How to Measure AI Search Visibility

Direct answer. Measure AI-search discovery with a fixed set of buyer questions, dated answer samples, brand mentions, cited source URLs, available platform reports and qualified downstream visits. Keep mentions, citations and leads separate. A sampled answer is an observation, not a stable rank or a complete count of exposure.

AI answers can vary with the prompt, user context, location, date and product surface. A single screenshot cannot support a claim that a company “ranks first in ChatGPT.” Build a repeatable observation method before making budget decisions.

Define what counts

Record four different events: the brand is named; an owned URL is cited; a third-party page about the brand is cited; or a visitor reaches the site and completes a meaningful action. Those events answer different questions. A mention may build awareness without a click, while a referral may not preserve every step that led to it.

For Vaquero, the business outcome is a completed, qualified review tied to an in-scope business. Do not call a form submission or an AI mention a qualified review. Connect visits to the same CRM stages used for organic and paid acquisition when the data is available.

Fix the prompt set and observation method

Choose a small set of neutral buyer questions from the keyword universe: service selection, problem diagnosis, cost and process, platform-specific discovery and comparison. Include non-branded and branded prompts. Keep the wording, location, account state and date with each observation. Sample the same set at a regular cadence and note product or interface changes.

For every answer, record whether Vaquero was mentioned, which URLs were cited, where a citation pointed, and whether the answer actually addressed the prompt. Review the source page for factual accuracy and relevance. Do not count repeated mentions in one answer as separate opportunities.

Use available platform evidence

Bing Webmaster Tools introduced an AI Performance view (in public preview as of early 2026) that reports cited pages and sampled grounding query phrases. Those observations can reveal which site pages are being used and what related phrases need better answers. Google Search Console and analytics provide useful search and referral context, but reports and attribution vary by product and time. Consult current platform documentation before assuming a particular metric exists for ChatGPT, Claude, Grok, Gemini or Perplexity.

Use landing-page and CRM data to connect any identifiable referral to booked and completed reviews. Preserve unknown sources instead of assigning credit based on a guess. A citation on an answer with no click can still be worth observing, but it is not a lead.

A sample tracking log

A spreadsheet is enough to start. Use one row per prompt, per surface, per date, with these columns:

Date and time of the observation
Platform and product surface
Account state and location used
Exact prompt wording
Prompt type (non-branded, branded, comparison or platform-specific)
Brand named (yes or no)
Owned URLs cited
Third-party URLs cited
Answer addressed the question (yes, partly or no)
Factual issues found in the answer
Identifiable referral visits in the same period
Booked or completed reviews tied to those visits
Reviewer and notes

Turn observations into editorial work

If another source is consistently cited for a buyer question, inspect what that source provides: a clearer definition, firsthand evidence, a comparison, a current fact or a useful format. Improve the one canonical Vaquero page that answers the question. If a relevant page is never accessible to a crawler, run the technical readiness check before rewriting it. If the issue is a weak decision answer, the content authority guide offers the editorial framework.

Limit: Prompt samples are not population-level impression data. Mentions and citations do not prove a platform endorsement or cause a sale. Report sample size, surfaces and missing attribution beside any trend.