An observation, not a promise.
AI answers vary. Seen42 uses repeated, versioned prompts and Cite42 measurements to show what was observed, where it came from and what can be compared.
Mentioned and cited are separate
A mention records a named brand and its first-mention position in an answer. A citation records an owned URL among the sources returned for that assistant. Neither measures a recommendation or a search ranking. Returned sources are limited, so no observed citation is not proof that the assistant never cited you.
Sampled answers have variance
Measurements use supported provider APIs and, where available, native search or live search results. They are not identical to every consumer chat session. Responses can differ by time, wording, geography and model. We record the assistant surface and methodology version.
Failed answers do not count as absence
Errors, empty or incomplete responses are excluded from the valid-answer denominator. A partial run is labelled partial. Views show their sample size so a single successful answer is not mistaken for broad coverage.
Prompt edits preserve history
Each run saves its prompts, control flags, assistants and competitors. Editing wording creates a new prompt identity. Trends compare matching prompt cohorts and methodology versions rather than rewriting previous results.
Keep a control group
Targeted prompts have pages built for them. Control prompts do not. Compare both over repeated checks to help distinguish a change in coverage from ordinary answer variation. This is an observational test, not proof that publishing caused a result.
Sentiment is scored where you are named
Once a month, for prompts where the latest check named you, Cite42 asks the assistants that named you again and scores how the answer talks about you: positive, neutral or negative, with the phrases behind it. The score runs from 0 to 100, where 50 is neutral. Answers that name you without describing you are counted but not scored, and competitors are not scored.
Answers are fetched on demand
Stored rankings include mentions, positions and sources, not full answer text. A separate paid request can fetch a fresh AI answer. That answer is a new observation, not a replay of the historical tracker result.