You check your visibility in AI answers by preparing a fixed set of questions, split into brand and non-brand, asking each one in a new session of every AI system, and recording for each answer whether your company is mentioned, whether it is recommended and which URLs the system cites. You then repeat the measurement with the same questions to see changes over time.
Tools for website owners show only part of the picture. Google Search Console and Bing Webmaster Tools report impressions, clicks or citations on Google and Bing. There is no such console for ChatGPT, Perplexity or Gemini. So the foundation is a manual (or automated) test with a predefined set of questions.
Which questions should you choose?
Split the questions into three groups:
- Brand questions (2–3). "Who is [company]?", "What do customers say about [company]?" They measure whether the system knows you and describes you correctly.
- Non-brand questions about providers. "Which company in Slovenia offers [service]?" They measure whether the system suggests you.
- Non-brand questions about the problem. Questions a buyer asks before they know any providers: "How do I calculate …", "Is it worth …". They measure whether your content is a source for the answer.
Write the questions in your buyers' language and in their words. Take them from sales conversations, enquiries you have received and queries in Search Console. Once the set is fixed, don't change it between measurements.
How do you run the test?
- Choose the systems. For example ChatGPT with web search, Google AI Overviews, Google AI Mode, Gemini, Perplexity, Copilot and the AI summary in Bing.
- Ask every question in a new session. Logged out or in a private window, and without mentioning the company first. Otherwise conversation history and personalisation affect the answer.
- Record the conditions. Date and time, system and model, language, location and search mode.
- Repeat key questions two or three times. Answers vary between runs; a single run is chance.
- A test that didn't run is not 0%. If the AI summary doesn't trigger or the system is unavailable, mark the test as not run and don't count it among the valid answers.
What should you record for each answer?
| Field | What you record |
|---|---|
| Mention | YES/NO: whether the company is named |
| Recommendation | YES/NO: whether the system suggests it as an option (not meaningful for brand questions) |
| Citation of your domain | The URL of your page, if it is among the sources |
| Other cited sources | All URLs the answer refers to |
| Competitors | Companies the system names |
| Accuracy | Whether the description of the company is correct |
The most useful column is the one with other cited sources. It shows which pages (directories, comparisons, media, competitors' articles) AI systems use as sources for the questions you would like to be answering. That way you see where your content or presence is missing.
How do you calculate the share?
For each question, calculate the share of valid answers with a mention. Then average the questions with equal weight, separately for brand and non-brand.
The example is made up. In one system you have four non-brand questions. In the first, the company is mentioned in one of two runs (50%); in the other three, in none (0%).
Mention share = (50% + 0% + 0% + 0%) ÷ 4 = 12.5%
Calculate the recommendation share and the citation share in the same way. A mention and a citation are not the same thing: a system can mention you without a link, or cite your page without recommending you.
What do Search Console and Bing Webmaster Tools show?
| Tool | What it shows | What it doesn't show |
|---|---|---|
| Google Search Console | Clicks and impressions from AI Overviews and AI Mode in the Performance report (Web search type); a generative AI report with link impressions in these features | A clear separation of clicks from AI answers and regular clicks |
| Bing Webmaster Tools, AI Performance report | Citations of your pages in Copilot and AI summaries in Bing, by page and over time, and the queries that produced them | Clicks |
| Web analytics | Visits from AI system domains such as chatgpt.com and perplexity.ai | Answers in which you were mentioned but nobody clicked |
Data from the consoles tells you what happened once you were already in the answer. The question test tells you whether you were in the answer at all.
What is our baseline result?
We used the same method on ourselves. On 15 and 16 October 2026 we asked 15 questions in Slovenian to seven systems: ChatGPT, Google AI Overviews, Google AI Mode, Bing, Perplexity, Copilot and Gemini.
- Brand questions: PIKA. Digital was mentioned in all seven systems (100%). In Bing, the AI summary triggered only once for brand questions.
- Non-brand questions: the mention share was between 0 and 12.5%. Google AI Overviews, Google AI Mode, Bing and Perplexity 0%, Copilot and Gemini 8.3%, ChatGPT 12.5%.
The questions, counting rules and sample limitations are published in the GEO test methodology. The result shows a typical picture: AI systems know the company when you ask by name, but they don't suggest it to a buyer who is still looking for a name.
What are the most common mistakes?
- Brand questions only. They show 100% and a false sense that all is well.
- One run as the result. The same system can answer the same question differently.
- Testing in a logged-in account with history. The system knows you from earlier conversations; the buyer doesn't.
- Changing the questions between measurements. Comparison over time loses its meaning.
- Counting tests that didn't run as 0%. The share comes out lower than it is.
How can you get a first picture quickly?
If you don't want to run the test by hand, use the free GEO report. You enter your domain and email address, and in about 15 minutes you receive a score for your visibility in AI search, a comparison with competitors, gaps in content and structure, and three priority fixes. Other free tools are collected on the tools page.
Sources
- Google Search Central, AI features and your website: https://developers.google.com/search/docs/appearance/ai-features
- Search Console Help, Generative AI performance report (Search): https://support.google.com/webmasters/answer/16984139