A client asks whether they show up in ChatGPT. Someone on the team types the client's name, sees it in the answer, and sends a screenshot. A week later a competitor's screenshot shows the opposite. Both are real and neither is a measurement. AI answers are generated fresh every time, so the only honest way to track visibility is a fixed set of questions, asked the same way, on a schedule, with the sources written down.

Why one check proves nothing

Every answer is composed on the spot from whatever the search step returned, and the model's wording and choice of sources shift between runs. Add personalisation from a logged-in account's history, the searcher's location, which model version is answering, and whether the search step fired at all, and two people asking the same question get different citations. How much it varies is observed rather than documented by any of the companies, but anyone who repeats a question will see it.

Three rules follow. Ask the same wording every time. Ask in a clean state: a fresh chat, logged out where you can be, and set to the market you sell in. Repeat on a schedule and read the trend, never the single result.

A prompt that contains your brand name ("tell me about Harbour Dental") will always find you and proves nothing. Measure with the questions a customer asks before they know your name.

Separate "an AI answer exists" from "it cites you"

These are two facts, and they need two columns. For any query there are three outcomes. No AI answer appears, so the blue links carry the query and there is nothing to lose. An AI answer appears and cites other sites, which is the gap worth working on. An AI answer appears and cites you, which means the cited pages are doing the work and deserve protecting.

Many transactional and local queries show no AI Overview at all, and that is a genuine zero rather than a defeat.

Record which domains get cited when it isn't you. For recommendation questions they are often roundups, directories, forums and review sites rather than your direct rivals, and they tell you where to earn a mention.

Build a baseline you can repeat

A spreadsheet is enough. The discipline is in keeping it the same from month to month.

Write the questions a customer asks before they know your name. Mix the sub-questions from the free AI query fan-out simulator with the questions your sales conversations start with. Keep the list short enough that you'll repeat it every month.
Pick the assistants that matter for your market: Google's AI Overviews and AI Mode, ChatGPT, Gemini and Perplexity cover most businesses.
For each question and each assistant, record four things: whether an AI answer appeared, whether your brand was mentioned in the text, whether your site was cited with a link, and which domains were cited.
Run your two closest competitors through the same questions in the same session, so the comparison shares the same conditions.
Repeat monthly with the same wording and the same settings, and add a column rather than editing old ones.

The payoff is watching "appeared, not cited" rows turn into "cited" after the work from the writing lesson.

What your analytics can and can't tell you

Google's Search Console Help says that impressions and clicks from AI Overviews and AI Mode are included in the Performance report, but not broken out as a separate filter. So your Search Console data already contains AI surface traffic and there's no way to isolate it there.

In your analytics, a click on a citation arrives as a referral from the assistant's own domain, and chatgpt.com, perplexity.ai, gemini.google.com and copilot.microsoft.com show up in referral reports. Group them into their own channel so you can watch them together. Treat the count as a floor rather than a total: not every assistant visit carries a referrer, and a click from an AI Overview looks like any other Google organic click.

Mentions of your brand on the indexed web are a leading indicator, because assistants cite third-party pages for recommendation questions. Count mentions of your distinctive brand phrase and watch the direction between checks. If your name is a common word, measure the distinctive part of it, or the count is mostly noise. The Brand mentions and entity SEO lesson covers how to grow that number.

The ways this measurement lies

A citation that appears in one run and vanishes in the next is noise until it repeats.
A mention in the text and a citation with a link are different outcomes; report them separately.
An AI answer that doesn't exist for a query is not a loss, and a report that counts it as one will send you chasing the wrong keywords.
Don't claim a citation as the result of last week's change unless the same query, in the same wording, was uncited the month before.

What to take away

A single screenshot of an AI answer is not a measurement; answers vary by run, account, location and model, so only repeated checks with fixed wording count.
Track whether an AI answer exists and whether it cites you as separate facts, because "no AI answer" is a neutral result, not a loss.
Search Console folds AI Overview and AI Mode traffic into the Performance report without a filter, and referral reports show assistant clicks only when a referrer is passed, so both are floors rather than totals.
Which third-party domains get cited instead of you is the most useful number in the sheet, because it tells you where to earn a mention.

The domains an assistant cites instead of you are usually sites that link to and talk about the businesses they recommend, so the next chapter starts with Backlinks that matter, and the ones that don't.

Measuring your AI visibility