How AI Search Proof measures brand visibility
A reproducible audit needs explicit surfaces, frozen prompts, timestamps, citations and honest failure states. It does not need to pretend an API response is a screenshot of a consumer chat app.
Freeze the question set
Your market, language, competitors and up to 15 buyer questions are stored with the paid order. That makes the baseline comparable.
Measure four API surfaces
Perplexity Sonar, OpenAI with web search, Gemini with grounding and Claude with web search. Production labels disclose routing when an aggregator is used.
Keep evidence and errors distinct
We parse brand presence, position, competitor names and URL citations from the same answer. A timeout, unavailable provider or invalid payload is unmeasured—never a zero.
Score transparently
Presence contributes 50%, position 20%, own-domain citations 20% and crawler readiness 10%. The evidence rows remain the source of truth.
What the audit can and cannot prove
It can show what the named API returned for a frozen prompt at a known time. It cannot guarantee what every user sees in ChatGPT, Gemini, Claude, Perplexity or Google AI Overviews. Consumer interfaces vary by account, location, personalization, experiments and time.
Privacy and retention
Pending unpaid orders expire after 24 hours. Raw provider payloads are retained for up to 30 days, derived audit reports for 12 months, subscription data until 90 days after cancellation and magic links for 15 minutes. Full IP addresses are never stored for locale detection.
Paid boundary
Form validation is free because it does not retrieve the site or call providers. A personalized scan begins only after a server-verified payment webhook. Public educational pages and the fictional Northstar sample are not personalized reports.