For years, brand visibility in search had a simple scoreboard. You tracked rankings, impressions and clicks, and you knew roughly where you stood. That scoreboard still matters, but it no longer tells the whole story. A growing share of research now starts with a question typed into ChatGPT, Claude or Google’s AI features, and the answer arrives as a short summary rather than a list of ten blue links.

If your brand is not part of that answer, a buyer may never reach your website at all. The difficulty is that most analytics tools were not built to show you what an AI assistant says about you. This guide explains how to measure brand visibility across the major AI platforms in a structured, repeatable way, so that you can see where you stand and decide what to fix first.

Why AI visibility needs its own measurement

Traditional search visibility is measured against a ranked list. AI visibility is different in three important ways.

First, the answer is generated, not retrieved. Two people asking the same question can receive slightly different responses, and the same person can receive a different answer next week. A single check proves very little.

Second, the unit of visibility is a mention, not a position. Your brand might be recommended first, listed as one option among five, mentioned with a caveat or left out entirely. Each outcome has a very different commercial value.

Third, accuracy matters as much as presence. An assistant that mentions your brand but describes your pricing, features or service area incorrectly can do more harm than one that does not mention you at all.

Because of these differences, measuring AI visibility is less like checking a ranking and more like running a small, recurring research study.

Step 1: Build a prompt set that reflects real buyer questions

Everything starts with the questions you test. A good prompt set mirrors how your buyers actually ask for help, not how your marketing team describes the product.

Group prompts by intent

Organise your prompts into a few clear groups:

·         Category discovery: “What are the best tools for [problem]?” or “Which agencies help with [service] in [city]?”

·         Comparison: “[Your brand] vs [competitor]” or “Alternatives to [competitor].”

·         Problem-led: “How do I fix [specific problem]?” where your product or service is a natural answer.

·         Brand-specific: “What is [your brand]?”, “Is [your brand] good for [use case]?” and “How much does [your brand] cost?”

Use your own data to choose the wording

Sales call notes, support tickets, site search queries and Search Console queries are the best sources for realistic phrasing. Long, conversational queries in Search Console are especially useful, because they often resemble the way people talk to AI assistants.

A practical starting point is 30 to 50 prompts. That is enough to show patterns without becoming impossible to review by hand.

Step 2: Test each platform consistently

ChatGPT, Claude and Google’s AI features do not behave in the same way, so test each one separately and keep the conditions as consistent as you can.

ChatGPT

Run your prompts in a fresh conversation, ideally with memory and custom instructions turned off, so that earlier chats do not influence the answer. Note whether the response used web browsing or answered from the model’s built-in knowledge, because the two can produce different brand lists.

Claude

Follow the same approach: a new conversation for each prompt group and no project context that could bias the result. Record whether the answer drew on web search, where that option is available.

Google AI Overviews and AI Mode

Search for each prompt in a clean browser profile, signed out where possible, and record whether an AI Overview appeared at all. When it does appear, capture which brands are named and which sources are linked. For local or regional businesses, note the location you searched from, because results can change by market.

Repeat, then record

Because AI answers vary, run each prompt more than once, for example three times per platform in each measurement cycle. Save the full response text or a screenshot along with the date, the platform and the exact prompt. Without this record, it is very hard to tell real change from normal variation.

Step 3: Score what you find

Raw answers are hard to compare, so turn them into a small set of metrics. Keep the scoring simple enough that anyone on the team can apply it in the same way.

Mention rate

The share of responses that mention your brand at all. If your brand appears in 12 of 60 responses, your mention rate is 20 percent. This is the headline number most teams track first.

Position and prominence

Where your brand appears relative to others. A simple scale works well: recommended first, listed among the main options, mentioned in passing, or absent. Prominence often matters more than raw mention rate, because buyers tend to act on the first one or two suggestions.

Share of voice

Your mentions as a proportion of all brand mentions across your prompt set. Tracking the same set of competitors each cycle shows whether you are gaining or losing ground in your category, even when overall answer patterns change.

Sentiment and framing

Whether the answer describes your brand positively, neutrally or with caveats such as “can be expensive” or “limited support”. These caveats are worth tracking closely, because they often point to a specific perception you can address.

Accuracy

Whether the facts are correct: pricing, features, integrations, locations and policies. Every inaccuracy should be logged with the platform and prompt, because each one is a concrete fix waiting to happen.

Citations

For platforms that show sources, which pages are cited when your brand is mentioned. Note whether they are your own pages, review sites, directories, forums or competitor content. This tells you which sources actually influence the answers.

Step 4: Connect AI visibility to real traffic

Mentions are useful, but leadership will eventually ask whether AI visibility sends real visitors and revenue.

Track AI referral traffic in analytics

In GA4, create a custom channel group or exploration that isolates referrals from domains such as chatgpt.com, claude.ai, perplexity.ai and gemini.google.com. This traffic is often small today, but its trend is a useful signal, and it is easy to compare engagement and conversion rates against other channels.

Watch for branded search lift

When AI assistants mention a brand, some people search for that brand name directly instead of clicking a link. A rise in branded search impressions in Search Console, alongside a rising AI mention rate, is a reasonable sign that the two are connected.

Ask customers directly

Add “AI assistant (for example ChatGPT or Claude)” as an option in your “How did you hear about us?” field on forms, checkout or onboarding. Self-reported attribution is imperfect, but it often captures influence that analytics cannot see.

Step 5: Turn measurement into action

Measurement is only valuable if it changes what you do next. Review the results by looking for patterns rather than individual answers.

Fix inaccuracies at the source

If assistants repeat outdated pricing or features, update your own pages first, then the most-cited third-party sources such as directory listings and review profiles. Make key facts easy to find and stated plainly, ideally near the top of the relevant page.

Strengthen the pages that answer real questions

AI systems favour content that answers a question clearly and specifically. Comparison pages, detailed FAQs and well-structured guides tend to be quoted more often than vague promotional copy. For online stores, the same principle applies to product and category pages, which is the focus of generative engine optimization for ecommerce.

Earn mentions where AI looks

If the citations for your category point to particular review sites, industry publications or community threads, those are the places where an accurate, consistent description of your brand will have the most influence.

Re-measure on a schedule

Run the same prompt set monthly, or quarterly for smaller teams, and keep the method unchanged so that results stay comparable. Add new prompts as your product and market change, but keep a stable core set for trend tracking.

Common mistakes to avoid

·         Judging visibility from one answer. A single screenshot is not a trend. Always repeat prompts and look at the totals.

·         Testing only brand-name prompts. Asking “What is [brand]?” shows accuracy, but category and comparison prompts show whether you are actually being recommended.

·         Ignoring competitors. Your mention rate means more when you can see how often competitors appear in the same answers.

·         Changing the method every month. Consistent prompts, platforms and scoring are what make the data useful over time.

·         Treating AI visibility as separate from SEO. Clear, accurate, well-structured content helps both, and most improvements serve both goals.

Final thoughts

Measuring brand visibility across ChatGPT, Claude and Google AI does not require an expensive platform to get started. A thoughtful prompt set, a consistent testing routine, a simple scoring sheet and a link to your analytics will already tell you far more than most brands know today.

Start small, measure consistently and focus on the patterns. Within a few cycles, you will know where your brand appears, how it is described and which changes actually move the needle.