How to Measure Your Brand’s Visibility in AI Search

Measuring your brand’s visibility in AI search takes three sources of evidence. The first is a fixed set of prompts you check regularly in ChatGPT, Google’s AI Overviews, Gemini, Perplexity and Copilot. The second is your analytics data for visits those tools send you, and the third is a read-through of the answers to see if they describe you accurately. No single dashboard covers all of it, so aim for a sample you can repeat the same way each month and watch the trend. A perfect score is not on offer.

Why AI visibility is harder to measure than rankings

Traditional SEO measurement rests on a fairly stable idea. A query returns a ranked list, and you can see where you sit in it. AI search breaks that in several ways.

• Answers are written fresh each time. An AI tool generates a response and may mention, recommend or cite your brand, or leave you out, differently on each run.
• Results vary. The same prompt can produce different answers on different days, for different users, in different locations, and depending on earlier messages in the conversation.
• Prompts are longer and more personal. People ask full questions with context (“I run a small bakery in Cardiff and need a new website, who should I talk to?”), so there is no neat keyword list to track.
• Some influence leaves no click. Someone may read an AI recommendation, then search your name or visit directly. The AI tool shaped the decision, but analytics may credit branded search or direct traffic.
• Platform reporting is limited. Google Search Console includes AI Overviews and AI Mode activity within its overall web search figures rather than reporting it separately, and most AI assistants offer site owners little or no reporting.

None of this makes measurement pointless. You just have to work in samples and trends, accept some noise, and link what you can see to leads and sales.

The metrics that matter

Before collecting data, decide what you want to learn. For most businesses these are the most useful measures.

Mention rate and share of voice tell you about reach, accuracy tells you about risk, and referral traffic and conversions tell you what it is worth. You need all of them before you can decide where to put effort.

Step 1: Build your prompt set

Everything else hangs off your prompt set. It is a fixed list of questions, phrased the way real customers might ask about your category, that you check on a schedule.

Where to find prompts customers would actually use

• Customer conversations. The questions people ask your sales and support teams are the best source. Write down the context they give as well as the keywords.
• Search Console queries. Longer, question-style queries show the language people already use.
• People Also Ask and related searches for your core topics.
• Your own services and locations. Combine what you sell with where and for whom you sell it.

Cover the full journey

Group prompts by intent so you can see where you are strong and where you are missing.

• Category and problem prompts: “How do I stop my website going to spam in Gmail?”
• Recommendation prompts: “Best accountants for small businesses in Swansea.”
• Comparison prompts: “Is a fixed-price website better than a monthly subscription?”
• Brand prompts: “What does [your brand] do?” and “Is [your brand] any good?”

Brand prompts get overlooked. They show what AI tools say about you to someone who has already heard your name and is checking you out before getting in touch.

For a small business, 20 to 50 prompts is usually enough to show patterns without eating your week. Larger organisations with many products or locations may need more and will probably want a tool to manage them. Keep the wording fixed between checks, because once you change the prompts the trend line stops meaning anything.

Step 2: Run your checks the same way every time

Because answers vary, how you check matters as much as what you check. Follow the same routine each time.

1. Choose your platforms. Focus on the tools your customers are most likely to use. For most UK businesses that means Google (AI Overviews and AI Mode), ChatGPT, and at least one of Gemini, Copilot or Perplexity.
2. Use a clean session. Log out, or use a fresh chat with memory and personalisation off where possible, so your own history does not skew results.
3. Note location settings. Local recommendations depend heavily on location. Record where the check was run from.
4. Run each prompt more than once. Two or three runs per prompt gives a fairer picture than a single answer. Record if you appeared in each run.
5. Record the details. For each answer, note if you were mentioned and in what position, which competitors appeared, which sources were cited, and anything said about you that was wrong.
6. Save evidence. Keep copies or screenshots of notable answers, particularly inaccurate ones, so you can tell later if they have improved.
7. Repeat on a schedule. Monthly checks suit most small businesses. Weekly checks make sense only if you are actively making changes and have the capacity to review them.

A spreadsheet is enough. Use one row per prompt per platform per run, with columns for mentioned (yes/no), cited (yes/no), position, competitors named, cited URLs and accuracy notes. From that you can calculate your mention rate and share of voice each month.

Should you use a tracking tool?

A growing number of SEO platforms and specialist tools now automate prompt tracking across several AI assistants. They save time and handle larger prompt sets, which starts to matter once you are past a few dozen prompts. Treat their numbers as estimates. Each tool samples answers in its own way, often through programmatic access that may not match what a logged-in user sees. Compare a tool’s findings with your own manual spot checks before relying on it.

Step 3: Track AI referral traffic in analytics

When an AI tool links to your website and someone clicks, the visit can show up in your analytics with a referring domain. In Google Analytics 4 you can see this in the traffic acquisition report by looking at session source.

Sources to look for

Common referrers include chatgpt.com, perplexity.ai, gemini.google.com, copilot.microsoft.com and claude.ai. Domains change over time as products are renamed, so review the list every few months rather than treating it as fixed.

Create an AI channel group

By default, GA4 usually lumps these visits into “Referral”, where they are easy to miss. You can create a custom channel group with a channel called something like “AI assistants”, defined by a source condition that matches the domains above using a regular expression. Place it above the Referral channel so it takes priority. You then get a separate line for AI-driven sessions, engagement and conversions that you can follow month to month.

Know what analytics will miss

• Google AI Overviews and AI Mode clicks generally arrive as Google organic traffic, indistinguishable from normal search clicks in GA4.
• Some app-based clicks do not pass a referrer and appear as direct traffic.
• Influence without clicks, where someone later searches your brand name or types your URL, cannot be attributed to the AI answer directly.

Because of these gaps, treat AI referral traffic as a minimum. Branded search impressions in Search Console and direct traffic are useful to watch alongside it, although both also move for plenty of other reasons.

Linking this data to leads and sales works the same way as for any other channel. Our guide on how to measure marketing ROI explains how to tie channel activity to revenue.

Step 4: Check accuracy and sentiment

Being visible is worth little if AI tools get the facts wrong. An answer that lists a service you stopped offering, quotes old prices or puts you in the wrong town can cost you enquiries without you ever knowing.

Go through the answers to your brand prompts and look for:

• Wrong services, locations, opening hours or prices.
• Outdated company details, such as a former name or an old address.
• Confusion with a similarly named business.
• Negative framing based on old reviews or a single complaint.
• Missing strengths, such as specialisms or accreditations you do hold.

When answers cite sources, check what those sources are. Inaccuracies usually trace back to something on the web, such as an old directory listing, an outdated page on your own site or a Google Business Profile that disagrees with your website. Fix the source and the answer usually follows, given time.

Step 5: Find out which sources drive your visibility

When AI tools cite sources, they show you what they lean on. Over several months, look for patterns in the cited URLs across your prompt set.

• Your own pages. Note which ones get cited, and for which prompts. Keep those pages current and be careful when changing their URLs.
• Third-party pages. Directories, review sites, comparison articles, trade bodies and local news. If competitors keep getting mentioned through a source where you are absent, that is an obvious place to start.
• Gaps. Prompts where nobody is cited well, or the answer is vague, point to topics where a clear page written by someone who knows the subject could fill the hole.

This analysis works much like a traditional competitor review, and the method in our guide to running a competitor analysis adapts well to AI answers.

You can also check your server logs for visits from AI crawlers, such as OpenAI’s GPTBot and OAI-SearchBot, PerplexityBot and ClaudeBot. Crawler visits show that a tool can reach your pages. They say nothing about your chances of being mentioned. They do confirm that your robots.txt or firewall is not blocking crawlers you want to allow.

Turning the data into a one-page report

Measurement only helps if it changes what you do next. For a small business, a monthly or quarterly AI visibility report can fit on one page and cover these points.

• Mention rate and share of voice across your prompt set, compared with the last period.
• Results by intent group, for example strong on brand prompts and weak on recommendation prompts.
• Differences between Google, ChatGPT and the other platforms.
• Anything inaccurate, where it probably came from and the fix in progress.
• Referral traffic and conversions from AI assistants.
• Two or three specific changes for the next period.

Keep expectations in check. Small prompt sets produce noisy numbers, and a swing of a few mentions may be nothing more than variation. Wait until you see the same direction over several periods before drawing conclusions. If you are setting targets, the approach in how to set marketing goals keeps them achievable and tied to business outcomes. Our marketing analytics and reporting service can build this reporting alongside your other channels.

An illustration

Imagine, purely as an example, a Bristol-based kitchen fitter tracking 30 prompts across three platforms. After a quarter, the spreadsheet shows they appear regularly for brand prompts but rarely for “best kitchen fitter near me” style recommendations, and that the answers mentioning them cite an old directory listing with a previous address. The next steps are obvious. Correct the directory listing, improve the location and service pages, and ask recent customers to leave reviews on the platforms being cited. The following quarter’s checks then show if any of it made a difference.

Mistakes to avoid

• Judging from one answer. A single check tells you almost nothing about visibility.
• Testing only from your own account. Personalisation and memory can make you appear more visible than you are.
• Changing prompts every month. Without fixed prompts, you have no trend to read.
• Counting mentions and ignoring accuracy. Being mentioned with the wrong information can do more harm than not being mentioned.
• Expecting precise attribution. Much AI influence happens without a trackable click.
• Measuring without acting. Each report should lead to specific improvements to content, listings or reputation.

Setting up your first month of tracking

The methods for measuring AI search are still settling down, but the basics will not change much. Write your prompt set this week, add the AI channel group in GA4, and run your first round of checks before the month is out. A few months of that will show you if you are being found, how you are described and if any of it turns into enquiries.

We set up prompt sets, analytics and the monthly reporting for clients through our AI visibility strategy service. Nobody can promise you AI citations, so we focus on accurate measurement and steady fixes instead.

Share:LinkedInXEmail