Context
An original study analyzes 14.44 million search requests and 3.97 million sessions from a shared research search API used by external agents, finding intent-dependent query reformulation and substantial reuse of terms from previously retrieved evidence.
What changed
The authors introduced a session-log analysis and Context-driven Term Adoption Rate (CTAR) measure to study how accumulated retrieval evidence relates to later agent search queries. They report over 90% of multi-turn sessions contain at most ten steps, and an average 54% of newly introduced query terms appear in accumulated evidence.
Why it matters
Agentic discovery involves sequences of queries informed by prior retrieval, so measurement limited to the initial prompt or a single cited result can miss how sources enter later searches.
Visibility OS interpretation
Visibility OS interpretation: evaluate search-agent discoverability as a trajectory, including cross-step evidence reuse, query reformulation and repeated searches; this is a research API observation, not a confirmed commercial AI-search ranking mechanism.
What remains uncertain
How well do these patterns transfer to consumer ChatGPT Search, Gemini, Perplexity or Google AI Mode? How much of term reuse reflects causal influence rather than shared topic vocabulary? How sensitive are results to sessionization and LLM-derived intent labels?