AI visibility measurement, version vis-2027.1
How AISeeWhy asks AI assistants about a brand, what it records for every answer, and what the numbers can and cannot tell you.
Version vis-2027.1, released 2026-11-22. First versioned release. Every answer is stored as a measurement record with the model the provider says it used, its response id, token usage, cost and the full raw response.
What is asked
- Each prompt you enter is sent as typed to each assistant you pick, through the provider's official API, with web search turned on.
- Prompts are never rewritten. A changed prompt is a new prompt with a new id.
- A run can ask each prompt several times (repeats). Results are shares of answers and median positions, so one random answer does not decide.
What is recorded for every answer
- Provider, the model requested and the model the provider returned, the endpoint and the response id.
- Parameters: web search on, the prompt's position in the run and the repeat number.
- Status (ok, refusal or an error code), latency, input and output tokens, search calls, and cost (reported by the provider, or our estimate when it reports none).
- The full raw response, compressed, in private storage for 13 months, with its SHA-256 hash, so any number can be re-extracted later without asking again.
How an answer is read
- Mentioned: the brand or one of its other names as a whole word, after a false-match guard (case, another company's domain, strict mode for common words).
- Position: the brand's place in the first list of the answer that names it.
- Sentiment: positive, neutral or negative from words in the sentences that name the brand.
- Cited: whether your domain is among the sources the assistant returned.
Known limits
- API answers are not the same as answers in the consumer apps (ChatGPT, Gemini, Claude, Perplexity): models, search and personalization differ.
- Answers change from one run to the next. Small samples have wide uncertainty; compare shares over several runs.
- Location and language are not set unless the prompt says so.
- When a provider changes the model behind a name, results can shift. The model id the provider returned is recorded so such breaks are visible.
How the score works · Other versions: audit-2027.1