The short answer
You measure AI answer visibility by asking the major AI systems the questions your customers actually ask, recording whether your brand is named, cited and described accurately, and repeating the same question set every month. There is no dashboard that does this for you. The measurement is the habit.
That absence of tooling puts most brands off, which is exactly why doing it is an advantage. The method below takes an afternoon to set up and an hour a month to maintain.
Why your analytics can’t see this
When ChatGPT recommends a competitor instead of you, nothing happens in your analytics. No impression is logged, no click is lost that you can count, no query appears in Search Console. The customer asked, the answer was given, and the journey moved on without touching your website.
That’s the defining property of AI answer visibility: it fails silently. So the measurement has to be active. You go and read the answers, because no report will bring them to you.
Step 1: Build the question set
Write down 20 to 50 questions a real customer would ask before buying what you sell. Not questions about you. Questions about the problem and the category. Four types cover most of it:
- Category questions. “Best payroll software for UK small businesses.” “Recommend a commercial cleaning company in Leeds.”
- Comparison questions. “X versus Y for a small team.” “Alternatives to X.”
- Problem questions. “How do I reduce returns in an online clothing shop?” The answers to these often recommend providers unprompted.
- Trust questions. “Is X reputable?” “Who are the leading firms in [category]?”
Pull the wording from real sources: sales call notes, support emails, the queries already in your Search Console, the questions prospects ask on first calls. The closer to real customer language, the more honest the measurement.
Step 2: Ask the systems
Run the set through the systems your customers use. For most UK brands that means ChatGPT (with web search enabled), Perplexity, Google’s AI Overviews and AI Mode, and Microsoft Copilot, which draws on Bing’s index.
Two rules keep the results honest. First, ask like a customer, not like a marketer: “who should a UK ecommerce brand use for email marketing?” measures visibility, while “tell me about [your brand]” only measures whether the system can look you up. Second, remember the answers are probabilistic. The same question can produce different answers on different runs, so ask the important questions more than once and treat single runs as noise.
Step 3: Record four things per answer
For every question, log:
- Named: does your brand appear in the answer at all?
- Cited: does the answer link to or reference your site as a source?
- Accurate: is the description of what you do, who you serve and where you operate correct?
- Company: which competitors, publishers or directories appear instead of or alongside you?
A spreadsheet with one row per question per system is enough. The fourth column matters more than it looks: the sources that keep appearing instead of you are the places your corroboration is missing.
Step 4: Turn it into a score
Three numbers summarise the sheet. Answer share: the percentage of questions where you’re named. Citation share: the percentage where your site is used as a source. Accuracy rate: the percentage of mentions that describe you correctly. Track them per system and overall.
Don’t expect big numbers at first. For a small brand in a competitive category, moving answer share from zero to appearing in a third of category questions is a meaningful commercial change, because each appearance sits directly inside a buying decision.
Step 5: Repeat monthly, same set, same method
Visibility in AI answers moves as models update, indexes refresh and your own signals improve. A single check is a snapshot; the value is in the trend. Keep the question set stable so months are comparable, add new questions at the end rather than replacing old ones, and note the date of each run.
The mistakes that spoil the measurement
The common ones: testing only brand-name queries (which measures recognition, not visibility), asking leading questions that smuggle your brand into the prompt, running everything in a logged-in account full of your own history, and doing it once then never again. Any of these will tell you a flattering story instead of a true one.
What to do with the results
The gaps point at the work. Low answer share with decent rankings usually means an entity or corroboration problem. Mentions without citations usually mean your content isn’t quotable enough to use as a source. Inaccurate descriptions mean the public record about you is thin or inconsistent.
That translation from measurement to fix is the core of our AI Search Visibility service, and the audit method behind it is written up in a practical AI visibility audit. If you’d rather see your numbers before deciding anything, request a free audit and we’ll run the first measurement for you.