What questions should you ask the AIs?
The ones a customer would ask before choosing: who recommends, how much it costs, how it compares, and the company's name. Write between 5 and 10 questions of three types:
- Discovery: "what is the best [what it does] in [city]?"
- Comparison and price: "how much does [what it sells] cost in Portugal?"
- Name: "who is [the company name]?"
Write them once and always use the same wording. If you change the words from one measurement to the next, you are no longer comparing the same thing.
Why three readings and the 2-in-3 rule?
Because answers vary from reading to reading. One reading is noise; what repeats in two of three is a signal.
Ask each question three times, each one in a new conversation and, if possible, without being signed in, so the AI does not use your history. Count the company as "present" only if it appears in at least two of the three. A single appearance does not count, either for or against. This is the method we use for Rui's businesses, and the measurements, with dates, are in SEO for AI.
Which AIs should you measure?
Several, because each one gets its information from different sites and answers differently. Measure ChatGPT, Perplexity, Gemini, Copilot and Google's AI answers, that is, AI Overviews and AI Mode (see AI Overviews and AI Mode).
Two cautions:
- The ChatGPT you use on the site may answer differently from the version you query through programming (API). To measure what customers see, use the consumer version.
- Not all AIs can be measured through programming. Some can only be checked by hand, with the session of the person asking the question.
What should you record in each reading?
Always record the same fields, so you can compare from one month to the next.
| Field | What for |
|---|---|
| Question and AI | Know what was measured, and where |
| Date | Answers change, so the date is part of the result |
| Appears (yes or no), per reading | Apply the 2-in-3 rule |
| Position in the answer | See whether it is first, second or just one more |
| How it describes the company | See whether the description is right and the same as yours |
| Sources cited | See where the AI gets its answers |
Sources are the most useful field: they show where you need to be. If the AIs cite directories and third-party lists, that is where the company has to appear. What to do next is in how to appear in ChatGPT.
What other signals help you measure?
Search Console shows traffic coming from Google, and GA4 shows the visits arriving from AI sites, such as chatgpt.com or perplexity.ai. For AI Overviews, Google says that traffic is included in the "Web" search type in Search Console.
The most direct signal is to ask the people who contact the company: "how did you hear about us?" It does not replace measuring the AIs, but it shows whether customers already come from there.
What is llms.txt and do you need it?
llms.txt is a proposed convention: a Markdown file at the root of the site, at /llms.txt, with a summary and links to the most useful pages, written for AIs. It was proposed by Jeremy Howard in September 2024 (source: llmstxt.org, accessed November 15, 2026).
It is not a standard that AIs are required to read, and Google says no special files are needed to appear in its AI answers (see AI Overviews and AI Mode). It costs little, and our site has one. Do not treat it as a solution: what counts is a clear site and a well-described company outside it.
How long does it take to see results?
There is no guaranteed timeframe: it depends on when each AI next reads the sources. That is why you measure every month, with the same questions, and compare with the previous measurement. Be wary of anyone who promises you a position.

