How to Measure Your AI Search Visibility

Each source answers a different question. None of them answers "how visible am I in AI search?" on its own.

Table · scrolls sideways

The sections below take each source in turn, then show how to put them into one monthly record.

Google's page on AI features and your website is explicit: traffic from AI features is "included in the overall search traffic in Search Console" and "reported on in the Performance report, within the 'Web' search type." There is no separate AI Overviews filter.

That makes Search Console useful as a baseline rather than an AI meter. Watch the pages you expect to be cited for: if their impressions rise while clicks fall, that may reflect answers satisfying the question on the results page, or it may be a dozen other things. Note the change and the date in your log; don't explain it with a guess.

Before measuring anything, use URL Inspection on your key pages. Google says a page must be indexed and eligible for a snippet to appear in its AI features. A page that isn't indexed can't show up, and no amount of measuring will fix that.

On 10 February 2026 Microsoft launched AI Performance in Bing Webmaster Tools as a public preview. It is the closest thing to a citation dashboard a site owner can get for free. It reports:

• Total citations: how many times your pages appeared as sources in AI-generated answers over a period.
• Average cited pages: the average number of unique pages from your site cited per day.
• Grounding queries: a sample of the phrases the AI used when retrieving your content.
• Page-level citation activity and trends over time.

Read it with Microsoft's own caveat: the cited-pages figure "does not indicate ranking, authority, or the role of any page within an individual answer." It covers Microsoft's surfaces only. It is not a ChatGPT or Google number, and it shows citations, not clicks. If you have not yet verified your site in Bing Webmaster Tools, that is the first job.

When someone clicks a link in an assistant's answer, the visit usually arrives with that assistant's domain as its source. In Google Analytics 4, the Traffic acquisition report shows Session source, "the publisher or inventory source from which traffic originated." Filter it for assistant domains such as chatgpt.com and perplexity.ai, check the list for any others that appear, and save the view.

Then look past the visit count. Which pages did those visitors land on, and did any of them send an enquiry? A handful of qualified enquiries matters more than a large referral number.

Two limits keep this honest:

1. Uncounted visits. Google Analytics files traffic "that doesn't have a clear referral source" under (direct) / (none), and lists causes such as redirects that strip campaign tags (GA4 help). Some AI-driven visits will land there. Your referral count is a floor.
2. Citations nobody clicks. Someone can read an answer that names your business and call you without visiting the site. Analytics never sees it. Ask new enquiries how they found you, and record the answer.

This is an access check, not a visibility metric, but it rules out the most basic failure. OpenAI's crawler documentation says OAI-SearchBot is "used to surface websites in search results in ChatGPT's search features", separately from GPTBot, which crawls content that may be used for training. Each setting is independent, and OpenAI says a robots.txt change for search takes about 24 hours to register. Perplexity's crawler page describes PerplexityBot as the crawler that surfaces and links sites in its search results, and Perplexity-User as a user-triggered fetcher that generally ignores robots.txt.

Ask whoever runs your hosting to check robots.txt, firewall and bot-protection settings, and to search the server or CDN logs for those user agents on your key pages. If the search crawlers are blocked or challenged, fix that before you read anything into the numbers above.

The tools above show traffic and counts. None shows what an answer actually says about you. For that, we use a manual panel, and you can run your own.

Our method: a fixed panel of buyer questions, run in the major assistants and logged. We describe the method here, not results. To run one:

1. Write the questions customers actually ask. Mix unbranded questions about the service and area with branded questions about the business. A fictional example for a Waterloo bookkeeping firm: "Who does small-business bookkeeping in Waterloo?" and "What does [firm name] charge for monthly bookkeeping?"
2. Fix the list and the setup. Run the same questions, worded the same way, in the same assistants, from a signed-out or fresh session where you can. Changing the wording changes the answer.
3. Log each answer with the same fields: date, assistant, question, whether you were named, which sources were cited, and whether what it said about you was accurate.
4. Repeat on a schedule you can keep, and log a page change on the date you make it, so later readings have context.
5. Never put customer details into test questions.

Treat the panel as a sample. Answers vary by user, location, wording and the day. A single favourable screenshot is an observation, not a position. The value is in the accuracy column as much as the "named" column: an assistant describing a service you stopped offering is a fix you can make on your own pages.

Be clear about the gaps before you report anything to a partner or a board:

• Every answer given about you. No platform shares that, and a panel only samples it.
• Placement inside an answer. Bing says its counts don't indicate it; the others don't report it.
• Google AI features on their own. They are blended into the Web report.
• A ChatGPT or Perplexity citation count for site owners. As of 15 October 2026, neither offers one.
• Cause and effect. If citations rise after a page change, that is a correlation in a small sample, not proof.

Keep the stages separate, because they are different events: being cited, being clicked, and becoming an enquiry. A one-page monthly record with five lines is enough:

Table · scrolls sideways

Add one line of notes for anything you changed that month. Over time, that record tells you more than any single dashboard. For the changes worth making in the first place, see how to help your website appear in ChatGPT search; for the definition and the decision, see what AI search optimization is.