You measure visibility in AI answers by putting a fixed set of customer questions to the same AI assistants at regular intervals and recording every answer. For each observation, note the question, the platform, the date, the settings, whether your business is named and whether your website is linked as a source. Google Search Console and Bing Webmaster Tools add their own figures on their AI features.
What are you actually measuring?
People who talk about “AI visibility” often mix up three different things. Keep them apart, because each one needs its own source of data.
Three things people call AI visibility
| AI mention | Citation | Traffic | |
|---|---|---|---|
| What it is | The assistant names your business, brand or product | The answer links to your website as a source | Someone clicks through and lands on your website |
| Where you see it | In the answer itself, once you record it | In the answer, and partly in Google’s and Bing’s own reports | In your visitor statistics, such as Google Analytics 4 |
| What it proves | That your name was in that one answer | That your page was used as a source | That a visit took place |
An AI mention and a citation can appear in the same answer, but they don’t have to. A mention without a link can still be useful: someone remembers your name and looks you up later. A source link shows where your content was used, but it doesn’t prove a visit. Only your own statistics tell you whether someone actually arrived.
Which bike shop in Zwolle repairs e-bikes from different manufacturers?
Several bike shops in Zwolle repair e-bikes from different manufacturers. De Spaak Cycles lists on its website which motors and batteries it can diagnose and whether you get a courtesy bike during the repair. Call ahead to check that your make is supported.
Bronnen: despaakcycles.example.com, bikes.example.com
Step 1: choose a fixed set of questions
The question set is the foundation of every measurement. Pick questions customers really ask, not the ones that happen to suit you.
Where to find real customer questions
- Questions from customer emails, phone calls and chat
- Quote requests and the doubts people raise in them
- Search queries in Google Search Console
- Questions your sales staff or engineers hear all the time
- Questions customers still ask after they have bought
Then split the set into two groups. Brand questions include your business name (“What does [your business name] sell?”). Non-brand questions don’t (“Which bike shop in Zwolle repairs e-bikes from different manufacturers?”). A brand question already steers the assistant towards your name, so a mention there says little about your visibility. Non-brand questions show whether you appear when someone has no provider in mind yet. Brand questions are still useful for checking whether the assistant gets your opening hours, prices or range right.
Say a bike shop in Zwolle picks twenty questions. Fifteen are about repairs, e-bikes and lease bikes without the shop’s name; five mention the shop. Those twenty questions stay the same for as long as the series runs.
Step 2: record every observation
An AI answer you don’t record is an answer you can’t compare later. For each question and each platform, note at least the following:
| Field | Example |
|---|---|
| Question (word for word) | “Which bike shop in Zwolle repairs e-bikes?” |
| Platform and mode | ChatGPT with search, Google AI Mode, Copilot |
| Date and time | 11 November 2026, 10:15 |
| Settings | Signed in or out, language, country, whether account memory is switched on |
| Businesses named | Which names, in what order |
| Source links | Which URLs the answer shows as sources |
| Is it accurate? | Does it contain errors about your business? |
| Evidence | A screenshot or the full text of the answer |
That last field matters. A spreadsheet full of “yes/no” entries without the answer itself can’t be checked. Keep the text or a screenshot so someone else can verify what the answer said.
Step 3: repeat and compare
A single measurement is a snapshot. AI answers shift, even within one platform. Google says AI Overviews and AI Mode may use different models and techniques, which means the links they show can differ too. With other assistants, the exact question, the context of the conversation and the sources available all play a part.
For your measurements, that means:
How to repeat a measurement so you can compare
- Ask more than onceAsk each question more than once per round and note whether the outcome stays the same.
- Fixed intervalsMeasure monthly, for example, in the same environment and with the same settings.
- Note the changesWrite down what changed between two rounds: a new page, a new model version the platform announced, a reworded question.
- New set, new seriesIf you change the question set, start a new series and don’t compare it with the old one.
How do you read the results?
Treat your measurement as a sample. Take the bike shop: it is named in five of the fifteen non-brand questions. That means five out of fifteen questions examined, on that platform, on that day. It does not mean the shop appears in a third of all AI conversations about bikes in Zwolle. To claim that, you would need to know which questions people really ask and how often, and the platforms don’t release those figures.
So focus mainly on:
- change within the same set: going from five to eight mentions on the same questions is an observation you can discuss;
- which competitors do get named, and which sources the answer relies on for them;
- errors: if the assistant gives an old address or a service you no longer offer, you know which information to correct.
What figures do Google and Bing provide themselves?
Alongside your own measurements, you can use the reports the platforms offer. They count different things from your question set, but the data comes straight from their systems.
Google Search Console. Clicks and impressions from AI Overviews and AI Mode are included in the standard performance report, under the Web search type. Search Console also has a separate generative AI performance report. Google announced it on 3 June 2026 and has made it available to all websites worldwide since 31 August 2026. The report shows impressions: how often links to your site appeared in AI Overviews and AI Mode, broken down by page, country, device and date. Google doesn’t mention clicks for this report. If you can’t see it, Google says your site usually hasn’t had enough impressions in those features yet.
Bing Webmaster Tools. In February 2026, Bing introduced AI Performance as a public preview. The report shows how often your pages are shown as a source in Microsoft Copilot, in AI-generated summaries in Bing and with some of its partners. It also shows grounding queries: the search phrases the AI used to retrieve your content. Bing describes this data as a sample of total activity.
Google Analytics 4. Since May 2026, GA4 has had an AI Assistant channel that groups visits from services such as ChatGPT and Gemini. How to find it and check it is covered in tracking AI traffic in GA4.
What can’t you measure?
An honest measurement also says what stays out of view.
- The total. Nobody outside the platforms knows which questions all users ask. You are measuring your set, not your market share.
- Answers without a click. Someone who reads your name and later goes straight to your site shows up in your statistics as direct traffic.
- Internal systems. In its guide to generative AI features, Google writes that no external tool has access to Google’s internal ranking or AI systems. Be wary of any service that claims to measure them.
- Cause and effect. If you are named more often after publishing a new page, that is an observation. The platform may have changed something itself in the same period.
A technical check of your website doesn’t measure AI mentions either. It shows whether search engines and other services can fetch and read your pages. That makes it a precondition for being cited, without saying whether you are. The checks you can run for it are in the GEO checklist.
How Outseen approaches this
Outseen currently checks AI answers by hand, once a month, in the assistants included in your plan. We agree with you in advance which questions we examine, in which environment and at what moment, and we record the answers we observe. Atlas, the research engine developed by Outseen, investigates the public technical side of your website; that research says nothing about rankings or AI mentions. You can read more about how we work on visibility in AI answers on our GEO page and in what is GEO.
What does this mean for your business?
Start small and keep it going. Choose a fixed set of questions customers really ask, with a clear share that leaves out your business name, and record each month what two or three assistants answer. Always keep the full text or a screenshot. After a few rounds you will see which questions never bring up your business and which incorrect information about you is circulating.
Also use what Google and Bing show you free of charge: the generative AI report in Search Console, AI Performance in Bing Webmaster Tools and the AI channel in GA4. Those figures don’t replace your own question set, but they show which pages already turn up in those platforms’ AI features. How to make the pages themselves easier for AI assistants to find is covered in how do you get visible in ChatGPT?