- IAB (the Interactive Advertising Bureau) published a formal, cross-industry AI-visibility measurement framework in August 2026, part of its Project Eidos initiative, built with a working group including Walmart, EMARKETER, Acxiom, WPP Media, and Microsoft Clarity.
- The problem it names explicitly: more than 20 companies now sell AI visibility tools, with no common definition of a "mention," no standard for what counts as a "citation," and no shared way to tell whether a tool's output is reliable enough to act on.
- Its first real contribution is a shared vocabulary: the "4 P's" (Presence, Prominence, Portrayal, Persuasion), applied separately to brand metrics and publisher metrics.
- Its second is a two-tier data-quality standard (Directional vs. Decision-Grade) with a concrete criteria matrix, including a hard floor: fewer than 50 queries in a measurement program is exploratory, not even directional.
- Its third is a required-disclosure checklist, with an explicit rule that a provider's refusal to disclose should itself be treated as a signal, and IAB states this disclosure framework is the foundation for a future provider certification program.
What vendors won't tell you about their own AI-visibility methodology compared three real, funded vendors on how much they disclose about their own scoring. This piece covers something different in kind: an actual cross-industry standards body (not a single vendor, not a single buyer) publishing a formal framework for the entire category, with named companies from brands, agencies, and measurement providers all signing onto the same vocabulary. Here's what it actually says, in full, and where it lines up with (or quietly differs from) definitions already used on this site.
Why is IAB publishing this now?
The scale numbers IAB opens with are the reason this can't wait for the market to sort itself out. ChatGPT has more than 900 million weekly active users. Google AI Overviews have over 2.5 billion monthly users and now appear on almost half of all searches. McKinsey projects unprepared brands could see traffic declines of 20% to 50% from traditional search channels as AI-powered discovery reshapes how people find things. Publishers are already living with the consequences: per Chartbeat data reported by Axios in March 2026, search referral traffic has declined 60% for small publishers, 47% for medium publishers, and 22% for large publishers over the past two years. AI chatbot referrals, despite growing more than 200% in a year, still account for less than 1% of total publisher page views. And per IAB's own citation, only 16% of brands systematically track AI visibility today, in part because there's no shared standard for what "tracking" it even means.
IAB's own framing of the resulting gap is blunt: more than 20 companies now sell AI visibility tools, each using different methodologies and prompt libraries, producing different answers for the same brand or publisher. Buyers have budgets ready to spend and no basis for evaluating what they're buying.
Rather than rating measurement providers, prescribing tools, or building measurement systems itself, the framework defines the shared vocabulary, quality criteria, and disclosure requirements the market needs to function at all.
What are the 4 P's?
IAB organizes AI visibility metrics into a causal hierarchy: each layer only matters once the one before it is established.
| Layer | Core question | Brand metrics | Publisher metrics |
|---|---|---|---|
| Presence | Does the brand appear, or is the publisher cited? | Mention Rate, Citation Rate, Share of Voice, Visibility Momentum | Citation Rate |
| Prominence | Where and how prominently? | Position | Content Utilization Rate |
| Portrayal | In what context, and with what accuracy? | Sentiment, Framing, Hallucination Rate, Factual Inaccuracy Rate | Attribution Clarity, Hallucination Rate, Factual Inaccuracy Rate |
| Persuasion | Does visibility drive action? | Recommendation Strength, Post-Citation CTR | Post-Citation CTR |
A few of these definitions are worth stating precisely, because the framework is explicit that the gap between two similar-sounding metrics is itself meaningful: Mention Rate is the frequency a brand appears in responses at all; Citation Rate is the narrower frequency it's cited as a source. A brand can have a high Mention Rate and a low Citation Rate, and IAB calls that gap itself "a meaningful signal": a model may talk about a brand constantly from training-data memory while almost never treating its own content as authoritative enough to cite. Hallucination Rate and Factual Inaccuracy Rate are also kept separate on purpose: a hallucinated mention is one the AI fabricates with no basis in any real source, while a factual inaccuracy is a citation that correctly points to a real source that itself contains the error. Different failure, different fix: one is an AI platform quality issue, the other traces back to inaccurate underlying content.
Persuasion metrics are flagged as a deliberate bridge: IAB defines them as part of the visibility vocabulary here, but full attribution methodology (actually connecting a citation to a business outcome) is being left to a forthcoming IAB attribution framework.
What's the difference between directional and decision-grade?
IAB's second contribution is a formal quality-tier distinction most tools quietly ignore. Directional measurement identifies patterns and trends (useful for early signal detection, internal briefings, and competitive awareness) but isn't sufficient for budget allocation or strategic decisions. Decision-grade measurement meets a higher bar across sample size, query volume, prompt-type coverage, testing cadence, reproducibility, data validation, methodology documentation, and platform coverage, and can support things like budget reallocation or agency performance reviews. Both tiers are legitimate. The framework's stated failure mode is treating directional data as decision-grade without recognizing the gap.
The criteria matrix backing this distinction gets specific. A few of the thresholds:
- Query volume: the framework treats fewer than 50 queries in an entire measurement program as exploratory, not even directional: decision-grade requires a large, diverse query set with disclosed category and subcategory coverage.
- Prompt type coverage: directional requires at least two of the four intent types (informational, comparison, recommendation, transactional) with the distribution disclosed; decision-grade requires all four, segmented and disclosed.
- Testing cadence: monthly or quarterly is suitable for directional trend monitoring; decision-grade calls for weekly or more frequent testing.
- Reproducibility: directional measurement will vary run to run, and the provider must document typical variation; decision-grade requires the provider to define acceptable variation ranges within a 7-day window and report confidence levels.
This is the same underlying problem AI visibility score covers from the academic side: the University of St. Gallen's 45-day study found cited sources turning over roughly 65% day to day, and identical prompts run simultaneously overlapping only 32–43%. IAB's framework doesn't cite that specific paper by name in the text extracted here, but it converges on the same conclusion from the standards side: a single-response measurement is a sample of one, not a value, and treating it otherwise is the core quality failure the whole directional/decision-grade split exists to catch.
What must providers disclose?
IAB's disclosure framework groups ten required items into three questions a buyer should be able to answer before trusting any provider's output:
What is being measured: Platform Coverage (which AI platforms, with model-version specificity), Prompt Library Construction (size, structure, refresh cadence, category weighting), Query Sourcing (synthetic vs. real consumer search data vs. observed behavior), and Source Type Classification (which source types are tracked and how they're weighted).
How it is being measured: Data Collection Architecture (active query simulation, passive panel observation, platform-native data, or a hybrid, providers must not present outputs from different architectures as equivalent without disclosure), Data Collection Methodology (access method, ToS compliance, live-retrieval configuration), Panel Validity (for passive approaches: panel size, recruitment, demographic representativeness), Data Validation (what external signals the data is checked against, how comparability is maintained across sources), and Mention Detection/Attribution/Sentiment Logic plus Factual Accuracy Classification (how hallucinated, accurate, and factually-incorrect mentions are told apart and reported, separately, not blended into one count).
How the data is maintained: Historical Data and Versioning (how far back data goes, how model updates are handled in the historical record, whether trend data gets re-baselined after platform changes).
The rule that ties it together, stated directly in the framework: "Where a provider cannot or will not disclose against a required item, that absence should itself be treated as a signal." That's a formal, cross-industry version of the same principle behind the four-question checklist this site has used to evaluate vendors in prior pieces (how many prompts, how many runs, over what window, blended or per-platform) except IAB's list runs to ten items with far more granularity on data architecture and validation specifically.
IAB also splits this into minimum disclosure (enough for a buyer to place output in context, appropriate for directional use) and enhanced disclosure (full documentation depth, expected for any provider positioning its output as decision-grade). And notably, IAB states this disclosure framework is explicitly the foundation for a future provider certification program, not yet built, but the direction the standards body says the market is heading.
Why isn't "your score is down" always true?
The framework's third section addresses something most vendor dashboards paper over: AI platforms are non-deterministic, and normal model behavior can look exactly like a real change in a brand's position if a program isn't built to tell the two apart. IAB's own illustration is concrete: a brand's Share of Voice can change five points overnight because a model was retrained, with zero change to the brand itself or its competitive position. Most dashboards would still report that as a win or a loss.
The framework's practical distinction is a simple diagnostic: a platform-driven shift appears across multiple brands or the entire category on one platform, tends to follow a model update or retrieval-configuration change, and shows up as a sudden, abrupt move outside normal variability ranges. A market-driven shift is isolated to specific brands or query types, follows an actual business event (a campaign launch, a new competitor, a piece of content gaining traction) and tends to be gradual or clearly event-linked. Mistaking one for the other means reporting platform noise to leadership as if it were brand performance, or missing a real competitive move because it got dismissed as "probably just a model update."
This is the same variability AI visibility score already documents from the University of St. Gallen study and the SparkToro/Gumshoe.ai consistency research (under a 1-in-100 chance two runs of an identical prompt return the same brand list). IAB's contribution is turning that known instability into an operational protocol: document a variability baseline before acting on period-over-period change, and when an abrupt shift appears, check whether it's isolated to one brand (market-driven) or patterned across the whole category on one platform (platform-driven) before reporting a cause.
Where does this line up with, and differ from, definitions already used here?
IAB's Persuasion category (Recommendation Strength, Post-Citation CTR) maps closely onto the "recommendation" outcome already defined in mention vs. citation vs. recommendation: both treat "the model actively recommends you over competitors" as a distinct, more competitively-loaded outcome than simply being mentioned or cited.
One real difference is worth flagging rather than smoothing over: IAB's Citation Rate counts a citation as either a linked reference or a named reference without a hyperlink. The definition used elsewhere on this site draws a narrower line: a citation specifically traces back to a retrieval step, with an actual audit trail (server logs, an API's sources field). A brand named as a source with no link, under IAB's definition, is still a citation; under the narrower definition, it reads closer to a sourced mention. Neither definition is wrong: they're drawing the boundary for different purposes (IAB for cross-provider comparability, this site for what's actually auditable from the outside), but it's exactly the kind of quiet definitional gap the rest of this framework exists to close, and worth knowing before assuming "citation" means the identical thing in every report you read.
What should you actually do with this today?
The framework isn't a certification program yet, and IAB is explicit that it doesn't rate any specific vendor. It's a checklist buyers, agencies, and publishers can start applying in procurement immediately, ahead of any formal certification. Before trusting the next AI-visibility number handed to you (a vendor's or your own), the practical version of IAB's framework is: ask which platforms and model versions were covered, how the prompt library was built and sourced, whether the data comes from live queries, a behavior panel, or platform-native reporting, how mentions get classified as accurate, hallucinated, or factually wrong, and whether the answer is directional or actually decision-grade. If a provider can't answer those, per IAB's own framework, that's the data quality, not a gap to give them the benefit of the doubt on.
Frequently asked questions
What is IAB's "Measuring Visibility in the AI Era" framework?
A formal, cross-industry measurement standard published by the Interactive Advertising Bureau in August 2026, part of its Project Eidos initiative. Built with a working group that includes Walmart, EMARKETER, Acxiom, WPP Media, and Microsoft Clarity, it defines a shared vocabulary, a two-tier data-quality standard, and a required-disclosure checklist for the AI-visibility measurement industry: it does not rate, certify, or evaluate any specific vendor.
What are the "4 P's" of AI visibility?
Presence (does the brand appear or get cited?), Prominence (where and how prominently?), Portrayal (in what context, and how accurately?), and Persuasion (does visibility actually drive action?), a causal hierarchy IAB applies to both brand metrics (Mention Rate, Citation Rate, Share of Voice, Position, Sentiment, Recommendation Strength, and more) and publisher metrics (Citation Rate, Content Utilization Rate, Attribution Clarity, Post-Citation CTR).
What's the difference between "directional" and "decision-grade" AI visibility data?
Directional data is useful for spotting trends and early signals but isn't rigorous enough to justify a budget decision. Decision-grade data meets a higher standard across sample size, query volume, testing cadence, and reproducibility, and can support things like agency performance reviews or executive strategy. IAB's own floor: fewer than 50 queries across an entire measurement program counts as merely exploratory, not even directional.
What must an AI visibility vendor disclose, under this framework?
Ten categories, grouped under what's being measured (platform coverage, prompt library construction, query sourcing, source type classification), how it's measured (data collection architecture and methodology, panel validity, data validation, mention/attribution/sentiment logic, factual accuracy classification), and how the data is maintained (historical data and versioning). IAB's own rule: if a provider can't or won't disclose against a required item, treat that refusal as the answer, not a gap to chase down.
Does this match how "mention" and "citation" are defined elsewhere on this site?
Mostly, with one real difference worth flagging. IAB's Citation Rate counts both a linked reference and a named reference without a hyperlink as a citation. Mention vs. citation vs. recommendation draws a narrower line, tying citation specifically to a retrieval step with an actual audit trail. Both definitions are internally consistent: they just draw the boundary in a slightly different place, which is worth knowing before assuming the term means exactly the same thing across sources.
Sources referenced
- IAB: Measuring Visibility in the AI Era (Aug 2026, full framework PDF linked from this page)
This piece extends what vendors won't tell you about their own AI-visibility methodology from a single-vendor comparison to a formal cross-industry standard. For the underlying volatility research this framework's quality tiers are built to guard against, see AI visibility score; for the precise definitions this piece cross-checks against IAB's own, see mention vs. citation vs. recommendation.
