Context
A research preprint studies the end-to-end web-search lifecycle across ChatGPT, Claude, Grok, and DeepSeek using real-world interactions and controlled experiments.
What changed
The study reports substantial differences in search-invocation decisions and query strategies across platforms, platform-specific domain preferences in returned search results, and cases where response claims rely on search results that are not cited.
Why it matters
The results provide empirical evidence that AI discovery is a multi-stage process: deciding to search, formulating machine-written queries, retrieving candidate sources, selecting evidence, grounding responses, and exposing citations can vary independently across platforms.
Visibility OS interpretation
Visibility OS should measure search invocation, query formulation, retrieved-domain availability, source use, and citation as separate observables rather than collapsing them into a single AI visibility score. More frequent search invocation should not be assumed to improve response quality.
What remains uncertain
How stable are these behaviors across model updates and time? How much of domain preference comes from the platform search backend versus model-side selection? Do uncited-but-used sources materially affect brand and publisher visibility measurement?