IAB AI Visibility Framework: 10 Questions to Ask Your AI Visibility Tool

ChatGPTPerplexityGeminiClaude

The short answer: in August 2026, the IAB (Interactive Advertising Bureau) published Measuring Visibility in the AI Era, the first industry framework for measuring how brands appear in AI answers from ChatGPT, Gemini, Google AI Overviews and other engines. It sets a shared vocabulary (the "4 P's": Presence, Prominence, Portrayal, Persuasion), separates directional measurement from decision-grade measurement, then lists what a measurement provider must disclose.

At BotRank, this is how we have measured AI visibility from day one: answers collected every day from the interfaces real users see, results split by engine and by market, prompts qualified by search intent, mentions kept separate from citations. This article turns the framework into 10 questions to ask any AI visibility tool, with our own answers.

More than 20 vendors now sell AI visibility tracking. Two of them can measure the same brand in the same category and report two different shares of voice. Marketing leaders have budget for Generative Engine Optimization (GEO), yet no way to tell whether the numbers they buy are reliable. According to the IAB, only 16% of brands track their AI visibility systematically, partly because no common standard existed.

The IAB framework changes that. For the first time, an organization the whole advertising industry recognizes has written down what serious measurement looks like. That is good news for brands. It is also good news for the tools that chose rigor over shortcuts.

What is the IAB AI visibility measurement framework?

The IAB framework Measuring Visibility in the AI Era is an industry reference published in August 2026 that defines how to measure the organic visibility of a brand or publisher in AI-generated answers. It was built by a working group of advertisers (Walmart), agencies (WPP Media, PMG, Tinuiti), measurement bodies (Alliance for Audited Media, EMARKETER) as well as tool vendors. It is part of Project Eidos, the IAB initiative to standardize advertising measurement.

The document does not rate vendors or prescribe tools. It provides four building blocks:

  • A shared vocabulary: precise definitions for each metric (mention rate, citation rate, share of voice, position, sentiment, etc.).
  • Quality standards: a criteria matrix that separates directional from decision-grade measurement.
  • A disclosure framework: what a provider must reveal about its methodology. A refusal to disclose should be read as a signal.
  • Stability practices: how to stay comparable week over week while models change without notice.

The framework covers organic visibility only. Paid visibility inside AI answers and traffic attribution are flagged as next priorities.

The 4 P's of AI visibility: the IAB reading grid

The IAB organizes AI visibility metrics into a causal hierarchy of four levels. If a brand is not present, nothing else matters. If it is present but poorly described, its presence can hurt it.

LevelCore questionIAB metrics (brands)
PresenceDoes the brand appear?Mention rate, citation rate, share of voice, visibility momentum
ProminenceWhere and how prominently?Position in the answer
PortrayalIn what context, with what accuracy?Sentiment, framing, hallucination rate, factual inaccuracy rate
PersuasionDoes visibility drive action?Recommendation strength, post-citation click-through rate

Two distinctions in this vocabulary are worth remembering. The first separates mention from citation: a brand can be named in an answer without its website being used as a source. We covered this in AI brand mentions vs citations: why one score misleads. The second separates hallucination from factual inaccuracy. A hallucination is invented by the model. A factual inaccuracy faithfully repeats a source that is wrong. The fix is different: in the second case, it is the source that needs correcting.

Directional vs decision-grade measurement: the real dividing line

The most important contribution of the framework is the distinction between directional and decision-grade measurement. Directional measurement shows a trend: "are we more visible than last quarter?". Decision-grade measurement supports budget allocation, agency reviews or a board presentation. Both are legitimate. The mistake is treating directional data as decision-grade.

CriterionDirectionalDecision-grade
Sample sizeMultiple responses per queryEnough responses per query to establish a stable distribution
Query volumeAt least 50 queries covering the categoryLarge, diverse query set with subcategories
Prompt type coverageAt least 2 intent types, distribution disclosedAll 4 intent types, results segmentable by intent
Testing cadenceMonthly or quarterlyWeekly or more frequent
Platform coverageOne or more platformsMajority of consumer AI traffic in the market, results per platform

The framework also states a statistical fact many dashboards ignore: single-response measurement is not measurement. AI engines never answer the same question exactly the same way twice. A brand's visibility on a prompt is a distribution, not a single value.

10 questions to ask your AI visibility tool (and our answers)

We turned the IAB criteria matrix and disclosure framework into ten concrete questions. Ask them to your current provider or to the ones you are evaluating. Here are our own answers.

1. Which AI engines are measured? Are results reported separately?

The IAB requires disclosure of the platforms covered, with results reported per platform, because each engine has its own biases. A score that blends every engine hides those gaps.

At BotRank: we measure ChatGPT, Gemini, Google AI Overviews, Google AI Mode, Perplexity, Microsoft Copilot, Claude and Mistral. Every metric is available engine by engine. As we showed in our analysis of why LLM optimization does not transfer across platforms, adding engines together without separating them distorts the diagnosis.

2. How are answers collected?

This is the first line of the IAB disclosure framework. A provider must say whether it simulates queries, observes a user panel or uses platform-native data. It must also state its access method (official APIs, browser-based observation, scraping) and whether web retrieval is enabled. Results collected in different ways should never be combined without disclosure.

At BotRank: we collect every answer by scraping the consumer interfaces your customers actually use: chatgpt.com, gemini.google.com, the Google results page for AI Overviews and AI Mode, perplexity.ai, etc. We do not use model APIs. This is a methodological choice: an API returns raw text, often without web retrieval, without displayed sources or sponsored modules. The IAB defines a brand's position as its place "in the formatted AI response as a user sees it, rather than in the AI's raw text output". Scraping measures exactly that: the rendered answer, with its sources, their order and its sponsored placements.

3. How many answers are collected, how often?

For the IAB, a single response is a sample of one: you need enough responses to describe an engine's normal variability. The framework also classifies monthly or quarterly cadence as directional. Weekly or more frequent measurement is expected for budget decisions.

At BotRank: by default, every prompt is run every day on every tracked engine, seven times more often than the IAB decision-grade threshold. On the main engines, a prompt therefore builds a series of around twenty answers per month. Trends are read on that series, never on an isolated answer. Cadence can be adjusted per engine when a customer prefers a slower rhythm.

4. Do the prompts cover all four search intents?

The IAB names "prompt-driven bias" as one of the two main sources of error. A query set dominated by recommendation prompts favors brands that perform well in that context. The framework therefore asks for coverage of four intent types (informational, comparison, recommendation, transactional) with the distribution disclosed.

At BotRank: every prompt is automatically qualified by search intent, with a confidence level, then remains editable by hand. We use the four IAB intents plus a fifth one for review searches. This makes the bias visible: across all prompts tracked in BotRank, 52% are recommendation prompts, versus 16% comparison, 15% transactional, 14% informational and 4% reviews. Without this qualification, a brand would not know its score is driven by a single type of question.

5. Does the tool separate prompts that already name a brand?

This extends the previous point. A prompt that names the brand ("Is BotRank a good GEO tool?") mechanically inflates its visibility. Mixing those prompts with generic ones means measuring awareness with questions that already contain the answer.

At BotRank: every prompt is also qualified on whether it names a brand. About 15% of the prompts tracked on the platform name an entity. Isolating them shows the visibility a brand actually earns on open questions, where it competes with others.

6. Are mentions and citations measured separately?

The framework defines mention rate (the brand is named) and citation rate (the brand is used as a source) as two distinct metrics. The gap between them is a signal in itself.

At BotRank: named entities and cited sources are analyzed separately for every answer. We know which sites and pages AI engines cite, how their share evolves, then whether the cited page actually talks about the brand. Our study of 1.2M AI responses relies on that separation.

7. Is the brand's position in the answer measured?

Being named first or tenth does not carry the same value. The IAB also asks providers to explain how position is computed in free-text answers.

At BotRank: every brand, product and cited source gets its order of appearance in the answer. Source position follows the order in which the engine displays them.

8. How is the brand described?

The IAB "Portrayal" level covers sentiment and framing: is the brand presented as the market leader, the budget option or an outdated choice?

At BotRank: every answer is split into extracts tied to the brands it names, then analyzed for sentiment, themes and keywords by our Perception module. You see how a brand and its competitors are described, not only whether they are named. Conflicting brand information is often the first cause of poor framing.

9. Are results measured market by market?

The same question asked from France or Germany returns different brands and different sources. The IAB asks providers to disclose the geographic origin of queries and to report results per market rather than as one global score.

At BotRank: every prompt is attached to a market and run in that context. Multi-brand, multi-country groups read their visibility country by country.

10. Is the raw data still accessible behind the score?

The IAB accepts composite scores on one condition: disclose their weighting and keep the underlying metrics accessible. "Composite scores are a presentation layer," the framework says.

At BotRank: the visibility score breaks down by engine, prompt, market and competitor. Raw answers, extracts and sources remain available. The data is also accessible through our API and our MCP server, so it can be joined with the brand's own first-party data.

What the framework leaves open: where we are going

The IAB presents its framework as a starting point. Three topics remain open.

  • Paid visibility. The framework notes that organic answers and sponsored placements now share the same response surface. BotRank already measures sponsored placements in AI answers, separately from organic visibility: our customers see both without ever confusing them. Scraping the real interfaces makes this possible, because APIs do not return those placements. For context: ChatGPT Ads vs GEO.
  • Hallucinations and factual inaccuracies. Measuring what AI engines get wrong about a brand requires a brand-supplied source of truth. This is the next frontier for the whole industry. At BotRank, we are working on it within our Perception module, which already analyzes how every answer describes the brand: the next step is to check those descriptions against reference information supplied by the brand, to tell hallucinations from factual inaccuracies. Our analysis showing that AI engines can cite fake websites explains why the topic is urgent.
  • Stability across model updates. A share of voice can drop five points overnight because a model changed, while the brand did nothing. The IAB recommends separating platform-driven shifts from market-driven shifts. A shift that hits a whole category on one engine almost always signals a model update. Tracking thousands of prompts across many categories makes those movements visible.

We also have a critical reading. The framework remains light on numbers: apart from the 50-query floor, the 7-day window and the weekly cadence, acceptable variability thresholds are left to the working group. And a shared vocabulary does not make numbers comparable across tools as long as prompt sets differ. That is exactly why prompt set transparency matters as much as the metric itself.

How to use the IAB framework today

  1. Decide what the data is for. Trend monitoring can live with directional measurement. Budget allocation requires decision-grade measurement.
  2. Send the 10 questions above to your providers. A vague answer or a refusal is information in itself.
  3. Require results per engine and per market. Be wary of a single score with no breakdown.
  4. Check how your prompts split by intent. If 80% of your prompts are recommendation requests, your visibility is overstated or understated depending on your profile.
  5. Report trends with a margin. The IAB recommends writing "approximately 22%, plus or minus 4 points" rather than "22%". It prevents overreacting to normal variation.

Go further on choosing a tool: the 5 essential features of a GEO tool and the AI visibility signals most brands still miss.

FAQ: the IAB AI visibility measurement framework

What is the IAB "Measuring Visibility in the AI Era" framework?

It is an industry reference published by the IAB in August 2026. It defines the metrics of brand visibility in AI answers, the quality criteria of reliable measurement and the information a provider must disclose.

What are the 4 P's of AI visibility?

Presence (does the brand appear?), Prominence (in what position?), Portrayal (how is it described?) and Persuasion (does visibility drive action?). These four levels form a hierarchy: without presence, the other metrics are meaningless.

What is the difference between directional and decision-grade measurement?

Directional measurement shows a general trend. Decision-grade measurement relies on enough responses, prompts, engines and at least a weekly cadence to justify a budget decision.

What is the difference between mention rate and citation rate?

Mention rate is the share of answers that name the brand. Citation rate is the share of answers that use the brand's content as a source. A brand can be mentioned often yet rarely cited.

Is BotRank certified by the IAB?

No. The IAB has not launched a certification program yet: the framework lays its foundations. BotRank is not part of the working group. We aligned our methodology and our answers with the published criteria so every customer can verify them.

AI visibility measurement aligned with the IAB criteria

Daily collection from the real AI interfaces, results per engine and per market, prompts qualified by intent, mentions and citations kept separate.

Try BotRank free for 7 days

Source: IAB, Measuring Visibility in the AI Era, August 2026. BotRank data as of November 17, 2026, computed across all prompts tracked on the platform. Spotted an error? Write to contact@botrank.ai and we will fix it.

AI Search & GEO expert

After nearly 15 years in digital strategy on the client side (including 10 years at Olympique Lyonnais, where he was notably in charge of SEO).
Florian co-founded BotRank.ai in 2025, the GEO (Generative Engine Optimization) tool used by more than 2,500 companies to manage their visibility in AI-generated search results. He writes regularly about GEO and AI Search.