GEO, AEO, AI visibility: engines cite the clearest, not the most visible
AI visibility, or GEO (Generative Engine Optimization), covers the practices that make a company visible and correctly cited in AI answers: ChatGPT, Perplexity, Google’s AI Overviews, Gemini. At Fast Growth Advisors, GEO extends our lifelong craft: a sharp message, structured pages, citable proof. Because answer engines do not rank pages, they extract statements.
What AI does with your sentence
“AI platform to optimise your processes”
No identifiable question. No named prospect. No metric. The AI has nothing to extract.
“We cut time-to-hire by 45% for staffing agencies with more than 50 recruiters.”
Quantified benefit, named target, precise segment. The AI knows which question you answer.
In short
- •The Fast Growth Advisors AI Visibility (GEO) offer makes your company citable by ChatGPT, Perplexity and Google’s AI Overviews.
- •Measured: on a typical B2B page, half the text survives extraction, but none of the six pieces of proof. The free diagnostic includes a citability test, answers archived.
- •Three axes decide your presence in generated answers: message clarity, the technical structure of your pages, and what third-party surfaces say.
- •Measured: 876 of the 1,129 requests declaring themselves an AI assistant on our servers were proven impersonations. Your AI visibility figures may well be wrong.
- •The no-commitment entry point is the free diagnostic.
Does Google cite you in its AI Overviews?
When Google writes the answer itself at the top of the results page, it keeps a handful of sources and drops the rest. We measure that choice, question by question, and we hand back the link that lets you run the search again.
Three outcomes are possible on any given question, and we never merge them: Google cites you, Google answers without citing you, or Google does not answer this question at all. The third one is not a failure of your site.
| What the measurement finds | What we write | Is it a fault of your site? |
|---|---|---|
| Google generates an answer and cites your page | Google cites you | No. Worth protecting |
| Google generates an answer without citing your page | Google answers without citing you | Yes. This is the finding that matters |
| Google generates no answer at all | Google does not answer this question | No. There is nothing to measure |
Most tools on the market merge the last two states, because all they read is the text of the answer. They then report an invisibility that nothing established. A question Google does not answer tells you nothing about your site: it is an absence of measurement, and we write it as such. An answer that cites three of your competitors and not you, on the other hand, is a finding, and that one can be fixed.
The scope, stated plainly: this measurement covers Google AI Overviews. Not ChatGPT, not Perplexity, not browsing agents. Those are measured differently, and we do that too. Every reading returns the state of the question, the text Google wrote, the pages it cites and the link to run the search yourself. We are currently establishing our own baseline on the UK market, and we will publish it here.
What does an engine keep from your page?
In an AI-composed answer, not being cited means not existing at all. That leaves one question: what does an engine actually keep from a page? We built a fictional B2B startup homepage carrying the most common defects: result figures, client logos, testimonial and pricing injected by JavaScript. Then we submitted it to the three extractors that prepare the corpora of large models.
share of text kept share of proof kept
| Reading | Bytes of text | Share | Proof kept |
|---|---|---|---|
| Seen by a human | 1,450 | 100% | 8 / 8 |
| trafilatura | 706 | 49% | 1 / 8 |
| jusText | 796 | 55% | 2 / 8 |
| Resiliparse | 949 | 65% | 2 / 8 |
Half the text survives, which reassures. None of the six proofs survives, which condemns. What gets through is the promise, and an old positioning left hidden in the code during a redesign. An engine can then say what the company does, never what sets it apart. The full protocol in our white paper →
How do AI engines choose their sources?
Three axes decide your presence in a generated answer. We measure them separately, because they are fixed separately.
What gets read
An AI handles the implicit very poorly. If your pages do not state in black and white who you serve, what you do and what sets you apart, they are unusable in an answer.
Concretely: questions asked the way your prospect asks them and their answers, quantified ROI proof, named client quotes. Measured: 0.91 out of 5. That is the weakest of the Observatory’s fifteen sub-criteria, 18% of the potential.
What stays invisible
Everything that must be in the page without a human reading it: Schema.org structured data aligned with the content type, llms.txt and llms-full.txt, crawl directives, declaration in Search Console and Bing, sitemap and the technical SEO that carries the rest.
It is half the work, and the half almost nobody does. Measured: 61% of the Observatory’s detailed audits have their value figures in the press, not on their own site.
What others say
AI engines cross-reference third-party surfaces heavily: Reddit, G2, Trustpilot, Gartner, the trade press, Q&A spaces.
Which ones matter depends entirely on your market: a SaaS vendor does not play out on the same surfaces as a manufacturer. Measured: a third of the queries ChatGPT issues for a French-language question go out in English. The surfaces that count are not always in the reader’s language.
And we check for real. Not a score in a dashboard: your category’s questions are asked to ChatGPT, Perplexity and Google AI Overviews, before and after, answers archived. You are cited, or you are not.
Which language does an engine search in on your behalf?
When a user asks a question, the engine does not search for the question: it breaks it down into queries it issues itself. Fast Growth Advisors asked 30 questions across 15 sectors, in French, to three engines through their official interfaces, and read the queries actually issued. Total cost of the study: 4.95 dollars.
| Engine | Queries issued | In French | In English | Mixed or undetermined |
|---|---|---|---|---|
| ChatGPT | 96 | 40 | 34 | 22 |
| Claude | 52 | 34 | 3 | 15 |
| Gemini | 44 | 35 | 0 | 9 |
A third of ChatGPT’s queries go out in English for questions asked in French, and the share rises on technical B2B subjects. Two consequences follow. If you sell into non-English markets, your English pages are read by engines answering local buyers, so they are not a secondary asset. And if you sell in English, your content is the reference literature those engines fall back on, which is an opening in every market where local material is thin.
Are your AI visibility figures true?
A bot is whatever it declares itself to be: the user agent is a free text field. Fast Growth Advisors verifies the identity of every bot, through published address ranges and reverse resolution confirmed in both directions. Over thirteen days, 876 of the 1,129 requests presenting themselves as an AI assistant were proven impersonations, and the 53.7% error rate a non-verifying tool would have displayed drops to 0.0% on authenticated requests alone.
| Identity verdict | Requests | Missing pages |
|---|---|---|
| Authenticated identity | 154 | 0.0% |
| Proven impersonation | 876 | 69.1% |
| Total, reading without verification | 1,129 | 53.7% |
The question to put to your current measurement tool: what does bot identification rest on in this report? If the answer is the user agent, the volumes shown mix genuine bots and impersonators. The method and the four verdicts, in our white paper →
How does Fast Growth Advisors build your AI visibility?
Diagnose
The free diagnostic includes a citability test: five of your category’s questions asked to ChatGPT, Perplexity and Google AI Overviews, answers archived, verification link included, a read of the causes on the message side and the surface side. You know where you stand, and why.
Fix
The message first: who you serve, what you do, what sets you apart, phrased for a hurried reader as much as for a machine. The structure next: extractable pages, questions and answers, quantified proof, markup. It is the core of our Message-Market Fit method.
Guide production
A clear site is not enough: you have to feed the surfaces AI engines consult, regularly. We define what to publish, where and in what format; your teams produce. To industrialise without sacrificing quality, our NOMO IA software is built exactly for that.
Discover NOMO IA → · Our AI & agents approach →
A point of honesty about tools: a GEO dashboard measures your presence, it does not create it. What creates a citation is clear, structured and sourced content, published in the right place. Measurement confirms the work, it does not replace it.
What does the Message-Market Fit Observatory show?
We audit the public surfaces of French post-funding startups with a published methodology. The result directly links clarity and citability:
75.9% of the 369 audited French post-funding startups fall below the critical clarity threshold, set at 37.5 out of 75. They are largely the same ones AI engines never cite.
Source: Message-Market Fit Observatory, Fast Growth Advisors, 2026.
Three further figures from the same report set the scale. The average clarity score is 5.33 out of 10, barely half the potential. No audited startup goes above 8 out of 10, which means the top of the field does not exist yet in France. And readability by AI engines is the weakest of the fifteen sub-criteria measured, at 0.91 out of 5.
That leaves the question every funded executive asks. Does capital fix the problem over time? It does not. The correlation between the amount raised and message clarity is close to nil, R² = 0.036, which means raising 50 million rather than 5 does not make a message any clearer. That is good news for whoever decides to deal with it: the gap closes through work, not through funding.
These figures cover French post-funding startups. A UK edition of the Observatory is in preparation.
This data is freely reusable with attribution. For the full methodology: Explore the Observatory →
FAQ
What is GEO (Generative Engine Optimization)?
GEO covers the practices that make a company visible and correctly cited in AI answers like ChatGPT, Perplexity or Google’s AI Overviews. Where SEO optimises page ranking, GEO optimises extraction: message clarity, structured content, sourced figures and presence on the surfaces AI engines consult (reviews, comparisons, Q&A).
Why doesn’t my company appear in ChatGPT or Perplexity answers?
Three causes dominate. An implicit message: if your pages do not state who you serve, what you do and what sets you apart, an AI can extract nothing. An unusable structure: no questions and answers, no clean definitions, no sourced figures. And an absence from the third-party sources AI engines consult. The diagnostic identifies the cause that affects you first.
Does GEO replace SEO?
No, it adds to it. The foundations are shared: clear, structured and credible content. What changes is consumption: Google ranks pages, answer engines compose an answer and cite their sources. A sharp message serves both. That is why the work starts with the message, not with the technique.
Google does not answer my question. Am I invisible?
No. A question where Google generates no AI Overview measures nothing at all. Google writes an answer on some queries and not on others, and that choice is its own. If a report tells you that you are invisible on those questions, it is confusing an absence of answer with an absence of citation. The question to put to your provider is simple: how do you tell the two apart? We keep them separate in every reading, and we apply no penalty to a question that got no answer.
What exactly is measured in Google AI Overviews?
The block Google writes above the results, and the pages it cites there. Every reading returns four things: the state of the question, the text of the answer, the pages cited with yours flagged if it appears, and your organic ranking on the same query. Added to that is the link to run the search yourself, because a measurement you cannot reproduce is not worth much. The reading is dated: answers vary from one day to the next, and we saw it happen twice in a single day on the same question.
GEO, AEO, LLMO: what are the differences?
They are mostly labels for the same discipline. GEO (Generative Engine Optimization) has taken hold in usage; AEO (Answer Engine Optimization) and LLMO (Large Language Model Optimization) name the same thing. Whatever the acronym, the mechanics are identical: publish clear, structured and verifiable content, on your site and on the surfaces AI engines consult, so an answer engine can extract and cite it.
Do you need a GEO tracking tool?
Not to begin with. A monthly manual protocol is enough for a startup: ask ChatGPT and Perplexity the three or four questions your customers ask you, archive the answers, note who is cited and why. Tracking tools bring scale to agencies and large brand portfolios. But no dashboard ever created a citation: the message, the structure and presence on the right surfaces create it.
How long before you see GEO results?
Two horizons. Engines with live retrieval (Perplexity, AI Overviews, ChatGPT with search) can reflect your fixes within weeks, as soon as your pages are recrawled. Presence in answers drawn from model training and pick-up by third-party surfaces build over several months. That is why the method starts with the fast-impact fixes: message, structure, markup.
The four pages in this series
Each one carries its own dated measurements, with the link to run them again.
Google AI Overviews → The monthly barometer, forty B2B buying queries, what Google writes and whom it cites.
ChatGPT visibility → What the engine sees of your site, and why it only searches three times in ten.
Perplexity visibility → The engine that cites the most, its two crawlers, and what a citation returns.
Measure AI visibility → The method exactly as we apply it to our own site, instrument by instrument.
Where to start?
With the measurement that costs nothing: the free diagnostic, citability test included. You will know whether your prospects and AI engines understand the same thing about you, and what to fix first.
Concretely, the diagnostic asks five questions from your category to ChatGPT, Perplexity and Google AI Overviews, archives the answers, and reads your pages with the three extractors that prepare the corpora of large models. You get two things. What an engine says about you today, verbatim. And what it actually keeps from your site, measured, including the proof that disappears silently.
There is nothing to install and no dashboard to watch. The starting point is a measurement, not a subscription.
For the detailed technical protocol: read our method for making a site citable by AI engines →