More than 20 tools now sell AI visibility numbers, and per the IAB only around 16% of brands track it systematically. The IAB's new measurement standard sets out the questions to ask before trusting any provider's data. Here is a practitioner's read of it, plus the buyer checklist to use before you sign anything.
The Questions to Ask Before You Trust Any AI Visibility Number: Reading the New IAB Measurement Standard
By Stephen Honight, Founder of Lmo7
More than 20 companies now sell AI visibility measurement tools. Each uses a different methodology, there is no shared definition of a “mention”, and the same brand can get a different answer depending on which one you ask, per the IAB. Only around 16% of brands track AI visibility systematically at all, also per the IAB, which leaves most of the market either without a number or trusting one they cannot explain.
That gap is what the IAB (the Interactive Advertising Bureau, the US ad industry’s standards body) has just tried to close. In August 2026, as part of its Project Eidos measurement initiative, the IAB published “Measuring Visibility in the AI Era”, a shared framework for judging AI visibility data rather than a tool that produces it. It does not rate providers or hand out a badge. It hands buyers a set of questions, and this piece is my read of what those questions are and what to do with the answers.
The short version
The IAB’s framework does two things. It gives the market a shared vocabulary for what “visibility” means, four categories, which I will come to, and it draws a line between measurement that is good enough to notice a trend and measurement that is good enough to move a budget. The report calls the first kind directional and the second decision-grade, and its own line on the difference is blunt: “The failure is in treating directional data as decision-grade without recognizing the gap.” My recommendation would be to read this piece less as an AI story and more as a procurement one. Before you buy or renew any AI visibility contract, ask where on that scale the data actually sits, because the sales deck will not volunteer it. The buyer checklist further down is the working part of this piece. The rest explains why it exists.
Why a standard was needed
Ask an AI model the same question twice and you can get two different answers, even with nothing about the brand, the market or the model itself changing. That is how these systems work, rather than a fault waiting to be fixed. The IAB puts it plainly: “single-response measurement is not measurement.” Your brand’s visibility on a given query is a distribution, and a single screenshot tells you where one draw from that distribution landed, not where you stand.
The practical consequence is that a bare percentage overstates what anyone actually knows. The IAB’s own example is worth borrowing: “approximately 22%, plus or minus 4 points” is an honest number. A flat “22%” implies a precision the method cannot support.
We saw this ourselves before we had read a word of the IAB’s report. Our Lexym Rufus Visibility Tracker ran 70,764 queries across 1,082 distinct prompts over 392 nightly runs on amazon.co.uk, amazon.de and amazon.es, between 25 July and 23 August 2026. We published the findings on this blog on 28 August (“We Ran 70,764 Queries Through Rufus in a Month”), and the clearest single result was that the brand leading a Rufus answer on any given night was often not the brand that led it the night before. AI shelf space gets redrawn nightly. That is exactly the kind of non-determinism the IAB is now asking the market to standardise around, and it is why a one-off check of your AI visibility tells you almost nothing.
The market context
None of this is a niche concern. The IAB’s report cites ChatGPT at more than 900 million weekly active users and Google AI Overviews at over 2.5 billion monthly users, now appearing on almost half of all searches and around 14% of shopping queries. McKinsey, also cited in the report, projects that brands who are not prepared could see traffic declines of 20% to 50% from traditional search channels as AI-powered discovery reshapes how people find products.
The publisher side already shows what that looks like in practice. Chartbeat data reported by Axios in March 2026 found publisher search referral traffic down around 60% for small publishers, 47% for medium publishers and 22% for large publishers over the past two years. AI-powered discovery is already showing up in traffic numbers, and the businesses measuring it properly are still a minority, on the IAB’s own figure above.
The 4 P’s: a framework you can map onto what you already report
The IAB organises AI visibility into four categories it calls the 4 P’s. The naming is ad-industry convention, so ignore the label and read them as four plain questions about your brand, in the order they matter commercially.
Presence asks the basic question: does the brand appear? This covers mention rate, citation rate and share of voice. Mention rate is the most fundamental metric in the whole framework, because nothing downstream matters if a model never says your name. Share of voice takes that further by measuring your mentions as a share of all mentions in your competitive category, so a mention rate that looks fine in isolation can still mean you are being drowned out by the rest of the category.
Prominence asks where and how visibly you appear once you are mentioned. In practice this is position: are you named first, or fourth, in the answer.
Portrayal asks what happens to the brand once it is on screen. This covers sentiment and framing, plus two metrics worth separating properly: hallucination rate and factual inaccuracy rate. A hallucination is the model inventing an association with no source behind it. A factual inaccuracy is the model accurately reflecting a source that was itself wrong. Those are different problems with different fixes, and reporting them as one blended number hides which one you are dealing with.
Persuasion asks whether visibility actually drives action. Recommendation strength is the key distinction here: a model that says a brand is “the best option for X because Y” is doing something commercially different from one that lists it as “a good option” among several. Post-citation click-through rate sits alongside it.
Presence and Prominence map onto what share of voice and ranking already try to measure, just organised more precisely. Portrayal and Persuasion are the two categories worth paying closer attention to, because they are where a mention can look good on paper and still not be doing any commercial work.
Directional versus decision-grade
This is the distinction I would spend the most time on if I only had one takeaway to give a client.
Directional measurement is useful. It finds patterns, flags early shifts, and gives a team something to talk about in a monthly briefing. What it is not, per the IAB, is sufficient grounds for reallocating budget or making a strategic call.
Decision-grade measurement is a higher bar on several fronts at once. It needs a large and diverse enough query set that the IAB treats anything under 50 queries per programme as exploratory, not even directional. It needs coverage across all four intent types, informational, comparison, recommendation and transactional, reported separately rather than blended. It needs testing on a weekly cadence or better. It needs reproducibility within a defined seven-day window, with stated confidence levels rather than a single figure presented as fact. And it needs documented methodology and results broken out per platform, because a blended cross-platform score hides exactly the differences you would need to act on.
Put those together and the IAB’s warning from earlier becomes concrete. A monthly dashboard that shows a line going up is directional. Deciding to move agency spend, or drop a channel, off the back of that same dashboard is a decision-grade action taken on directional evidence, and that mismatch is what the standard is trying to stop.
The buyer checklist
The most useful part of the report, for anyone actually paying for AI visibility data, is the disclosure framework. It groups what a provider should tell you into three buckets.
What is measured: which platforms are covered, how the prompt library was built, and where the queries came from. Query sourcing sounds like a technical detail until you see what it does to the output. A provider with excellent reproducibility running the wrong 200 prompts, built without your input, “may produce precise answers to the wrong questions”, in the IAB’s phrase.
How it is measured: which of the four data-collection architectures the provider uses, active query simulation, passive panel observation, platform-native data, or a hybrid of these, plus how they validate their data and the logic they apply to decide what counts as a mention or a positive sentiment.
How the data is maintained: what history exists, and how the provider handles versioning when a model update changes the baseline overnight.
The IAB’s sharpest line in this section is also the most useful one to put to a sales team directly: “where a provider cannot or will not disclose against a required item, that absence should itself be treated as a signal.”
Two examples from our own work show why this matters in practice, not just in theory.
The first is the Publicis Commerce “Decoding Rufus” study, published in March 2026, which we wrote about on this blog. Its headline chart showed content-optimised products gaining an 11 percentage-point advantage over a control group. Its statistical appendix, buried further in, told a different story: on the study’s own measure of how often the AI agent recommends a product (it calls this SOAR), the test group’s mean moved by 0.001, near zero, while the control group fell by 0.072. Most of the reported gap came from the control group getting worse, not the test group getting better. Neither number is wrong on its own, but without a stated baseline for how much this metric moves with no intervention at all, a reader has no way to tell a real effect from ordinary variation. That is the variability baseline the IAB is asking every provider to establish before anyone reports a period-over-period change.
The second is closer to home. A September piece on this blog looked at Trakkr’s own read of its Google Search Console data and found that a large share of the long, prompt-shaped queries showing up in Search Console were not typed by buyers at all. They were monitoring prompts fired by other vendors’ AI visibility tools, some carrying leftover config strings like raw country codes, landing in Search Console because they run through Google’s own surfaces. If that is happening upstream of the visibility number itself, in the raw input a tool is built on, then the input needs interrogating before the output is trusted, not after.
Where Lmo7 already sits
None of the above is an IAB endorsement of Lmo7, and I would not present it as one. It is a bar we set for ourselves before this standard existed, and the standard now gives us a shared language to describe it against.
We report trends and ranges rather than single point-in-time ranks. We report per platform rather than blending everything into one score, because a brand can be strong on ChatGPT and invisible on Gemini and a blended number would never tell you that. We publish our methodology rather than asking a client to trust a closed system. And every figure we hand a client states its window and its scope, because a number with no dates attached is not a number you can act on.
Two live examples, both current clients, both stated with their windows. With Trip Drinks, a 60-day AI search engagement moved average position across the major models from 7th to 3rd, with AI referral traffic up 33% over that period. In the IAB’s terms, that is a Prominence result with a Persuasion result sitting right behind it. With Haleon, our Share of Model analysis on Voltarol found a 100% mention rate and the number one average position in its category on both ChatGPT and Gemini, a Presence result and a Prominence result respectively.
What a standard cannot fix
A measurement standard tells you where you stand. It does not do the work of moving you.
The work behind AI visibility runs on two tracks that move at different speeds. The quick-win track is content clarity, PDP structure, entity schema and answering the actual questions shoppers ask, which you can move in weeks. The slower track is domain authority, third-party mentions, reviews, the citations models actually pull from when they answer a comparison question. That track moves in months, not weeks, and no amount of measurement rigour speeds it up.
My sense is the risk with a report like this is that a reader takes away either that the numbers are too unreliable to bother tracking, or that content work alone is enough without measurement. Neither holds up. Measurement tells you whether the work on both tracks is landing. It does not replace either one.
What I would do next
If you have no baseline at all, start there. Our DaaS tier gives you direct data access across Amazon, Meta, Google and Shopify at £250 plus VAT a month per marketplace, and it is built to the decision-grade bar this piece describes: repeated measurement, not a single snapshot. Commercially, this is the cheapest way to find out whether you have a visibility problem and where the fixing budget should go first.
If you already know you have a gap and want it closed, that is Challenger. The Agentic Tracking & Content Optimisation module runs the track-diagnose-fix loop for you at £1,000 a month, or £2,500 a month for the Full Agentic Stack if you want both tracks, quick wins and the longer authority build, running together. This is the option for a team that wants the work done, not just reported on.
If you run multiple brands or need several stakeholders aligned before anyone signs off a budget, that is Enterprise, from £5,000. This tier standardises measurement across brands and runs the workshop that gets everyone looking at the same numbers before the spending conversation starts, which is usually where these projects stall.
My recommendation would be to pick whichever of those matches where you actually are today, not where you would like to be, and to put the IAB’s disclosure questions to your current provider before your next renewal, regardless of which tier you choose.
Stephen Honight is the founder of Lmo7, an AI-native agency helping consumer brands win in AI-powered discovery and agentic commerce across Amazon and D2C. Lmo7 works with brands including Trip Drinks, Veloforte, Brown-Forman, Haleon, Pelotan and Symprove.