Direct answer
Optimizing your visibility within AIs and measuring it are two distinct disciplines. Optimization — GEO — acts on your content so that ChatGPT, Gemini, or Perplexity cite you better. Measurement observes, quantifies, and dates what AIs actually say about you, without changing anything. The same split as between an SEO agency and Google Search Console: one acts, the other measures. Both are useful. But the measurer should never be the one selling you the optimization — otherwise they're judge and party.
The problem
You're being sold dozens of tools to "boost your visibility in AI." They all promise the same thing: measuring where you stand, and telling you what to do, and selling you the service to do it. The same player makes the diagnosis and pockets the treatment.
Ask yourself a simple question. If the mechanic inspecting your car is paid by the number of repairs they bill you, do you still trust their diagnosis?
That's exactly the situation in the AI visibility market today. The one who measures sells the optimization. And no one finds it troubling, because the sector is young and the confusion suits everyone — except you.
The idea to grasp
In any market that has matured, two disciplines eventually separate: the one that acts and the one that measures. It's not theoretical elegance, it's a condition of trust.
- In web search: SEO agencies optimize. Search Console and Analytics measure.
- In advertising: ad sales houses sell the space. Nielsen counts the audience.
- In finance: the finance department keeps the books. The auditor certifies them.
In none of these cases does the measurer sell the action. Search Console doesn't bill you for SEO services. Nielsen doesn't place your ads. The auditor doesn't keep the books they sign off on. And that's precisely what makes their measurement credible: they have nothing to sell behind the number.
AI visibility has its two disciplines too — but they've been glued together:
- Acting is GEO (Generative Engine Optimization, sometimes called AEO or AIO). You work on content, presence, and structure to raise your chances of being cited.
- Measuring is observing what AIs really answer, query by query, model by model, in numbers.
These are two different skill sets. Conflating them means asking the mechanic to grade their own repair.
What you hear everywhere
"Our tool measures your visibility AND recommends the actions." Convenient. Too convenient. If the same hand holds the thermometer and the invoice, the thermometer always ends up showing whatever justifies the invoice.
"We give you an AI visibility score." Fine. Computed how? Over how many measurements? With what margin of error? Most of the time, it's a single query, on a given day, served up as a tidy round number. But an AI never answers the same way twice. A score with no repetition and no uncertainty margin isn't a measurement — it's a photo taken in the dark.
"AI visibility is tracked like a Google ranking." No. There's no fixed position in an AI answer, no ordered SERP. Claiming otherwise is selling reassurance, not measurement.
And that's where my core stance comes in, the same one that drives all my work: no trust in the AI, nor in the vendor, but only in the facts. Not the spotless score. Not the self-declared expertise. The verifiable, dated facts you can produce if a dispute arises.
My take: measurement must be a neutral instrument
From here on, the register shifts deliberately: we leave the diagnosis behind to describe what the instrument does. Plain, factual — that's the very nature of a measurement.
A measurement only has value if it's independent of the action it evaluates, and rigorous enough to withstand challenge. Concretely, measuring what AIs say requires four conditions:
- Repeating each query several times and showing an uncertainty margin — because a model varies from one time to the next.
- Computing the scores mechanically, without an AI "judging" the result.
- Dating and sealing the measurement, so it's verifiable and can be invoked as evidence later.
- Distinguishing what the AI knows from its own knowledge from what it pulls off the web at a given moment.