If you have ever been shown two “share of voice in ChatGPT” numbers for the same brand and wondered why they disagree, the answer is that until August 2026 there was no agreed way to measure any of this. Now there is. The Interactive Advertising Bureau (IAB), the trade body that sets most of the measurement standards the advertising industry runs on, has published Measuring Visibility in the AI Era, a 36-page framework that defines what good AI visibility measurement looks like. We have read it end to end. This is the short version.

Who wrote it, and when

The framework was published on 3 August 2026 by the IAB’s AI Visibility Measurement Framework Working Group, led by Caroline Giegerich, the IAB’s Vice President for AI. The working group is a cross-industry mix: measurement scientists from Walmart and WPP Media, product leads from Microsoft Clarity, agency heads from PMG and Tinuiti, the Alliance for Audited Media, EMARKETER and a couple of measurement start-ups. It sits inside Project Eidos, the IAB’s wider programme for how measurement gets defined and compared across the industry. In other words, this is not a vendor white paper. It is the buyers, the sellers and the auditors agreeing on a common language.

Why it was written

The IAB’s own framing is blunt. More than 20 companies now sell AI visibility tools, each with its own methodology, prompt library and scoring, and they produce different answers for the same brand. There was no shared definition of a “mention”, no standard for what counts as a citation, and no way for a buyer to tell whether a tool’s output was reliable enough to act on. Only 16 per cent of brands track AI visibility systematically, largely because nobody could agree on what to track.

The scale of the shift explains the urgency. The IAB cites ChatGPT at over 900 million weekly users, Google AI Overviews at over 2.5 billion monthly users and appearing on almost half of all searches, and McKinsey’s estimate that unprepared brands could lose between a fifth and a half of their traditional search traffic. Publishers are already feeling it, with search referrals down sharply over two years while AI referrals, though growing fast, remain under one per cent of traffic. Everyone is optimising for AI discovery. Almost nobody can measure whether it is working.

What it covers, and what it deliberately leaves out

The framework covers organic, unpaid visibility only: how a brand or a publisher shows up in AI-generated answers without paying for the privilege. It does not tell you how to optimise for AI search, it does not cover paid placements inside AI answers, and it does not attempt attribution, which the IAB has promised in a separate framework. It also does not rate or certify any tool. What it does is give buyers a basis for evaluating the tools they are being sold.

The four P’s: a shared vocabulary

The core of the document is a set of named metrics arranged as a chain of cause and effect. The IAB calls them the four P’s, and the order matters, because if the first one is zero nothing further down the chain can happen.

  • Presence: does the brand appear at all? Measured as Mention Rate (how often the brand is named across a set of questions), Citation Rate (how often it is used as a source, with a link or by name), Share of Voice (its mentions as a share of all brands in its category) and Visibility Momentum (how those figures are moving over time).
  • Prominence: where and how prominently? Measured as Position: first or fifth in a list, a standalone recommendation or one of several, and how far down the answer it sits.
  • Portrayal: in what context, and how accurately? Sentiment and Framing (is the brand described as the leader, the budget option, the outdated one, or filed under the wrong category entirely), plus two accuracy measures the IAB is careful to keep apart. Hallucination Rate counts the things the AI simply made up about you. Factual Inaccuracy Rate counts the things it got wrong because a source it read was wrong. The first is the platform’s problem. The second is fixable, because the source is usually a page you control.
  • Persuasion: does visibility drive action? Recommendation Strength (actively recommended with a reason, or merely listed) and Post-Citation Click-Through Rate, which is the bridge to attribution and depends on data most AI platforms do not yet share.

Publishers get a parallel set built on citations rather than mentions, including how substantively their content is used and how clearly they are credited.

Two grades of data: directional and decision-grade

This is the section most worth remembering. The IAB says not all AI visibility data is fit for the same purpose, and it draws a line between two tiers.

Directional data tells you whether visibility is broadly going up or down and whether you appear for your category at all. It is valuable for early warning, internal briefings and keeping an eye on competitors. It is not good enough for moving budget, choosing a supplier or presenting to a board.

Decision-grade data can support a statement like “our share of voice fell four points this quarter, driven by one competitor in recommendation-type questions”. To earn that label a programme has to clear a set of minimums: enough repeated responses per question to see the normal spread, a large and varied question set covering all four kinds of intent (what is X, X versus Y, best X for Y, where to buy X), weekly measurement, defined limits on how much results may vary between runs, and full documentation of how raw answers were turned into numbers. The IAB even sets a floor: fewer than 50 questions in a programme is “exploratory”, which is a polite way of saying it is not measurement yet.

The IAB’s line is that both tiers are legitimate. The failure is treating directional data as decision-grade without admitting the gap.

What a tool vendor now has to tell you

The framework’s practical teeth are a disclosure list. Anyone selling AI visibility measurement should be able to say which platforms and model versions they test, how their question library was built and whether it is grounded in real search behaviour or invented, whether they collect data by running their own queries, by watching a panel of real users, or from the platforms’ own reporting, how they detect mentions and sentiment, whether they separate hallucinations from genuine inaccuracies, and what they do to their historical data when a model updates. The IAB’s phrase for a vendor that will not answer is that “the absence should itself be treated as a signal”.

Why the numbers move when you have done nothing

The last section deals with something anyone who has tracked AI visibility will recognise: the same question asked twice gives different answers. The IAB’s position is that a single response is not a measurement, that visibility is a range rather than a number, and that a change only means something if it is bigger than the normal wobble. It also explains the two reasons a figure can shift overnight with no change to your brand: the platform retrained its model, or the platform changed the way it lays out answers. Either should trigger a fresh baseline rather than a continuous trend line. The tell-tale sign of a platform change is that every brand in the category moves at once on the same engine.

There is even suggested wording for reporting this to leadership, along the lines of “we report ranges, changes inside the range are not meaningful, changes outside it are”. Anyone who has spent twenty years reading Google Analytics as exact figures will need that reframing.

What we are doing with it

We have adopted the IAB’s vocabulary across our AI visibility work, so that the numbers we report can be compared with anyone else’s. Every report we produce now states which tier it belongs to. Our fortnightly visibility cycles are directional by the IAB’s definition, and we say so, because that is the honest label and it is the right tier for what clients use them for: spotting movement early and knowing which questions to fix. We report ranges rather than single figures, we report each AI platform separately rather than blending them, and we keep hallucinations and factual errors in separate columns, because they need different remedies. And any measurement tool we use or recommend has to answer the IAB’s disclosure list first.

The framework is a starting point and the IAB says as much. Thresholds will tighten, paid placements and image and voice answers are on the list for later, and attribution is coming separately. But for the first time there is a common language, and that changes the conversation from “which tool is right” to “which tier is this, and what is it good for”.

Source: IAB, Measuring Visibility in the AI Era, published August 2026. The market statistics above are the IAB’s own citations.

FAQ

Does this tell me how to get my brand into ChatGPT answers?
No. It is about measuring visibility, not achieving it. Optimisation advice is explicitly out of scope.

Do I need a paid tool to follow it?
Not necessarily. The framework applies equally to an in-house programme. What matters is meeting the tier’s minimums and documenting your method.

Is a “visibility score” out of ten meaningful?
Only if the provider discloses how it is weighted and shows the underlying metrics. The IAB treats composite scores as a presentation layer, not a measurement.

Which AI platforms should be measured?
Those with a meaningful share of consumer AI use in your market, reported separately and weighted by that share. In the UK that currently means Google AI Overviews and AI Mode, ChatGPT, and usually Perplexity and Gemini.