To track brand visibility in AI search, ask the AI tools your buyers use a fixed set of real buyer questions, save every answer, and log whether your brand is mentioned or cited. Then repeat the same set on a schedule and compare.
By Kristijan GjorgjioskiIndependent SEO consultantUpdated 18 October 2026 · 7 min read
To track brand visibility in AI search, you build a fixed list of buyer questions, run it through the AI tools your buyers use, and record whether each answer mentions or cites you.
Most businesses check once, screenshot a good answer and stop.
That tells you nothing, because AI answers change between sessions and have no position one to defend.
What you can track is a pattern across many prompts, saved as evidence and repeated over time.
Here's the seven-step process I use.
You need very little to begin:
Budget two to three hours for the first round of 20 to 30 prompts across four tools.
Later rounds go faster.
One warning before Step 1: keep this separate from the work of improving visibility. Tracking tells you where you stand. Earning mentions is AI search optimisation, and mixing the two makes both worse. If you want the improvement side first, start with how to get mentioned by ChatGPT.
Your results are only as good as the questions you ask.
Most people type their own brand name into ChatGPT. That checks one thing: whether the tool has heard of you. It says nothing about whether you come up when a buyer has not decided yet.
Start from the questions buyers really ask. Pull them from:
Then sort them into groups like these:
| Prompt type | Example | What it tests |
|---|---|---|
| Recommendation | "Best accountant for contractors in Leeds" | Whether you get suggested |
| Comparison | "Fixed fee vs hourly bookkeeping, which is better?" | Whether your approach is described |
| Cost | "How much does a commercial roof survey cost?" | Whether your pricing content is used |
| Problem | "Why is my GA4 data different from Shopify?" | Whether your guides get cited |
| Branded | "Is [your brand] any good?" | How you are described when named |
Keep branded prompts in their own group.
They answer a different question, and mixing them in flatters the results.
Pick the tools your buyers use, not every tool that exists.
For most businesses I check four:
Add Microsoft Copilot or Perplexity if your audience leans that way.
Buyers in larger companies may have Copilot sitting inside the Microsoft tools they work in all day.
Do not assume one tool predicts another.
In my runs, the same prompt regularly gets a different set of businesses on each surface, because each tool picks its sources in its own way.
Before running anything, write down what counts as "you".
This sounds fussy. It saves arguments later.
List:
Then define the outcomes you will log for every answer.
I use two yes/no fields:
That gives four possible results per answer. Mentioned without being cited means other sites are describing you. Cited without being mentioned means your content is feeding the answer while a competitor gets the recommendation.
Both are fixable, but with different work.
AI answers vary with account, location, history and time.
You cannot remove that variation, but you can stop adding to it.
For every round:
Run your most important prompts more than once. If you appear in two of three runs, that is a real signal. If you appeared once and never again, it was noise.
Pro tip: platforms throttle repeated questions. If a tool starts refusing or slowing down, stop and resume later rather than switching accounts mid-round.
A result you cannot show someone did not happen.
Save three things per prompt, per tool:
Your sheet should look something like this:
| Date | Tool | Prompt | Mentioned | Cited | Competitors named | Sources used |
|---|---|---|---|---|---|---|
| 2026-10-14 | ChatGPT | Best bookkeeper for builders | No | No | Firm A, Firm B | Directory X, Blog Y |
| 2026-10-14 | Gemini | Best bookkeeper for builders | Yes | No | Firm A | Your review page |
The "sources used" column is the one people skip, and it is often the most useful. It shows which third-party sites the tools trust in your market. If the same directory or review site keeps appearing, getting accurately listed there is a job for off-page SEO.
Your prompt runs show what buyers see. Platform data shows what happens around it. Add three sources.
Google says traffic from AI Overviews and AI Mode is counted in the Search Console Performance report, under the "Web" search type, as part of your overall search traffic.
That is stated in its guide to AI features and your website.
For impressions, there is now a dedicated view. The Generative AI performance report counts how many times links to your site were shown in AI Overviews and AI Mode, and lets you group them by page, country, device or date.
It does not report clicks, and it does not tell you which of the two features showed the link. So AI Overview clicks still cannot be pulled out on their own.
I use it to see which pages Google's AI features surface, then check those pages' topics in my prompt runs to see what the answers actually say.
If the report is missing, Google says your site usually has not had enough impressions in these features yet. It also shows nothing if the property's Search generative AI setting, under Settings, has been switched from the default of include to exclude.
Bing launched an AI Performance report in public preview in February 2026.
It counts how often your pages are cited across Copilot, Bing's AI summaries and some partner integrations, and lists the grounding queries behind those citations.
Citations there are not clicks.
They are still the closest thing to a first-party citation count any platform offers.
In GA4, look at referral traffic from AI tool domains such as chatgpt.com, gemini.google.com, perplexity.ai and claude.ai.
ChatGPT also tags many of its links with utm_source=chatgpt.com.
Treat these numbers as a floor. Visits from apps often arrive with no referrer and land in Direct. If your GA4 numbers already look odd, my guide to why GA4 data does not match explains the usual causes.
One round is a snapshot.
Two rounds are a trend.
Rerun the same prompt list monthly, or after any big change: a new service page, a site migration, a PR push.
Change the list only between rounds, and log what changed.
When you compare rounds, look for patterns rather than single answers:
Each pattern points to a different fix. Missing from whole prompt groups usually means you lack a page that answers those questions, which is a content strategy gap; my guide on appearing in Google AI Overviews covers that side. Cited but not mentioned often means your pages explain the topic well but never make the case for your business.
If you are unsure which numbers deserve a place in a report, I break them down in AI visibility metrics worth measuring.
You do not need a tool or a budget to begin.
Write 20 buyer questions, run them through four AI tools, and log mentioned, cited and sources for each answer.
That first round will show you more than any visibility score.
When the list grows past what you can run by hand, or you want it repeated every month without the screenshots eating a day, that is the job my AI search visibility tracking pipeline does: it runs your prompts in a real browser on each surface and hands you every saved answer.
For the bigger picture of how prompt tracking fits with the rest of the research tooling, see the AI visibility tracking service page and the wider SEO automation overview.
Yes, at a small scale. A spreadsheet, a browser and a screenshot tool are enough to run 20 to 30 prompts a month by hand. Search Console, Bing Webmaster Tools and GA4 are free and add the data the platforms share. Paid tools and automation become worth it when the prompt list grows past what one person can run in an afternoon.
Start with 20 to 30 prompts covering your main services, locations, comparisons and cost questions, plus a handful that name your brand. That is enough to see patterns without drowning in screenshots. Add prompts when you launch a service or notice buyers asking a new kind of question on sales calls.
AI tools generate each answer fresh, so wording, sources and recommendations can change between sessions, accounts and locations. That is why a single check proves little. Running a fixed prompt set in a consistent setup, and repeating it, shows which results are stable and which were one-off variations.
Some of it does. Visits where the browser passes a referrer show up as referral traffic from the AI tool's domain, and ChatGPT adds a utm_source=chatgpt.com tag to many of the links it shows. Visits from apps or copied links can arrive with no referrer and land in Direct, so GA4 undercounts AI-driven visits.
About the author
Kristijan Gjorgjioski
Independent SEO consultant · 6+ years in SEO · 60+ client websites worked on
I'm an independent SEO consultant with 6+ years in SEO. I run audits, migrations and content programmes for businesses in Europe, the UK and the USA, and I build the automation behind my own research.
Book a free 30-minute call. I look at your site first and tell you plainly what I would fix, and in what order.