GEO methodology: how we measure AI visibility
At Elevatec Techlearn, an artificial intelligence agency and fractional CTO, we measure AI visibility by asking ChatGPT, Gemini, Perplexity and Claude the same questions a customer would, via API and with web search turned on. Each answer is scored by where you appear: 100 if you come first, 80 if second, 65 if third, 45 if lower down or only as a cited source, and 0 if you are absent. We repeat the same measurement every 6 to 8 weeks to see what has changed.
1. The questions: the ones your customer would ask
It all starts by writing the questions a real customer would ask an assistant before choosing. They are not loose keywords. They are sentences like "which dental clinic do you recommend in Valencia?" or "I need an agency to set up GoHighLevel in Spain". We write them by sector, by city and in each market's language.
That question bank stays fixed. It is what lets us compare one measurement with the next: if we changed the questions, we would not know whether you improved or we simply asked something else. The free audit uses a short bank; in a full diagnosis, in our cases, it has ranged from 30 to 211 questions.
2. How we run them
We send each question to the four engines through their API, with web search turned on, and store the full answer and the sources it cites. Using the API has an advantage: the query is clean, with no history or preferences from a personal account, and it can be repeated identically two months later.
It also has a limit worth stating. What the API returns with web search is very close to what a user sees in the app, but not identical: the app may use a different default model, the user's location or their history. That is why we talk about a measurement that is comparable over time, not an exact snapshot of what every person will see.
3. How we score
Each answer gets a score based on where the brand appears:
| Where the brand appears | Points | What it means |
|---|---|---|
| First recommendation | 100 | The AI suggests you before anyone else. |
| Second | 80 | You are in the answer, right behind the first. |
| Third | 65 | Still inside the group people usually read. |
| Lower down, or only as a cited source | 45 | It names you or uses you as a source, but does not recommend you in the top spots. |
| Absent | 0 | For that question, you do not exist. |
Each engine's score is the average of its questions, and the overall score is the average of the engines. Beyond the number we record how each AI describes you, who it recommends instead, which sources it uses and whether it confuses you with another company.
4. What we change on your website
- The entity. Full name, what you do, where you are and who is behind it, consistent across the site, with a description that separates you from companies with similar names.
- Structured data. Organization with its external profiles, Service, FAQPage, BreadcrumbList and Article, in both languages if the site has them.
- llms.txt. A plain-text summary for language models, with the company's key facts and its important pages.
- Citable content. Pages that answer one specific question and open with a direct answer, with prices, timelines and limits where they exist.
- Crawler access. HTML that can be read without running JavaScript, a robots.txt that does not block AI bots and an up-to-date sitemap.
- Authorship. Who writes and with what experience, linked to profiles that can be verified.
5. What we change off your website
- Profiles and directories. Google Business Profile, LinkedIn and the directories in your sector, with the same details as the website.
- Real reviews. We help you ask real customers for them at the right moment. We do not buy reviews or write them.
- Third-party mentions. Media, associations or partner websites that talk about you with your full name and a link.
- Consistency. Same name, address and phone everywhere. Models distrust a company that shows up with different details on every site.
The method in practice: clinics in Spain
We have applied this methodology in a pilot with 78 dental implant and aesthetic medicine clinics in Madrid, Valencia and Málaga. The results, the method in detail and its limits are in our pilot study on clinics' AI visibility.
How you can use this method
The free AI visibility audit applies this method with a short bank of questions for your city and sector. The GEO audit with an action plan costs 490 euros (590 USD in the United States) and adds an action plan to the measurement. A full diagnosis, with a custom bank, several markets and follow-up, is quoted at a fixed price after a free diagnosis. It is all on our pricing page.
What we cannot promise
We do not control what an AI answers, and nobody can guarantee you the top spot. Models change version, sources and criteria without warning, and each one re-crawls your website at its own pace. What we can do is fix what depends on you, always measure it with the same questions and show you how it evolves.
Nor do we take shortcuts that could backfire: fake reviews, text hidden for bots or paid mentions dressed up as opinions. More context in what GEO is and how to get ChatGPT to recommend your company.
Frequently asked questions
We use a fixed bank of customer questions sent via API, with web search, to the four engines. Each answer scores 100, 80 or 65 if the brand comes first, second or third, 45 if it appears lower down or only as a source, and 0 if it is absent.
Because the query is clean, with no personal account history, and it can be repeated identically in the next measurement. The result is very close to the app's, but not identical, and we say so in every report.
That the AI names you or uses you as a source, but does not recommend you among the top three. It is better than being absent, because the model already knows you, and it is usually the starting point for moving up.
Every 6 to 8 weeks, with exactly the same questions. Less time does not give the engines a chance to re-crawl the site; more time makes it hard to know which change had an effect.
No. Nobody controls what a model answers. We fix what depends on you, on and off your website, and measure the effect with the same questions.
On the website: the entity, structured data, llms.txt, citable content, crawler access and authorship. Off it: profiles and directories, real reviews, third-party mentions and consistent details everywhere.
Ready to work together?
Schedule a meeting or tell us about your project. We respond quickly, no strings attached.