ChatGPT and Gemini mostly cite pages that answer a customer question directly, contain their own facts, figures or evidence, and are indexed and readable. An AI answer names only a few sources, and anyone missing from them doesn't appear in the answer. That is exactly what AI visibility is about.
AI visibility is the share of answers from ChatGPT, Gemini, Perplexity and similar AI answer engines in which your brand is named or your page is linked as a source. It isn't decided in one place, but in four.
Key takeaways
- Check your brand at four points of an AI answer: question, source selection, answer, citation. If it is already missing at source selection, better copy won't help; the technical side comes first.
- Write pages that answer a customer question directly in the first sentences and contain your own facts, figures and evidence. Interchangeable texts are probably cited less often.
- Treat SEO as the foundation: what isn't indexed and crawlable can't be cited by any AI.
- Measure with fixed questions that you ask several times. A single test says little, because the same question can be answered differently each time.
How does a brand end up in an AI answer in the first place?
An AI answer comes about in four steps: question, source selection, answer and citation. Your brand can drop out at any of them. The best way to see this is to follow a single customer question: "Which software creates quotes and invoices for an electrical business with ten employees?" Let's assume you sell exactly this kind of software.
Step 1: The question
Customers phrase their questions at length and in specific terms, with industry, size and goal. Depending on the system, the AI splits this question internally into several sub-questions and looks for material on each. Your brand can only appear if your pages match these sub-questions: writing quotes, creating invoices, trades business.
Action: As a rule of thumb, collect 10 to 20 real customer questions from sales, support and emails. Write them down the way customers say them, not the way your product team names things.
Step 2: Source selection
This is where a brand can drop out silently. The AI draws on pages it can find and read. If your service description exists only in a PDF price list, or its content appears only after a script loads it, it may be missing from the selection, no matter how good it is.
Action: Check whether the page for the question is indexed, whether your robots.txt locks crawlers out, and whether the core text is in the HTML.
Step 3: The answer
The model writes the answer from the sources it has chosen. What matters is that your brand is unambiguous as an entity: name, offer and target group appear in one clear sentence on your page and match what third parties write about you.
Action: Start your main pages with a sentence along the lines of "X is software for Y that solves Z."
Step 4: The citation
At the end there is either a mention in the text or a link to your page, sometimes both. What can be lifted out cleanly is probably more likely to be cited: a clear paragraph, a table, a figure with evidence.
Action: As a rule of thumb, make one or two statements per page citable, meaning they are understandable on their own and come with a figure or evidence.
Which content do AI systems cite, and which do they skip?
AI systems cite pages that answer a question directly and contain something that is missing elsewhere: their own figures, evidence, expert statements.
In the electrical business example, the page that says how long it takes to create a quote with the software and what happens automatically along the way probably has the better chances. The page that says "Our innovative solution for your success" will probably be skipped.
What gets skipped includes, for example:
- texts that repeat a keyword but make no statement,
- generic AI texts without their own facts, which any competitor could have written just as well,
- pages where the answer only comes in the fourth paragraph.
Then there is the wider context: systems probably also draw on sources outside your domain, such as review portals, forums and trade media.
Why is classic SEO still the foundation?
Without indexing, clean technology and a page that matches the search query, your brand never even makes it into source selection. SEO and GEO are therefore not opponents.
A hypothetical example shows the gap: a business ranks in the top 10 for "quoting software for trades" but isn't named in AI answers. The reason can be the text: it opens with a company history instead of the answer and contains no verifiable figure.
Anyone who knows the search results page for the customer question can see which pages already serve as sources.
How do you measure AI visibility instead of rankings?
You measure AI visibility with a fixed set of questions that you run through the systems regularly and several times. A single test isn't enough, because answers depend on chance and personalization. Only the average over many runs tells you anything.
Three steps are enough to start:
- Define 20 questions that customers really ask, in every language of your markets.
- Ask each question several times in ChatGPT, Gemini and Perplexity and note which brands and links appear.
- Repeat this at the same rhythm and compare only values for the same questions.
This gives you four metrics: the mention rate (share of answers that name your brand), the citation rate (share with a link to your domain), the share of voice compared with direct competitors, and the tone of the mention. Keep tables of raw data at hand so you can trace every number and don't have to rely on an index value that nobody can check.
Who is AI visibility worth it for, and when not yet?
AI visibility is worth it when customers ask questions that need explaining before they buy and you can answer those questions with your own knowledge. That applies to software, consulting, specialized trades and B2B offers, especially in several markets.
It isn't worth it yet in three cases:
- Your website has hardly any indexed pages. SEO comes first here.
- You don't yet have a clear offer that can be described in two sentences.
- Your customers decide almost only on price or location and don't ask questions beforehand.
Small agencies face an extra task: for each client, they have to answer whether the client's site shows up in Google AI Mode and AI Overviews. For this we offer the solution for agencies with GEO and SEO checks per client, Bing Webmaster Tools connection and publishing for up to 10 client sites.
How do you find out where you stand today?
Start with an audit that checks technology, content and AI citation together before you write a single text.
UPGeoSEO offers a free SEO and GEO audit for this, which analyzes your website and shows where you stand in classic search and in AI answers.
Frequently asked questions
Do I have to optimize separately for ChatGPT, Gemini and Perplexity?
At the core, no. Clear answers, your own facts, clean technology and mentions by third parties should help in all systems. You should still measure separately, though, because each system may prefer different sources.
What should I do if an AI gives false information about my company?
First find the source of the error: for example an outdated page or a directory entry, not just your own website. Correct what you can change there, and put the right information as a clear sentence on a page of your own. Then check again with the same questions. There is no guarantee that the model will adopt the information right away.
Is it enough to translate existing texts into other languages?
Not without adaptation. Customers in other markets phrase their questions differently, name different competitors and expect different terms. A pure translation often misses these questions. So check a separate list of real customer questions for each market and write the pages natively for them.
