What is generative engine optimization (GEO)?
How AI assistants choose which sources to cite, and what you can change.
On this page
Short answer
AI visibility is how often AI assistants bring up your company when buyers ask them about your market. You measure it by putting a fixed list of buyer questions to ChatGPT, Perplexity and Google AI Overviews, several times each, and counting the answers that name you or link to your site. You raise it with clear pages on your own site and with mentions on the sites these assistants read.
Mentions, citations and share of answers
Three counts make up the picture. Track them apart, because each one moves for its own reasons.
- Mention. Your company name appears in the answer text. A buyer who reads “look at A, B or C” sees you even without a link.
- Citation. The answer links to one of your pages as a source. An assistant can link your article and still recommend a rival, so a citation is not the same as a recommendation.
- Share of answers. The part of all answers in your test where you show up. Ask 10 questions three times, get named in 9 of the 30 answers, and your share is 30 %.
We also check accuracy: does the assistant get your prices, services and location right? Being named with last year’s price can cost you more than being left out.
Why one answer tells you little
The same question gets a different answer on the next try. In our November 2026 study we asked ChatGPT the same agency questions three times each. The three lists of named agencies shared only 17 % of names on average (Jaccard index). Country matters too. When we asked the same question from other countries, the list overlapped with the US list by only 9 to 21 %.
So one good screenshot is not a result, and one bad answer is not a verdict. You need many answers to the same questions, collected the same way every time.
How to measure AI visibility
- Write the question set
Take 10 to 30 questions in the words your buyers use. Mix three kinds: category questions without your name (“best payroll software for small teams”), questions with your name (“is [brand] good for agencies”) and comparisons (“[brand] or [rival]”).
Read more
The first kind counts most. That is where a buyer who has never heard of you builds a shortlist.
- Pick the engines
We use ChatGPT with web search, Perplexity and Google AI Overviews. Add Copilot or Gemini if your buyers use them.
- Fix location and language
Ask from the country, and the city if you sell locally, where your buyers are. Use a clean session with memory off, so your past chats do not tilt the answers toward you.
- Run each question several times
We do three runs per question in each engine. With fewer runs, the count jumps around by chance.
- Log every answer the same way
For each run, write down: named or not, linked or not, which URL, who else was named, which sources were cited, and whether your facts were right.
Read more
A spreadsheet is enough to start.
- Repeat every month with the same set
If you change the questions, you lose the comparison. New questions go into a separate group, and the old ones stay.
Google AI Overviews need a slightly different routine. There is no chat: you search the query in Google and note whether an overview appears and whether it links to you. In our study an overview appeared on 146 of 155 results pages, including commercial “best agency” searches. So track Google overviews for sales queries too, not only for “what is” questions.
ChatGPT also rewrites your question into its own searches. In our study it added the current year to 44 % of them. That is why the log keeps the cited sources, not only the final text.
What webmaster tools show
The search engines’ own webmaster tools give part of the picture. Bing Webmaster Tools has an AI Performance report, in public preview, that shows when your site is cited in Copilot and in Bing’s AI summaries. Google Search Console counts appearances in AI Overviews inside its usual Performance report, together with regular results under the Web search type, and does not show them as a separate line. Neither tool covers ChatGPT or Perplexity, so the question set stays the main method.
Track them apart, because each one moves for its own reasons.
Metrics we report
| Metric | How we count it | What it tells you |
|---|---|---|
| Named rate | answers that name you ÷ all answers, per engine | how often buyers see your name |
| Linked rate | answers that link to your site ÷ all answers | how often you serve as a source |
| AI Overview coverage | queries with an overview ÷ tracked queries | how much of your demand Google answers with AI |
| Overview links to you | queries where the overview links you ÷ tracked queries | your place among Google’s AI sources |
| Rivals named | how often each competitor appears | who holds the spot you want |
| Top sources | domains cited most often | which outside sites to work on |
| Fact errors | answers with a wrong price, service or place | what to correct on and off your site |
Keep every number per engine. A blended “AI score” hides it when engines move in different directions, as they did on our own site.
What moves AI visibility
On your site
- Lead with the answer. The opening two sentences of a page should state the service, the buyer, the place and the price. Assistants tend to quote that part.
- A page for each service and topic. Assistants search with short, literal phrases. “Payroll for small teams in Texas” matches them. “Our solutions” does not.
- Facts as text. Prices, service areas, founding year and team size belong in the page copy, not in pictures, PDFs or sliders.
- Let search bots in. Check robots.txt and the bot rules of your CDN or firewall. OpenAI’s OAI-SearchBot and the crawlers of Perplexity and Bing need access. A “block all AI” rule copied from a forum can take you out of search answers.
- Indexed in Google and Bing. Google links pages in AI Overviews only if they are indexed and allowed to show a snippet. ChatGPT search also relies on outside search partners, so Bing indexing is worth checking too.
More on page layout is in our guide to AI citations and our ChatGPT SEO guide.
Off your site
Assistants read what other people write about your market, and that often weighs more than your own pages.
- Round-ups and comparisons. Find the “best X in Y” articles that rank for your category and write to the authors with a reason their readers would care about you.
- Directories and review sites. One full profile on a site that shows up for your searches is worth more than a dozen empty ones.
- The same facts everywhere. If your site says one price and a directory says another, an assistant may repeat either.
- Discussions. Forums and Q&A threads get read too. Post under your own name, and only where you can add something useful.
Our own numbers
We track aoe.kz, our agency site in Almaty, with Russian-language questions. ChatGPT and Perplexity get five “recommend an agency in Almaty” questions, three runs each, through the API with web search on and location set to Kazakhstan.
| What we counted | Oct 6, 2026 | Nov 2, 2026 |
|---|---|---|
| ChatGPT names AOE | 5 of 15 | 12 of 15 |
| Perplexity names AOE | 6 of 15 | 5 of 15 |
In Google, AI Overviews linked to aoe.kz in 0 of 17 tracked queries on September 22, 2 of 20 on October 6, and 2 of 17 collected on November 2.
ChatGPT moved a lot in one month. Perplexity stayed where it was. We edited pages and earned mentions over the same weeks, so we cannot tie the rise to a single change. This is one market and a small set of questions. It shows the method at work, not what will happen in your market.
A blended “AI score” hides it when engines move in different directions, as they did on our own site.
Common mistakes
- Testing in your own account. Memory, chat history and location lean the answers your way.
- One run per question. When runs share only 17 % of names, a single answer can point you the wrong way.
- A new question set every month. Then the months cannot be compared.
- Counting mentions and citations together. They rise and fall for different reasons.
- Asking the assistant how visible you are. That gets you its opinion, not a count.
Questions
What is a good AI visibility score?
There is no shared benchmark. Every question set gives its own baseline. Compare your numbers with your own last month, and with the rivals named in the same answers.
Can a tool track AI visibility for us?
Several tools run question sets on a schedule. Check how many runs per question they make, which country they ask from, and whether they save the cited sources.
How often should we measure?
Once a month with the same set suits most companies. After a big change to a key page, an extra check a week or two later shows whether Google has started linking it.
Can you measure it for us?
Yes. Our free AI visibility check asks 10 buyer questions once each and sends you a table. The AI visibility audit asks 30 questions in ChatGPT, Perplexity and Google AI Overviews, three runs each, and ends with a ranked list of fixes. Monthly tracking and the fixes themselves are part of our AI search optimization work.
