Robby Stein confirmed that the reduced/absent citation behavior observed with Gemini 3.8 Flash in AI Mode was not working as intended and that a fix would be rolled out.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Strong EvidenceConfidence 93%
In Google Research experiments, Retrieve-for-Train outperformed single-query search, zero-shot expansion and a Best-of-N baseline across the evaluated set-valued retrieval tasks, while the distilled diffusion retriever was reported as 12–20× faster than autoregressive fan-out approaches in the tested setup.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Strong EvidenceConfidence 90%
In CITECHOICE controlled replays, structured rendering increased a target source's citation count after retrieval, while evidence that it increased the probability of being cited at all was inconclusive.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Strong EvidenceConfidence 88%
The original GEO paper reported that its tested optimization strategies could improve measured visibility by up to 40% in its benchmark setting.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 85%
In the GEO-SFE study, the authors report that structural feature engineering across six evaluated generative engines improved citation rate by 17.3% and subjective quality by 18.5% relative to their study baselines.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Strong EvidenceConfidence 82%
In the 2023–2024 audit, the authors classified more than 5% of C4 corpus tokens, and more than 28% of tokens from its most actively maintained critical sources, as fully restricted from use.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Strong EvidenceConfidence 82%
The Pinterest GEO paper reports that its production framework delivered a 20% increase in organic traffic after deployment across billions of images and tens of millions of collections.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Strong EvidenceConfidence 82%
In the DeepTRACE evaluation, citation accuracy ranged from roughly 40% to 80% across evaluated generative-search and deep-research systems, while deep-research configurations improved citation thoroughness but still contained substantial unsupported statements.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Strong EvidenceConfidence 82%
In the evaluated controlled experiments, generative AI search systems differed materially in source mix, freshness, domain diversity, cross-language stability and sensitivity to query phrasing, so visibility outcomes were engine-, language- and query-dependent.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Strong EvidenceConfidence 82%
In experiments reported by AgenticGEO across three datasets and two representative generative engines, its adaptive agentic framework outperformed 14 baselines and demonstrated cross-domain transferability.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Strong EvidenceConfidence 82%
In the Beyond the Vacuum experiments, competitor-aware combinatorial strategy selection outperformed the evaluated existing agentic and single-heuristic GEO baselines on GEO-Bench and the paper's synthetically augmented competitive dataset, with reported transfer to out-of-distribution datasets.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Strong EvidenceConfidence 82%
Under the paired, volume-controlled experiments reported in Scoring Without the Engine, quotation addition, statistics addition and cite-sources increased citation on none of the ten evaluated modern generative-engine families.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Strong EvidenceConfidence 82%
Cloudflare reports that as of June 2026, 52% of crawler requests it classified by purpose were for AI training, while mixed-use crawlers represented more than 36% of crawler activity.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Strong EvidenceConfidence 82%
A 2026 critical survey of 45 GEO studies finds that the evidence base spans distinct stages of generative-engine visibility and has methodological limitations for establishing durable, general cross-platform discoverability effects.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Strong EvidenceConfidence 82%
In the evaluated geo-citation-lab dataset, citation selection and citation absorption behave as distinct measurable outcomes, so citation breadth does not by itself establish equivalent answer-level influence.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Strong EvidenceConfidence 82%
In the study setting evaluated by Schulte, Bleeker and Kaufmann, AI-search visibility varied across repeated observations, making a single run an unreliable basis for estimating brand or source visibility.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Strong EvidenceConfidence 82%
In the repeated-sampling study reported by Sielinski (2026), single-run citation visibility measurements can give a misleadingly precise picture because citation distributions and domain rankings vary across repeated samples.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Strong EvidenceConfidence 82%
In the authors' experiments across seven GEO attacks and five evaluated LLMs, GEO Defender reduced average attack success rate from 50.32% to 6.20% while retaining 94.12% of benign-evidence use.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Strong EvidenceConfidence 82%
Longpre and coauthors audited robots.txt and Terms-of-Service data-use preferences across approximately 14,000 web domains associated with AI training corpora in a 2023–2024 longitudinal study.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Strong EvidenceConfidence 82%
In Counter-GEO-Bench, three evaluated off-the-shelf defenses reduce information-distorting GEO attack success by at most 5.7% relative, while the proposed C-GEO Guard baseline reports a 47.6% relative reduction with near-zero utility loss.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Strong EvidenceConfidence 82%
The audit reported Terms-of-Service crawling restrictions covering approximately 45% of C4 tokens and documented inconsistencies between sites' Terms of Service and robots.txt preferences.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 78%
In two practitioner-led GEO experiments covering 15 commercial-intent queries, third-party placements accounted for most observed citation visibility, while citation concentration and decay varied substantially by source and platform.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 65%
The VHop-Router paper reports 36% set success on a 50-query realistic transfer test and about 1% success on a harder four-hop extrapolation test.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 65%
In the reported VHop agentic-search comparison, replacing the retrieval tool increased task success by 52.7% and reduced average generated tokens by 61%, from 1,886 to 728.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 65%
On the authors' controlled VHop evaluation, the trained VHop-Router achieved 76.3% standalone retrieval success, compared with under 5% in the specified baseline configuration.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 65%
Chen, Zhou and Wu introduced VHop, a symbolically verified benchmark with five controlled difficulty levels for multi-hop visual evidence retrieval, and VHop-Router, which retrieves linked image chains without intermediate agent-generated text queries.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
publishedConfidence 0%
On October 28, 2026, Shopify announced Shop.dev and shop.md access to Shopify Catalog, sign-in, cart, checkout and Shop Pay, alongside WebMCP tools for agent interaction with storefronts and checkout.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
publishedConfidence 0%
On October 1, 2026, Cloudflare made AI Search generally available and documented native image embeddings, scanned-PDF OCR, a 10 MiB file limit, and hybrid vector/keyword retrieval with optional reranking.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
publishedConfidence 0%
In 1,000 ranking-style queries spanning ten consumer topics, Chen and colleagues measured mean cited-domain Jaccard overlap with Google top-ten results of 4.0% (GPT-4o), 11.1% (Gemini 2.5 Flash), 12.6% (Claude 4.5 Sonnet) and 15.2% (Perplexity Sonar Pro).
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Strong EvidenceConfidence 0%
In SAGEO Arena, a controlled research pipeline using 171,003 web documents and 2,700 queries, the authors report that body-text-only GEO modifications can reduce retrieval and reranking visibility while stage-aware use of structural fields and content improves overall visibility.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
On the 162-question MERRIN benchmark, the authors report mean accuracy of 22.3% across tested search-agent configurations and 40.1% for the strongest agentic multimodal search configuration; issuing more searches did not consistently improve accuracy.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
publishedConfidence 0%
On June 17, 2026, Google publicly announced Agentic Resource Discovery (ARD) as an open specification using publisher-controlled catalogs and federated registries for discovery and verification of agentic capabilities.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
publishedConfidence 0%
The ARD v0.91 proposal dated August 26, 2026 specifies JSON-LD agentic resource entries and requires conformant domain resolvers to fetch /.well-known/ard.json, while the earlier ai-catalog.json lookup is optional.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
In a controlled study of 12 LLM agent models across three domains, the less requirement-satisfying item was selected about two-thirds of the time when it came from a preferred source and the better item from a dispreferred source; source-name masking or relabeling also changed selection.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
In three controlled QA collections, SIEVE's fielded Boolean search–inspect–fetch workflow achieved higher accuracy than the strongest conventional Search–Visit configuration on each collection while using 20.7–50.6% fewer distinct tokens.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
In the authors' revised ITER preprint, interaction-aware retrieval improved on the LRAT baseline by average relative 6.9% on InfoSeek-Eval and 15.4% on BrowseComp-Plus across six agent backbones.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
publishedConfidence 0%
In Carragher and Carley's October 2026 study, sampled AI news citation distributions were 1.23–2.08 times as concentrated as estimated audience traffic at the outlet level across the reported AI-system comparisons.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
publishedConfidence 0%
In the WARP study's tested open deep-research agent and topic configurations, inserting as few as 13 words into a frequently retrieved UGC page led to the inserted text being retrieved in 57–76% of agent executions and cited in 38–51% of generated reports.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
In a 55,393-query longitudinal observational audit, the authors reported AI Overview activation of 13.7% overall and 64.7% for question-form queries; 29.8% of cited domains were absent from first-page results, and 11.0% of 98,020 assessed atomic claims lacked support in retrieved cited source text.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
In a randomized Chrome-extension field experiment, conditional on an AI Overview appearing, the AIO-exposed group generated 39.8% fewer outbound organic clicks and 34.5% more zero-click searches than the AIO-hidden group, according to the authors.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Strong EvidenceConfidence 0%
In controlled retrieval experiments, the authors report that a source pool with 67% synthetic-content contamination yielded over 80% exposure contamination; in a separate adversarial setting, BM25 exposed approximately 19% harmful content.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
EstablishedConfidence 0%
On October 21, 2026, Walmart announced Agent Profiles with per-agent approvals and spending controls, and an MCP-based Agent Gateway for third-party shopping agents, initially described as a developer/company beta.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
In Ma et al.'s historical Google search sample of approximately ten thousand websites, AI Overview cited sources exhibited lower measured language-model perplexity and greater semantic similarity than the compared conventional search results.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
In a six-month DeepResearchGym research-API log analysis of 14.44 million requests grouped into 3.97 million sessions, the authors report that 54% of newly introduced query terms appeared in accumulated earlier evidence, on average.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
The LRAT authors report constructing 26,482 agent search trajectories and 91,713 training query-document pairs from 10,000 InfoSeekQA questions using four retrievers.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
Under a controlled threat model permitting one attacker-controlled result per query, the Breadcrumbing Search Agents study reports 55.9% attack success and 83.3% MaxN attack success for its Authority-Chain Hijack strategy on the SafeSearch full test split.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
publishedConfidence 0%
In a simulated generative-search evaluation of 100 source samples and 500 held-out query instances, the RAID authors report 3.00% Objective Overall and 1.32% Subjective Average relative improvement scores for query-unavailable content rewriting, with 95% confidence intervals of 1.49–4.62% and 0.64–2.05%, respectively.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
publishedConfidence 0%
The SIRA authors report that an LLM-generated, corpus-frequency-validated single weighted BM25 query outperformed selected dense and multi-round retrieval-agent baselines across ten BEIR-style retrieval tasks and in their BrowseComp-Wikipedia test of 232 questions over a 25,587,229-document English Wikipedia index.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
In the PSA paper's shared-substrate evaluation on InfoSeek-Eval and BrowseComp-Plus across five policy backbones, PSA improved macro-averaged task success relative to a query-based agent by 4.00 and 7.56 percentage points, respectively; mean within-backbone final-step token reductions were 28.3% and 33.9%.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
The DCI paper reports that direct raw-corpus interaction using terminal-style operations outperformed selected sparse, dense and reranking baselines on several BRIGHT and BEIR tasks and achieved strong results on BrowseComp-Plus and multi-hop question answering without a conventional semantic retriever.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
In GEO-Flag's evaluated 10,095 usable pages from released Google Search and Gemini-grounded retrieval datasets, its detector flagged 898 pages (8.90%) as GEO-like; this is a classifier-based estimate without ground-truth GEO labels.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
In CHASE's 20-round fixed-ranker simulation across six domains, ranking-feature proximity became less aligned with independently judged document quality; Spearman rho changes ranged from -0.107 to -0.018, averaging -0.068.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
In LASAGNA's controlled study of 450 RAG configurations across eight benchmarks, three retrieval techniques whose individual gains predicted +47% improvement delivered only +17% when combined.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
The IndexAct authors report that persistent candidate-set refinement with separate passage inspection outperformed evaluated baselines across five agentic-search and multi-hop QA benchmarks, with higher evidence coverage and smaller average live context on BrowseComp-Plus versus terminal-based interfaces.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
In the controlled SafeGEO benchmark of 600 recommendation cases and 22 GEO attack variants, flawed targets gained up to 83.2 percentage points in top-three recommendation placement; evidence-breakdown mitigation reduced harmful promotion by up to 39.2 points in tested conditions.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
In the controlled FORGE evaluation of 12 LLMs using frozen retrieved web evidence, a single locally polluted page produced fake-product recommendation rates up to 27%, and replacing the top three pages raised the maximum reported fooled rate to 73.8%.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Strong EvidenceConfidence 0%
On August 3, 2026, the Interactive Advertising Bureau published an AI visibility framework defining Presence, Prominence, Portrayal and Persuasion and distinguishing directional from decision-grade measurement.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
In the PailitaoMMSearch2.0 authors' approximately 14-day controlled production A/B test, with approximately 10% experimental traffic, they report +17.90% item-page views, +27.04% transactions and +8.94% gross merchandise value versus production baselines (all reported p<0.05).
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Strong EvidenceConfidence 0%
In controlled ViDoRe v3 and BRIGHT retrieval evaluations, Esfandiarpoor and colleagues reported an average 8.7-point nDCG@10 improvement for agentic retrieval using the same embedding model, with 107.4-second versus 0.67-second average query latency and 764.1K input tokens per agentic query.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Strong EvidenceConfidence 0%
In the ATLAS-Bench local-search evaluation (1,200 queries, three regions, 14 configurations across 11 models), GPT-6-Luna scored 71.5% on constraint satisfaction among returned recommendations and 95.6% on factual accuracy; GLM-5.3 covered more satisfiable queries with at least one acceptable recommendation (85.3% versus GPT-6-Luna's 67.3%).
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
In Tannenbaum's observational audit of 15 fixed commercial prompts on June 6, 2026 across ChatGPT, Microsoft Copilot, Google and Perplexity, 589 citations mapped to 528 unique URLs and 356 domains; mean same-prompt pairwise exact-URL Jaccard overlap was 0.0079, and 84.9% of compared engine pairs shared no cited URL.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
In the authors' fixed 100-query BrowseComp-Plus evaluation with a 100,000-document corpus and gpt-5.4-mini, RISE and the direct-corpus-interaction baseline each achieved 78% judged accuracy, while reported mean agent costs were $0.28 and $1.10 per query respectively.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
publishedConfidence 0%
On October 7, 2026, Perplexity announced publicly available pplx-embed-v2-late multi-vector retrieval models at 0.6B and 9B scales, supporting text and image retrieval with a shared token-level embedding space.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
publishedConfidence 0%
The October 12, 2026 Glasseek preprint describes open deep-search agent training and evaluation with domain-level URL filtering and sandboxing to reduce benchmark contamination; the authors report 61.6% BrowseComp accuracy for their 9B agent under the stated harness.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
publishedConfidence 0%
The October 21, 2026 peer-reviewed PLOS ONE study How AI agents search reports approximately one million ASQ agentic search traces spanning three corpora, five agent variants and six retrieval settings, with controlled traces containing up to 186 retrieval calls after an initial query.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
publishedConfidence 0%
In controlled GEO-Bench experiments across three generative engines, Liu and Xu report that FeatGEO improves citation visibility versus token-level rewriting baselines while maintaining or improving content quality.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
publishedConfidence 0%
In the XBCP controlled benchmark, four deep-research agents and multilingual retrievers were tested on language-mismatched supporting evidence; the authors report degraded evidence recall and citation reliability, with some answer-accuracy degradation persisting even when gold evidence is provided.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
In the authors' controlled E-GEO product-ranking evaluation with 900 held-out queries, GLOBE achieved a 61.6% rank-improvement success rate versus 44.1% for the evaluated monolithic prompt-optimization baseline.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
In the study's offline WANDS and held-out US ESCI product-search evaluations, lexical-semantic hybrid retrieval and cross-encoder reranking significantly improved nDCG@10 over BM25, while pure dense retrieval showed no significant overall gain after Holm correction.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
In an offline re-ranking experiment on 1,887 unique query-product pairs drawn from 266 ESCI queries, the authors reported nDCG@10 of 0.9382 for a human-feature oracle, 0.9258 for a human-free Q+P configuration and 0.9150 for synthetic approximations.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
publishedConfidence 0%
In seven English and Russian fixed-corpus benchmarks with gold evidence annotations, the T-Search authors report mean Recall@10 of 56.0 for one rollout, 14.4 points above the base model, and 61.3 with three fused rollouts.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
The PailitaoMMSearch2.0 authors describe a deployed Taobao Pailitao system that generates product Semantic IDs directly from image-plus-text requests for similar, editing and complementary matching shopping intents.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
In the authors' RollingEval-Jun25 benchmark of 250 computer-science arXiv papers, vanilla scholarly API retrieval remained around or below 15% reference-list recall at very large candidate counts, whereas bibliography-expanding Deep Research variants exceeded 80% recall by K=10,000.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
The MC-Search authors constructed 3,333 annotated multimodal agentic-search tasks across five reasoning structures, with an average of approximately 3.79 retrieval hops and explicit intermediate evidence and modality annotations.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
publishedConfidence 0%
In SAGE's controlled DR Tulu corpus-search experiment at top-k=10, BM25 achieved 81.2 average short-form exact match versus 63.0 for gte-Qwen and 49.3 for ReasonIR; on open-ended weighted recall, gte-Qwen scored 33.0 versus BM25 at 30.7.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
publishedConfidence 0%
In INSPIRE's 476-target scholarly search evaluation, the strongest tested agent achieved 0.284 nDCG@10, and the rate of targets where no graded antecedent was exposed ranged from 17.4% to 56.1% across agents; the largest observed deficit was before source selection.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
publishedConfidence 0%
In a difference-in-differences analysis of 105,012 Reddit communities and 60,696,936 subreddit-day observations, Zhang, Cui and Zhang estimate 12.0% higher daily comments and 12.4% higher daily unique commenting authors for Google AI Overview-eligible communities relative to ineligible controls after the August 15, 2024 international rollout.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
publishedConfidence 0%
In the same observational study, the authors report that the engagement advantage for experience-based Reddit communities is substantially attenuated after the introduction of Google's conversational AI Mode relative to the earlier AI Overviews interface.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
In EComAgentBench's 662 valid shopping tasks, the seven evaluated model configurations achieved overall task accuracy ranging from 19.5% to 57.1%; Claude Opus 4.6 had the highest reported accuracy (57.1%) while GPT-5.4 had the highest rubric-satisfaction score (81.0%).
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
Across all seven models evaluated by EComAgentBench, satisfaction of query-visible shopping constraints was higher than satisfaction of requirements stored in a tool-accessible profile or revealed through scripted clarification; for GPT-5.4 the reported rates were 88.1%, 69.8% and 70.9%, respectively.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
publishedConfidence 0%
Kakimov et al. (2026) propose an observational framework that measures retrieval and citation separately over query-document pairs and conditions citation audits on retrieval rank and document provenance.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
publishedConfidence 0%
In the authors' Google AI Overviews YMYL sample, documents classified as AI-generated were cited more frequently than human-authored documents after controlling for observed retrieval rank; the reported difference was concentrated among citations outside the observed retrieval results.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
publishedConfidence 0%
In Sünkler et al.'s 900 English-language Google US health queries collected February 2–4, 2026, AI Overviews appeared for 87.7% of queries, while the average observed domain-level overlap between AI-cited sources and top-ten organic results was 46.2%.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
publishedConfidence 0%
In the same Google US health-query sample, YouTube accounted for 6.8% of observed AI Overview citations versus 0.6% of top-ten organic results, according to the authors' domain-share table.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
publishedConfidence 0%
In TREC Deep Learning 2019 experiments, Huo and colleagues reported that retrieval pipeline and exploration policy both materially affected high-recall performance; query reformulation with Move-to-Front achieved the highest recall in their reported policy comparisons, with judgment-cost tradeoffs.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
publishedConfidence 0%
SearchHorizonBench reports 45,632 distinct executions and 8,880 forecasts across 3,168 questions, using paired SEARCH and THINK continuations from the same agent state.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
publishedConfidence 0%
On 1,632 held-out questions, SearchHorizonBench reports a 16.2% reduction in Brier error after historical calibration, while tested historical estimators explain less than 2% of question-specific search-gain variance.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
The SEEDS research framework uses lightweight retrieved-result previews and the agent's reasoning state to choose an arbitrary, potentially non-prefix subset of candidates for full inspection, rather than admitting only the top-ranked prefix.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
In the SEEDS paper's QwQ-32B live-web evaluation across BrowseComp, BrowseComp-zh and HLE, Seeds-o1 reported 11.78% mean accuracy and 8.30 full web-page reads per question, versus 9.28% accuracy and 27.81 reads for Search-o1.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
In Wadi and Ma's controlled Tool-Lab shopping experiments, charging for product-attribute inspection under vague goals led most tested LLMs to inspect fewer diagnostic attributes and sometimes select products with inferior unit prices.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
In the authors' complete-information validation, seven of eight LLMs selected the lowest-unit-price alternative in 99.7% of Study 1 and 98.6% of Study 2 sessions; in truncated-information validation, those models matched a conditional-information normative policy in 99.3% of sessions.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
AgentSearchBench reports 9,759 real-world agents, 3,211 benchmark tasks (2,452 single-agent queries, 500 multi-agent queries and 259 high-level task descriptions), and 66,740 agent executions.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
On AgentSearchBench high-level task descriptions, BGE-Large v1.5 achieved nDCG@5 of 23.08 versus BM25 at 16.35, while on executable task queries ToolRet achieved 37.52 versus BM25 at 32.41. The study reports that agent descriptions alone do not reliably reflect execution performance.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
In McKinsey's June 2026 11-market Agentic Commerce Consumer Survey, 53% of respondents in the weighted shopping-adoption analysis reported using AI chat/search or retailer AI assistants for at least one shopping task in the previous three months (n=9,988).
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
Among AI shoppers describing their most recent purchase in McKinsey's June 2026 survey, AI users were 3.2 times more likely to rank AI search as the biggest influence on their final decision than traditional search engines (weighted 11-market recent-purchase base n=5,095).
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Strong EvidenceConfidence 0%
OpenAI's October 1, 2026 ChatGPT release notes announce a Try on button on clothing and accessory product listings, allowing users to generate a virtual try-on using ChatGPT Images and a selfie.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Strong EvidenceConfidence 0%
OpenAI's October 1, 2026 ChatGPT release notes announce that users can save shopping products to Favorites or organize them in folders in the ChatGPT Library, available on ChatGPT web and mobile.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
The DailyReport paper reports a benchmark of 150 open-ended daily search tasks, 3,546 associated rubrics, and evaluation of 17 agentic system configurations.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
Across the 17 tested agentic system configurations, the DailyReport authors report stronger instruction-following than factuality and reasoning performance, with limited user-preference scores under their cascade rubric.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
In the authors' evaluated deep-research benchmarks, retrievers trained with LRAT improved evidence recall and end-to-end task success across tested agent architectures compared with their baselines.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
In the authors' controlled continued-pretraining comparison against five objectives, xPRT averages a 3.2% gain over the strongest baseline across 41 MTEB English v2 tasks and a 6.1% nDCG@10 gain across ten retrieval tasks.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
On 300 HotpotQA questions, the Glance authors report that its selected context improves answer F1 by 0.045 versus cross-encoder-selected context under the same base generator (paired bootstrap 95% CI 0.014 to 0.076).
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
The xPRT authors introduce a self-supervised retriever pretraining objective in which representative document passages are trained to predict and align with related passages from the same document.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
Metinov and Arziev describe Glance, which scores retrieved passages with passage-specific relevance probes using a shared generator backbone and a lightweight decision adapter instead of serving a separate reranking backbone.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
publishedConfidence 0%
SGR-Bench comprises 100 expert-curated state-gated retrieval tasks spanning 12 public data ecosystems and six source families, and evaluates eight CLI-based agentic systems plus three commercial search-agent products.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
publishedConfidence 0%
In an audit of 156 analyzable failed CLI search-agent trajectories, the SGR-Bench authors classified 37.2% as retrieval-scope drift, 27.6% as criterion mismatch, and 10.3% as final answer composition errors.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
publishedConfidence 0%
On the 100-task SGR-Bench, the authors report overall item-level F1 scores of 14.87% for Google Search AI Mode, 29.93% for Gemini Deep Research, and 54.20% for OpenAI Deep Research.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
publishedConfidence 0%
Hirsch et al. evaluated three open-source deep-research systems with a local diagnostic method that distinguishes hallucination, uncited input reliance, uncited output, and insufficient citation errors.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
publishedConfidence 0%
In the paper's AI-Q evaluation, 84.7% of final-report errors were attributed to the orchestrator, approximately 31% of its errors were hallucinations, and two tested interventions improved citation recall by 5% without reducing output quality.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
publishedConfidence 0%
Jinadu et al. held paper titles and abstracts constant while varying author, venue and citation metadata across original, flipped and boosted conditions for eight LLMs (five open-weight, three closed-weight) in single-turn top-one paper recommendation.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
publishedConfidence 0%
In Jinadu et al.'s tested academic recommendation setting, authority metadata produced directional, model-dependent choice bias; debiasing instructions suppressed explicit authority mentions more than they prevented authority-driven recommendation flips.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
publishedConfidence 0%
CoSearch jointly trains a multi-step reasoning agent and a generative document ranker using GRPO, while its first-stage dense retriever remains fixed.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
publishedConfidence 0%
In seven QA benchmarks, CoSearch authors report +6.6% relative F1 over Search-R1 for a Qwen2.5 7B reasoning agent and +10.8% for a 3B agent; these are relative gains, not percentage points.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
publishedConfidence 0%
NASA-EO-Bench, derived from NASA Earth Observation Knowledge Graph, contains approximately 47,000 query–dataset pairs including approximately 21,000 task-based queries.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
publishedConfidence 0%
In the revised v2 evaluation, the authors report that a zero-shot agentic reranking stage improves MRR by 16% over the supervised retrieval pipeline, with paired-bootstrap significance testing on an expanded 600-query reranking evaluation.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
publishedConfidence 0%
In March 2026, Zhu and Chang ran 156 English- and Japanese-language Tokyo hotel prompts on Gemini 2.5 Flash with Google Search grounding and classified 1,357 grounding citations into OTA and non-OTA source categories.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
publishedConfidence 0%
In that Tokyo Gemini hotel audit, 55.9% of grounding citations for experiential queries came from non-OTA sources, versus 30.8% for transactional queries, a 25.1 percentage-point difference.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
Shopping Companion Bench comprises 1,000 author-constructed shopping tasks over a corpus of 1,298,797 Lazada products, with single-product and add-on-deal tasks and preference histories checked by two e-commerce experts.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
The Shopping Companion authors report that their GPT-5 zero-shot baseline achieved under 70% task success on their benchmark, and that tool-wise reward training improved their 4B agent versus their tested training baselines.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
Volpini and colleagues evaluated seven standard and agentic RAG conditions across editorial, legal, travel and e-commerce domains, comparing plain HTML, JSON-LD and enhanced linked entity pages.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
In the authors' controlled RAG experiments, enhanced entity pages yielded reported accuracy improvements of 29.6% in standard RAG and 29.8% in the full agentic pipeline; the incremental Enhanced+ gain was not statistically significant.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
Zhang (2026) compared six retrieval structures on 345 AppWorld queries with expert dependency labels and 103 MCP-Bench tasks spanning 257 MCP tools.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
In Zhang's ToolRet stratification, completeness@10 was 0.58–0.67 for single-target queries versus 0.08–0.17 for queries needing at least four tools.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
In a 60-task downstream AppWorld check, the tested agent succeeded on 21 of 21 tasks with complete retrieved tool sets versus 2 of 39 tasks with incomplete sets.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
EstablishedConfidence 0%
Shopify announced a merchant-admin Catalog eligibility view and a Knowledge base in Sales channels > Agentic for publishing non-product store information used by AI shopping agents.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
EstablishedConfidence 0%
Shopify documents that published Knowledge base facts are accessible through Storefront MCP; storefront-scanned facts are unpublished until merchants publish them, while applicable Shopify-admin-derived facts can be published when created.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
EstablishedConfidence 0%
Shopify's Knowledge base lets merchants publish or unpublish individual information fields and attach reference URLs that are available to AI agents alongside the corresponding facts.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
In ToolBench G1 single-tool queries (n=200, k=5), the authors report Retrieve-then-Plan recall of 0.372 versus Plan-then-Retrieve recall of 0.289 ± 0.010.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
In ToolBench G2 intra-category multi-tool queries (n=200, k=5), the authors report Plan-then-Retrieve recall of 0.188 ± 0.007 versus Retrieve-then-Plan recall of 0.143 (p < 0.001); the G3 recall advantage was not individually statistically significant.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
On ToolRet (7,961 queries, 44,453 tools), Arcade.dev researchers report Completeness@10 of 45.2% for Tool Recommendation versus 28.8% for BM25 within their own evaluation harness (57% relative improvement).
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
Arcade.dev researchers describe a gateway exposing roughly 8,200 catalog tools through three agent-facing interfaces, while filtering recommendations by eligible tool scope and user authorization.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
In Arcade.dev's paired seeded-workflow evaluation against a static 100-tool list, the authors report 13.6% lower total billed cost overall across Codex and Claude Code clients, with no statistically significant difference in task completion.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
Banerjee et al. introduce evidence-based generation with line-level citations and an Ans+Evid metric that jointly evaluates answer correctness and grounding.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
Across eight proprietary and three open-weight LLMs in the authors' evaluations, similarity-based chunk retrieval more readily surfaced some relevant evidence than all passages required for multi-fact questions.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
The EBG evaluation identifies loss of document-level metadata from chunking and incomplete citations that answer-only metrics can overlook.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
EvidenT verifies extracted passages against retrieved documents using structured passage extraction and deterministic lexical alignment before answer generation, without model retraining.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
On approximately 500 real enterprise queries, EvidenT reports an average 29% improvement in gold-source hit rate relative to prompting baselines.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
In the authors' enterprise evaluation, EvidenT produced no citations to URLs absent from retrieved evidence and achieved near-saturated answer-to-source lexical coverage.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
Soudani and colleagues analyze seven deep-research agent architectures and report that many later reasoning and search iterations yield little or no improvement in intermediate retrieval recall or final-answer performance.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
RAAC uses unsupervised signals for document novelty, successive query diversity, and original-question coverage to choose whether a deep-research agent continues, redirects, or stops searching.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
On BrowseComp-Plus, Soudani et al. report RAAC reduces search calls by 14 on average across evaluated agents (up to 44 fewer), with answer-accuracy gains up to 10% and 3% on average as stated by the authors.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
MCite-RL combines iterative image retrieval and cropping with process-level and outcome-level citation rewards to train multimodal RAG for both answer accuracy and localized visual evidence attribution.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
Zhao et al. report average gains of 26.16% in visual citation intersection-over-union and 5.89% in answer accuracy relative to a vanilla RAG baseline across their evaluated benchmarks.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
publishedConfidence 0%
ProductResearch uses User, Research and Supervisor agents to generate and refine shopping research trajectories combining open-web tools with product-catalog retrieval.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
publishedConfidence 0%
In the authors' RACE evaluation, a Qwen3-30B-A3B model fine-tuned on ProductResearch trajectories scored 45.40 compared with 31.78 for its base model.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
publishedConfidence 0%
The ProductResearch paper reports effective distinct product coverage rising from 3.58 to 12.45 between the base and fine-tuned Qwen3-30B-A3B models.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
In the Wiki-VISA 7B ablation, full MCite-RL scored 60.00% on answers and 36.05% on citation precision, compared with 54.20% and 20.56% when all visual citation rewards were removed.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
Ming et al. introduce LLM-Wiki, which compiles documents into structured, bidirectionally linked Wiki pages and exposes search/read tools for iterative agent traversal; its persistent Error Book records and repairs recurring compilation errors.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
In author-reported tests of 500 questions each from HotpotQA, MuSiQue and 2WikiMultiHopQA, LLM-Wiki achieves F1 scores of 0.839, 0.739 and 0.911 versus LightRAG's 0.819, 0.659 and 0.847, respectively.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
In the authors' ablation, removing progressive traversal reduces LLM-Wiki F1 to 0.722, 0.601 and 0.789 on HotpotQA, MuSiQue and 2WikiMultiHopQA, while removing the Error Book reduces F1 to 0.801, 0.699 and 0.877.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
Walters, Chen and Toni evaluated 225 post-cutoff changed facts across six domains using one proprietary search-enabled model and six open-weight model configurations under controlled retrieval.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
In the study's pooled frozen-retrieval evaluation, 6.4% of snippet-only episodes (87/1366) produced stale answers versus 1.3% of episodes citing an opened page (8/632); this is an observational association, not a causal estimate.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
In the same study, 6.4% of graded snippet-only answers (62/962) had unfaithful citations versus 2.3% of graded opened-and-cited answers (29/1258).
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
Nian and colleagues use a locally served 31B open-weight language model to evolve semantic-index programs and a 1M-parameter proxy decoder to evaluate candidates for a 1B target generative retriever.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
On NQ320K, the evolved semantic-index and collision-handling program increased reported Recall@1 from 55.82% to 59.28% and Recall@100 from 85.82% to 90.57% against the study's seed generative retriever.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
Across 21 semantic-identifier programs on NQ320K, the paper reports Spearman correlation 0.90 between proxy and target Recall@100, while noting weaker transfer when evolving training objectives or the entire pipeline.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
SearchOS-V1 formulates long-horizon information seeking as relational schema completion and records each populated value with a source URL and anchored evidence excerpt.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
In the authors' Max@3 WideSearch evaluation, SearchOS achieved 80.3 item-level F1 versus 76.0 for the strongest evaluated baseline, A-MapReduce (+4.3 points).
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
In the authors' Max@3 GISA set-question evaluation, SearchOS reached 76.5 F1 versus 63.1 for the strongest evaluated baseline (+13.4 points).
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
A SearchOS hierarchical-skills ablation on 100 WideSearch questions reported 36.6% lower session time, 39.1% fewer search calls and 42.7% fewer page calls with skills enabled.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
On 6,808 HotpotQA examples, the paper reports that an intervention increased strict correct-citation rate from 0.001 to 0.024; absolute correct-citation performance remained low.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
In a controlled PopQA-derived RAG-style experiment with Llama-3.1-8B-Instruct, van Dort and Heuss identified a distributed set of attention heads and MLP components influencing inline citation-marker emission through activation patching.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
In targeted PopQA intervention tests, amplifying selected citation-related components recovered over 90% of tested missed citations, while suppressing selected components removed 9 of 13 spurious citations (69%).
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Strong EvidenceConfidence 0%
Yang's analysis of March–May 2025 AI Search Arena data extracted 366,087 citation URLs from 24,069 conversations; 32,865 citations (9.0%) were classified as news-domain citations.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Strong EvidenceConfidence 0%
Among news-domain citations in the 2025 Arena sample, the 20 most-cited news domains accounted for 67.3% of OpenAI-family citations, 31.9% of Google-family citations, and 28.5% of Perplexity-family citations.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Strong EvidenceConfidence 0%
In 1,534 judged AI Search Arena head-to-head comparisons where both responses cited news sources, the study did not find statistically significant associations between user preference and the measured news-source political-leaning or quality variables after its specified controls.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
The Agentic-R paper proposes training a search retriever using both LLM-scored relevance of candidate passages to intermediate agent queries and final-answer correctness, followed by alternating agent and retriever optimization.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
In the authors' seven-benchmark evaluation with their trained search agent, Agentic-R achieved a 45.00 average exact-match score, compared with 41.78 for the strongest evaluated baseline REPLUG.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
In the authors' Table 1 cross-agent evaluations, Agentic-R achieved average exact-match scores of 43.64 on R1-Searcher and 39.43 on SimpleDeepSearcher, exceeding the best listed comparison scores of 41.67 and 37.69 respectively.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Strong EvidenceConfidence 0%
AURA trains retrieval-oriented embeddings using intermediate reasoning-state specificity against relevant documents and hard distractors, and uses a learned utility estimator to choose when to stop refinement.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Strong EvidenceConfidence 0%
On 12 BRIGHT retrieval domains, Luo et al. report average nDCG@10 of 34.7 for AURA-4B versus 32.1 for DIVER-Retriever-4B with external GPT-4 query expansion.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Strong EvidenceConfidence 0%
Under matched same-backbone inference settings, AURA utility-guided stopping reduced average reasoning latency from 4.60 to 0.76 seconds and generated tokens from 332 to 96 while improving BRIGHT nDCG@10 from 33.0 to 34.7.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Strong EvidenceConfidence 0%
LoRE combines knowledge-augmented query understanding, query-relevant visual captioning, explicit e-commerce rules, supervised fine-tuning and reinforcement learning for four-level query-product relevance assessment.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Strong EvidenceConfidence 0%
The LoRE authors report 29.1% higher Macro-F1 than their GPT-5 baseline on the RAIR e-commerce relevance benchmark.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Strong EvidenceConfidence 0%
The LoRE paper reports a cumulative relative 27.0% GoodRate improvement in Taobao online A/B tests across cache deployment (+4.8%), knowledge distillation (+9.5%) and system/strategy updates (+12.7%).
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
In the matched synthetic corpus, the primary gpt-5.4 agent published a false value for 71% of 90 prior-clean claims with one false page, versus 83% with a correlated five-page cluster and 82% with apparent-independent false pages.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
Across eight evaluated model settings from five providers, correlated false pages raised elicited credence compared with a single false page; reported Dependence Discount Index values ranged from 0.71 to 1.01.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
A masked-skeleton provenance gate reduced correlated clusters to 1.03 effective sources on average while retaining 4.94 of five apparent-independent control sources; a separate natural-news provenance classification test reported precision 0.949 and recall 0.359.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
MiRAGE defines InfoF1 to assess factuality and information coverage and CiteF1 to assess citation support and completeness using atomic subclaims, with reference-based and source-collection-based variants.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
Across three evaluated datasets, human-computed MiRAGE metrics aligned with human quality judgments and offered claim-level grounding assessment beyond evaluated text-centric and sentence-level RAG metrics.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
The authors report that automated MiRAGE variants did not match human-judgment agreement on audiovisual sources because video-grounded verification remained weak; refusal assessment and multi-video inference are additional limitations.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Strong EvidenceConfidence 0%
The SPRA authors report a scholarly resource covering 1,059,112 AI papers, 19,949,289 core units and 24,073,381 typed internal relations.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Strong EvidenceConfidence 0%
In SPRA's controlled source-conditioned scientific QA evaluation, Schema-RAG scored 2.342/5 versus 1.961/5 for Text-RAG and 2.420/5 for Full Compact on the G-U1 measure.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Strong EvidenceConfidence 0%
Across 3,070 paired SPRA RQ2 tasks, Schema-RAG used 28.5% fewer input-plus-output tokens than Text-RAG, while Top-5 synthesis grounding and qualification scores were 0.145/5 and 0.325/5 lower than Text-RAG.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Strong EvidenceConfidence 0%
ReCite combines citation-location detection, intent-aware retrieval-query planning and reflective claim-evidence verification that can trigger a revised retrieval query after a mismatch.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Strong EvidenceConfidence 0%
The ReCite study constructed data from 10,893 computer-science LaTeX papers and used an end-to-end evaluation set of 200 paragraphs containing 1,023 unique target test papers disjoint from training decisions.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Strong EvidenceConfidence 0%
The ReCite authors report a best Strict citation F1 of 39.15% for their CAP-8 configuration and a Lenient F1 of 55.14% on their end-to-end academic citation benchmark.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
publishedConfidence 0%
In a cohort of 31,142,353 Ctrip users logged in during July 10–24, 2025, 1,904,368 (6.1%) had initiated at least one Wendao AI chat by July 10, 2025.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
publishedConfidence 0%
Among Ctrip shopping journeys in the study containing both AI chat and conventional search, the most common sequence was interleaving, accounting for 53% of those journeys.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
publishedConfidence 0%
Attraction-related questions represented 42% of observed Wendao AI chat requests in the Ctrip study.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
publishedConfidence 0%
In the Ctrip cohort, Wendao adoption was 10.7% for users aged 50 or older and 4.6% for users under 24.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
Eslami and coauthors introduce pplx-embed-v2-context in 9B and 0.6B variants, training contextual chunk embeddings with query-aware graded relevance and evaluating answer and verification-evidence retrieval using Context-bench and Q2D-Chunk.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
EstablishedConfidence 0%
Cloudflare reported on October 21, 2026 that nearly 100,000 Commerce.com storefront domains had WebMCP shopping tools enabled through its edge integration, with per-merchant opt-out controls.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
publishedConfidence 0%
The ALCE benchmark evaluates citation quality separately from answer correctness and fluency in retrieval-grounded LLM answers.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
publishedConfidence 0%
The E-GEO preprint introduces a benchmark containing over 7,000 consumer product queries paired with listings and reports evaluation of 15 rewriting heuristics.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
publishedConfidence 0%
On July 29, 2026 Google announced that Search Console platform properties for measuring social and video content performance on Google Search, Discover and Google News were globally available.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
publishedConfidence 0%
OpenAI announced Business Hub with eligible business claiming, product feeds and aggregated mentions, clicks and cited-topic insights; Business Agents initially serve eligible US ecommerce and online service businesses.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
publishedConfidence 0%
On November 13, 2026, OpenAI announced ChatGPT Wallet for an initial rollout through dots to eligible adult US ChatGPT Pro subscribers, with user-approved tokenized card payments and merchant guidance for agent access and checkout compatibility.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
publishedConfidence 0%
Google states that Gemini 4 Hydrogen in Search can conduct simultaneous searches across subtopics and provide a wider range of relevant website links.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Strong EvidenceConfidence 0%
In the studied Chinese-language generative-search dataset, presence in the citation pool did not guarantee answer-level brand exposure: the authors report an overall brand-selection rate of 8.3%, and source sets differed systematically between app and web interfaces.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
publishedConfidence 0%
Perplexity says its AI-search processing pipeline cleans crawled pages, splits them into semantically coherent passages, computes embeddings, stores passages and embeddings together, and at query time selects query-relevant text for the model.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
publishedConfidence 0%
Huo and colleagues (2026) formalized multi-step LLM-guided search as selecting the next document from an existing ranked list or spawning a new ranked list through query reformulation or reranking.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
publishedConfidence 0%
In the tested ChatGPT, Claude, Grok, and DeepSeek configurations, web-search decisions and query strategies varied substantially; platform-specific search systems returned results with differing domain preferences, and some response claims relied on search results that were not cited.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Strong EvidenceConfidence 0%
Google documents that participating merchants can use Universal Commerce Protocol integrations to enable native transaction experiences on Google AI surfaces, including AI Mode in Search and Gemini; during early access, native checkout is available only to select merchants.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Strong EvidenceConfidence 0%
Google states that Gemini 3.8 Live began rolling out to everyone in Search Live on September 15, 2026, powering real-time conversational Search Live experiences.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Strong EvidenceConfidence 0%
In its October 15, 2026 announcement, Google said it was increasing inline-link visibility in AI Search and testing a dedicated top-of-response display for official website links, alongside related source-navigation changes.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
In Chen and Koudas's two-market study, ChatGPT API retrieval explained 32% of within-site citation variation compared with 2.8% explained by measured page properties; these are study-specific estimates.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
In Liu et al.'s marker-controlled publication experiments, 8 of 10 tested AI-search platforms cited a fabricated test concept within seven days.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Strong EvidenceConfidence 0%
Google announced on September 24, 2026 that Search Console Performance reporting includes a multimodal search type filter for Search results and Generative AI features, covering Lens, Circle to Search, image uploads and Chrome image search.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Strong EvidenceConfidence 0%
On October 30, 2026, Microsoft Bing announced independent AI Training and AI Grounding preferences for publishers, planned to be honored on recrawl starting February 1, 2027, without changing eligibility for traditional Bing search results.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Strong EvidenceConfidence 0%
On October 30, 2026, Google announced plans to support a shared AI-training policy token in robots.txt in 2027; absent an explicit Google-Extended instruction, Google plans to treat a shared-token training opt-out as a Google-Extended disallow, while retaining provider-specific overrides.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
publishedConfidence 0%
On October 13, 2026, Microsoft Bing announced Topics Portfolio (Beta) in Bing Webmaster Tools AI Performance, with personalized thematic clusters, approximate topic-level ecosystem metrics, and query-level opportunities.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
publishedConfidence 0%
On October 21, 2026, OpenAI announced that ChatGPT Ads had expanded to 26 additional countries, bringing availability to 89 countries, and reported more than 100,000 advertisers had run campaigns.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
publishedConfidence 0%
OpenAI's October 21, 2026 engineering description states that ChatGPT Ads uses a separate, LLM-orchestrated retrieval, eligibility, ranking and auction pipeline that can rewrite multi-turn intent into multiple semantic queries, while organic ChatGPT answers are generated independently.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
publishedConfidence 0%
In Jiang et al.'s October 11, 2026 observational study of four ChatGPT Web settings, combined Google Search and OpenAI Responses API proxies recovered only about one quarter of ChatGPT Web cited URLs, and proxy agreement was better at hostname and entity levels than exact-URL level.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Strong EvidenceConfidence 0%
On October 30, 2026 OpenAI announced that it intends to honor a common robots.txt signal for foundational AI training preferences while retaining existing crawler controls; implementation was described as forthcoming.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
In a controlled 31,200-answer audit, Chen and Koudas observed different citation-source and business-recommendation outcomes across locations; the tested ChatGPT and Gemini configurations used different declared-versus-network location signals.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Strong EvidenceConfidence 0%
Cloudflare announced its Web Search API through AI Gateway on October 2, 2026, initially naming Ceramic.ai, Exa and Linkup and stating that partner search responses must link to crawled sources and crawlers meet its Verified Bots requirements.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Strong EvidenceConfidence 0%
On October 28, 2026, Google announced a preview of the Connected Apps Platform for Search and Gemini, with cloud-hosted MCP or Android MCP integrations and a console for registering, testing, submitting and monitoring apps.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Strong EvidenceConfidence 0%
On October 28, 2026, Google announced scaling ads in AI Mode to ten named countries and launching sponsored multiturn ads for follow-up queries in the United States.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Strong EvidenceConfidence 0%
On June 16, 2026, Microsoft announced global preview availability of Intents, Topics, Citation Share and Compare within Bing Webmaster Tools AI Performance; Citation Share measures a site's fraction of cited-source appearances for a grounding query, not rankings or traffic.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
In a controlled five-source citation benchmark of 229 GEO-Bench queries and 1,220 non-degenerate engine-query groups, the released 4B distilled citation ranker achieved mean within-group Spearman correlations of 0.542–0.549 across three seeds; selected comparative decision models achieved 0.575–0.578, with differences not statistically distinguishable on the reported paired intervals, but formal equivalence was not established.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Strong EvidenceConfidence 0%
In a controlled two-document RAG testbed with 252,000 trials across six LLMs and 18 manipulated factors, Vishwakarma et al. found that topical relevance and candidate list position dominated first-citation selection; this does not test live web retrieval.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
EstablishedConfidence 0%
On April 7, 2026, Google India announced Shopping Graph-backed shoppable product listings and comparisons in the Gemini app in English and Hindi, and conversational AI Mode shopping results in English with Hindi support planned.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
EstablishedConfidence 0%
On February 10, 2026, Microsoft introduced Bing Webmaster Tools AI Performance public preview with total citations, average cited pages, sampled grounding queries, cited URLs and time trends across supported AI experiences.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
EstablishedConfidence 0%
Google Search Central states that Googlebot fetches up to 2 MB of a supported non-PDF URL including HTTP response headers, up to 64 MB for PDFs, and does not fetch, render or index bytes beyond the cutoff.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
In the AgentGEO paper's controlled simulated in-context generative-engine setup, the authors report a citation rate of 79.52% for AgentGEO versus 56.58% for unchanged webpages and 68.80% for AutoGEO; these are benchmark results, not live search-product outcomes.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Strong EvidenceConfidence 0%
On January 27, 2026, Google announced Gemini 3 as the global default for AI Overviews and the ability to move directly from an AI Overview to a context-preserving AI Mode follow-up conversation on mobile worldwide.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Strong EvidenceConfidence 0%
On April 16, 2026, Google announced that Chrome desktop AI Mode could open clicked webpages side by side with the AI conversation and that recent Chrome tabs could be selected alongside images or PDFs as AI Mode context, initially available in the U.S.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Strong EvidenceConfidence 0%
On May 6, 2026, Google announced five source-discovery updates for AI Mode and AI Overviews, including related-article links, subscription labels, community-perspective previews, additional inline links and desktop link-hover previews.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
In a study using 11,000 Natural Questions queries, Huang and colleagues reported that the GPT-4o-mini API web_search condition shared 25% of its top 100 cited domains with Google organic results and 24% with Google AI Overviews; the study also observed uneven representation of cited sources within synthesized answers.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
In an October 15, 2026 joint technical announcement, Cloudflare and OpenAI reported that a pilot begun in July 2026, involving approximately 1% of Cloudflare's network, enabled OpenAI to discover and index hundreds of millions of previously unknown web pages; OpenAI said it independently evaluates and ranks the pages.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
In a 11,500-query comparison of Google Search, AI Overviews and Gemini Flash 2.5, Grossman and colleagues reported AI Overviews for 51.5% of sampled queries and average Jaccard overlap of retrieved sources below 0.2 across the compared systems.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
In a randomized field experiment with 37,561 Washington Post archive readers, Lee and colleagues found that adding cited AI answers above conventional results shifted search and article-consumption behavior, increased shared topic exposure, and was associated with a small increase in articles consumed per reader.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
Sharifymoghaddam and colleagues report that ACARe, a 4B context-aware reranker for iterative agent search, improves dataset-averaged nDCG@10 by 0.9–2.0 percentage points over its base reranker across the reported benchmark configurations, with fewer searches for the five main agents.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Strong EvidenceConfidence 0%
Google's October 8, 2026 UGC Fresh Data Program documentation states that approved UGC platforms can submit fresh content and interaction signals through a dedicated Search ingestion pipeline separate from the Indexing API and ordinary crawling; Google does not guarantee Search inclusion.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
In their August 11, 2026 preprint, Xu, Guo and Xiong report that verifiable-content rewards (VCR) outperform the strongest comparison defense by an average 12.1 percentage points on their composite Net defense-utility metric across three controlled benchmarks.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
In their 2026 preprint, Wu and colleagues introduce MAGEO, a reusable-strategy multi-agent GEO framework, a twin-branch comparison holding the retrieval list fixed, and a DSV-CF metric that jointly evaluates semantic visibility and citation fidelity; their tested configurations outperform evaluated heuristic baselines.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
Yu and colleagues' Agent2UCB research describes evaluating nine GEO strategies per content item, selecting among them with a bandit policy informed by LLM priors and online feedback, and reporting GEO-Bench visibility improvements while monitoring text-only SEO-readiness measures.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Emerging EvidenceConfidence 0%
In the EMBL AI Librarian preprint, the authors report an improvement exceeding 16 points in Citation F1 on ScholarQABench over strong prior baselines, and approximately 8 points on open-form LitQA2 when a GPT-5.4 agent uses the Librarian retrieval layer instead of web search.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Strong EvidenceConfidence 0%
Perplexity reported that after migrating to Photon its production p99 retrieval-and-ranking latency declined from about 800 ms to 65 ms; this is a first-party reported benchmark.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Strong EvidenceConfidence 0%
In 13,929 LMArena search-log observations from March-April 2025, the researchers reported that 15.6% of responses had no logged web visits and 30% provided no citations, with substantial provider-specific variation.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
Strong EvidenceConfidence 0%
In a laboratory eye-tracking study published at SIGIR 2026, the researchers observed greater engagement with generative search content while traditional search links continued to attract attention.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
publishedConfidence 0%
In a 3,611-question health benchmark, Agnoli and colleagues report that 44.4% of traditional Google top-five results were cited by the corresponding AI Overview, and estimate that 56.7% of cited AIO statements were supported by their cited pages after classifier-error correction.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
publishedConfidence 0%
The OpenDS-World research benchmark contains 200 independently human-verified live-web deep-search tasks; across 12 evaluated frontier models, the top-performing model achieved 39.0% accuracy in the authors' common tool environment.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
publishedConfidence 0%
Cloudflare announced that Redirects for AI Training converts qualifying same-origin non-self-referencing canonical tags into HTTP 301 redirects for verified AI training crawlers, excluding human visitors, search indexing and AI agent traffic.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.
publishedConfidence 0%
On May 15, 2026, Google Search Central announced a guide for generative AI features in Search covering original content, local and shopping content, images and videos, GEO misconceptions, AI agents and the continued relevance of foundational SEO.
Evidence status lives at claim level so contradictory sources can change the conclusion without rewriting the historical record.