How to Measure Search and AI Visibility (and Track Whether ChatGPT Mentions You)

'Are we visible in AI search?' sounds like a single question with a single answer. It isn't. A business can be indexed by Google, absent from any AI-generated answer, getting steady organic traffic, and still converting none of it into enquiries — four different outcomes, four different causes, and four different ways to check them.

This article sets out the five stages worth measuring separately, what each genuinely tells you, which tools to use, and a realistic manual method for checking whether tools such as ChatGPT or Perplexity mention your business — along with its real limitations. There is no single dashboard that answers this properly, and anyone claiming one exists is overstating what is currently measurable.

Five different things, often confused as one

'Visibility' is used loosely to mean indexing, ranking, traffic, AI mentions and conversions — five separate stages that each fail for different reasons and need different fixes. A page can be indexed but never rank; rank but never be clicked; be clicked but never convert; and a business can convert perfectly well through Google while never once being mentioned by an AI assistant. Treating these as one metric hides where the actual problem is.

What Google Search Console actually tells you

Search Console is free, official, and the most reliable instrument a business has for the first three stages. It is worth understanding its core metrics in plain terms rather than treating them as a single health score.

• Impressions — how many times a page appeared in a Google results page for any query, whether or not anyone saw it on screen
• Clicks — how many of those impressions resulted in someone clicking through to the site
• CTR (click-through rate) — clicks divided by impressions; a rough indicator of whether the title and description are compelling for that query
• Average position — the average ranking position across all impressions for a query or page; useful as a trend, less useful as a single number because it blends very different queries
• Queries — the actual search terms that triggered an impression, which often reveals buyer language a business has not written content for
• Landing pages — which specific pages are earning impressions and clicks, useful for spotting pages that rank but never convert

A sustained drop in impressions for a page usually points to an indexing or ranking problem. Impressions holding steady while clicks fall usually points to a title/description or competitive-results problem, not a content problem. Clicks holding steady while conversions fall points away from search entirely and into the page itself — see the related article on traffic without enquiries.

What analytics tells you — and where referrer data quietly runs out

Standard web analytics will show a traffic source such as 'chatgpt.com' or 'perplexity.ai' when a visitor clicks a link from an AI assistant's answer and the platform passes an HTTP referrer header. This is a genuine, measurable signal when it appears, and worth setting up a dedicated channel or segment for. The limitation is real: not every AI platform, surface or app passes a referrer reliably, and a user who reads an answer, closes the app and later searches for the business by name will show up as 'direct' or 'organic', with no trace of the AI interaction that originally prompted it. Analytics referral data is a useful lower-bound signal for AI-driven traffic, not a complete count of AI influence.

Checking whether AI systems mention your business: a manual method

Because no major provider publishes an 'AI mention' report, the only reliable way to check is to ask the questions yourself, repeatedly, and keep a record. This does not produce a definitive score, but it produces something more honest than guessing.

1. 01Write a fixed set of real buyer questions — the ones a prospective customer would actually ask, not brand-name searches
2. 02Choose the platforms to test (for example ChatGPT, Google AI Overviews/AI Mode, Perplexity, Microsoft Copilot) and keep the set consistent
3. 03Log the date, the platform, the account state (logged in or not, location setting, chat history on or off) and the exact question wording
4. 04Run each question more than once, on different days, because answers are not deterministic and will vary between runs even with no change to the underlying website
5. 05Record whether the business is mentioned, whether it is cited with a link, and which specific URL (if any) the citation points to
6. 06Repeat the same set on a schedule — monthly is reasonable for most B2B businesses — so that change over time, not a single snapshot, is what gets acted on

The real limitations of manual testing

• Personalisation and location — logged-in state, account history and inferred location can all change what an assistant answers, so your test may not reflect what a prospect sees
• Non-determinism — the same question, asked twice in the same session, can return different sources and different wording, which is a documented characteristic of how these models generate answers, not a testing error
• No official reporting — most providers do not publish how often a business is mentioned, so there is no ground truth to check a manual sample against
• Small sample size — testing ten or twenty questions a month tells you about those questions, not about every way a buyer might ask
• Third-party 'AI visibility' tools estimate mentions using their own sampled queries and their own methodology; treat their numbers as directional, not authoritative, and ask any vendor exactly how they generate them

Building a simple, honest measurement routine

• Monthly: review Search Console impressions, clicks, CTR and queries for key pages; flag anything moving sharply
• Monthly: check analytics for referral traffic from known AI domains and treat it as a minimum, not a total
• Monthly or quarterly: run the fixed question panel across your chosen AI platforms and log the results in a simple spreadsheet
• Quarterly: review conversions from organic and AI-referred traffic against enquiries actually logged in the CRM, not just form submissions
• Avoid chasing a single number; look at direction of travel across all five stages together