5 minute read

Key takeaways

  • Since 3 August 2026, an AI visibility figure can be judged against a standard.
  • The IAB, the body that standardises online advertising measurement, published “Measuring Visibility in the AI Era”: a shared vocabulary and two quality tiers, “decision-grade” and “directional”.
  • A figure fit for decisions requires published parameters: which surface and model, how many passes, what denominator, which answers are kept. A sentence that says “standardised protocol” is not a protocol.
  • Across the 19 providers in our market register, read on 18 and 19 August 2026: 14 claim to measure, none publishes its full parameters, 7 make a firm result promise, none publishes a complete failure. Every finding carries its exact quote, source and date.

In this article

Until now, a business owner handed “51% visibility in ChatGPT” could do nothing with it: no denominator, no date, no definition. Since 3 August 2026 there is a reference text for sorting such figures. This page says what that text requires, where the French market stands when measured piece by piece, and what it changes for a buyer.

What the IAB published on 3 August 2026

The IAB (Interactive Advertising Bureau) put “Measuring Visibility in the AI Era” online on 3 August 2026, freely accessible. It establishes three things: a common metrics hierarchy for AI visibility across brands and publishers, a shared vocabulary, and above all a distinction between two levels of measurement, in its own words: “Decision-Grade vs. Directional measurement standards to help assess data quality and fitness for use”. In plain terms: a figure fit for decisions, or a mere trend signal.

A decision-grade figure must announce its parameters: platforms covered with models specified, how the question set is built, measurement cadence, reproducibility. A figure that does not announce its parameters stays directional: it may feed an intuition, never a budget.

What this changes when you buy

The standard hands the buyer a set of questions nobody can brush aside any more. Before paying for an AI visibility figure, ask:

  • Which exact surface answered: the consumer application, an API, a third-party tool? And which model, spelled out?
  • How many times was each question asked, and on what dates?
  • What is the denominator behind every percentage?
  • Are the full answers kept and handed over?
  • Does a failed test count as failed, or does it vanish from the figure?
  • Is the figure decision-grade in the IAB sense, or directional? If the provider does not know the distinction, you have your answer.

This grid matches the questions to ask before signing, published with our own answers.

Where the French market stands, measured

Our market register records what each player publishes on its own website, exact quote, URL and capture date included. At the readings of 18 and 19 August 2026, across 19 providers: 14 claim to measure AI visibility, and none publishes the parameters of its measurement, neither questions, nor passes, nor models, nor weightings. One publishes its prices in full. Seven make a firm result promise. None publishes a complete failure; one publishes part of one.

Revision of 24 August 2026: these figures describe the reading of 18 and 19 August. The register has moved since (version 1.5, 21 August); for the current state, the market register is authoritative.

The observation is not ours alone. The largest agency on the market writes, in a December 2025 article, that measurement solutions “rarely state which prompts are used, which models are queried, what data the answers rest on or how results are weighted” (our translation), and calls this an opacity that “complicates the audit of results”. We agree. The IAB standard now says the same thing, with a specification attached.

What the standard requires that we do not have yet

Our measurement charter covers most of the decision-grade tier: questions frozen before the test, surface and model named, three passes per question, raw fractions with denominators, answers kept and handed over, zeros published, corrections published, fees independent of the measured figure.

And here is what we still lack, said plainly. Three passes show variation; they do not estimate a stable frequency, and our pages say so. The concordance of our instrument with the ChatGPT application is established only on the OpenAI surface, for a configuration that has changed since; the check of the current pair is promised for the November 2026 reading. And our readings do not yet publish confidence intervals. A standard is also for that: measuring the gap you still have.

Check us against the standard

Do not take our word for it; that is the whole point of the IAB text. Read the charter, open the complete diagnostic example with its 270 counted answers, and ask us the awkward question. A figure that cannot survive its own parameters did not deserve to be sold.

Frequently asked questions

Does the IAB standard make figures comparable across providers?

Not yet. It provides a vocabulary and two quality tiers; two providers can invoke it with different protocols. What it does make impossible is politely dodging the question “what are your parameters?”.

Can a proprietary tool be decision-grade?

Yes. Proprietary does not mean opaque: an in-house tool can publish its questions, passes, models and denominators. What downgrades a figure is the absence of published parameters, not who owns the tool.

What should I do with a figure that has no denominator?

Treat it as a directional signal, never as a basis for decisions. “51%” says nothing until you know 51% of what, measured when, on which surface, over how many passes.

Sources

  • IAB, “Measuring Visibility in the AI Era”, published 3 August 2026, accessed 19 August 2026: iab.com.
  • L’Agence WAM, “Outils IA : limites, opportunités et méthode pour mesurer votre visibilité”, 16 December 2025, accessed 19 August 2026: agence-wam.fr.
  • HENUSSE, the market register, readings of 18 and 19 August 2026, dated quotes and captures.
  • HENUSSE, the honest measurement charter.