How to measure your AI search visibility

Most claims about AI search visibility rest on a screenshot. Someone types a question, sees their business named, and calls it a result. The next day the same question names someone else.

This guide sets out a simple method you can run yourself, or use to check anyone who reports numbers to you. It needs a spreadsheet, a browser and a little time each month.

1. Write a fixed question set

Write the questions your customers really ask before they choose. Use their words, and include the details they'd include:

• The place: a town, a district or "near me".
• The service: the specific job, treatment or product, as a customer would say it.
• The worry or constraint: price, speed, nerves, accessibility, a deadline.
• The comparison: "best", "cheapest", "alternatives to", "who specialises in".

Good sources for questions are your enquiry emails, what customers ask on the phone, your reviews, and the "People also ask" boxes in Google. A good set runs from broad ("Who is the best accountant in Leeds?") to specific ("Which Leeds accountants help landlords with Making Tax Digital?").

Once the set is written, freeze it. Changing questions between measurements makes before-and-after comparisons meaningless. Add new questions as a separate set if you need to.

2. Ask each question more than once, on different days

AI answers vary. The same question can name different businesses at different times. One answer can make you look better or worse than you are, so ask each question on more than one day and use every answer, not the best one.

3. Use the tools your customers use, logged out

For most UK businesses that means Google AI Mode and ChatGPT, the AI search tools customers are most likely to use for "who should I choose" questions. For some B2B buyers, Microsoft Copilot matters too.

Run every question logged out, in a private browser window, from a UK connection. Signed-in accounts can carry history, memory and preferences that shape the answer; OpenAI's help pages say saved memories can influence how ChatGPT rewrites search queries. A logged-out answer is closer to what a new customer sees.

Location matters for local questions. Include the town in the question, and run every measurement from the same general location.

4. Record the same fields every time

For every answer, record:

Save the full text of every answer. If anyone questions a number later, you can show the answer it came from.

5. Report counts, with the method stated

From those records, 4 numbers tell you most of what you need:

• Appearance rate: answers naming you out of all the answers you recorded.
• Citation count: answers that cite one of your own pages or your Google profile.
• Competitor counts: how often each competitor is named, so you can see who AI prefers and for which questions.
• Errors: each wrong fact, where it appears to come from and whether it's fixed.

Always write the method next to the number: the questions, the tools and the dates. "Named in 9 of 40 answers, logged out, 3 to 10 November 2026" can be checked. "Visibility up 200%" can't.

6. Treat small changes with caution

With a few dozen answers, normal variation can move the count by 1 or 2 without anything having changed. Before calling something a result, look for a change that holds across days and tools, and check that the answers now repeat details you added. When the count doesn't move, say so.

Give changes time. Search engines and AI tools need to re-crawl changed pages before answers can reflect them, so leave a gap before measuring again.

7. Add what the platforms report

The platforms give partial data, and it's worth collecting alongside your own measurements.

• Google Search Console. Google says sites appearing in AI features are included in overall search traffic in Search Console, reported in the Performance report under the "Web" search type. You can't separate AI Overviews or AI Mode clicks from other search clicks there.
• Bing Webmaster Tools. Since February 2026, the AI Performance report (in public preview) shows how often your pages are cited in Microsoft Copilot, AI-generated summaries in Bing and some partner integrations, including total citations, cited pages and a sample of the grounding queries used.
• Your analytics. OpenAI says ChatGPT automatically adds utm_source=chatgpt.com to referral links, so visits from ChatGPT answers can be counted in Google Analytics or similar tools.
• Your enquiries. Add "an AI tool such as ChatGPT" to your "how did you hear about us" options. It's the closest link between AI answers and revenue that most businesses can get.

Common measurement mistakes

• Testing while signed in to your own Google or ChatGPT account.
• Searching for your own business name, which tells you little about whether new customers will find you.
• Changing the questions between measurements.
• Reporting the best of several answers.
• Counting a mention in a list of "other options" the same as a recommendation, without noting the difference.

If you'd rather not run this yourself, the AI visibility audit does it for you and hands over every answer, and the monthly plan keeps measuring. Our research on London gyms shows what a write-up looks like.