The marketing industry finally has a vocabulary for AI visibility. That doesn't mean it has a measurement system marketers can trust.
On August 3, the Interactive Advertising Bureau released a framework called Measuring Visibility in the AI Era. Its timing is right. More than 20 companies now sell tools that claim to track how brands appear in ChatGPT, Gemini, Perplexity, and other AI-powered discovery systems. The problem is that those tools can produce different answers for the same brand because they use different prompts, samples, platforms, and scoring systems.
IAB's answer is the 4 P's: Presence, Prominence, Portrayal, and Persuasion. It's a useful map. It is not proof that the map leads to revenue.
The metric market arrived first
Search marketing had decades to build conventions around rankings, impressions, clicks, and conversions. AI discovery has had a fraction of that time, yet brands are already being asked for dashboards, benchmarks, and quarterly targets.
That pressure creates a familiar commercial pattern. A new channel appears. Vendors create a score. Agencies give the score a polished name. Executives ask whether the score went up. Nobody has agreed on what the score is allowed to mean.
Recent coverage from Marketing Dive's report on the framework makes the same point: the industry needs shared definitions before it can compare vendors. The IAB framework matters because it draws a line between a measurement that spots movement and one that can support a decision. It calls the first category directional. Directional data can help a team notice a pattern or compare early signals. It shouldn't decide where the next million dollars goes.
Decision-grade data has a higher bar. The framework points to query volume, sample size, prompt coverage, testing cadence, reproducibility, and platform coverage. Those requirements sound obvious until you ask a vendor to show the raw methodology behind a brand visibility score.
That is where the friendly dashboard usually gets quiet.
Presence is not demand
The first P is Presence. Does the brand appear in an AI response? Metrics include mention rate, citation rate, share of voice, and visibility momentum.
Presence is the easiest layer to sell because it produces a clean number. Your brand appeared in 38 percent of tracked answers this month. A competitor appeared in 31 percent. The chart moves up and to the right. Everyone feels productive.
But an appearance is not an intention signal. A brand can be mentioned as an example of poor service, an outdated option, or a product that should be avoided. A citation can be technically present while buried in a response nobody reads. A model can include the brand because the prompt asked about it directly, not because the brand earned a recommendation.
This is the same mistake marketers made with raw traffic. They treated being seen as evidence of being wanted.
The difference is that AI answers hide more of the journey. A person may ask an assistant for three vendors, accept the shortlist, and never visit the cited pages. Your analytics won't see the consideration event, while a visibility platform may claim a win because your company appeared in the answer.
Both datasets are incomplete. Pretending otherwise is how a new vanity metric gets promoted into a budget metric.
Prominence changes the story
The second P is Prominence. Where does the brand show up, and how much weight does the answer give it?
This is a better question than simple mention rate. A brand named first in a recommendation is not in the same position as a brand added in a final sentence. A product used as the central example is not equivalent to a product listed beside twelve alternatives.
Prominence also forces marketers to inspect the substance of the citation. Did the system draw a meaningful claim from your content, or did it attach your URL as a decorative footnote? Was the page actually read by the model, or did another source repeat the same claim more loudly?
A useful report should show the prompt, the response, the position of the mention, the cited source, the model, the date, and the sampling method. A single score hides the exact evidence a strategist needs to challenge it.
Portrayal is the brand safety layer
The third P is Portrayal, and this is where AI visibility becomes more than an SEO problem.
A model can mention your company accurately, inaccurately, or with a confidence level that the source material never justified. IAB includes sentiment, framing, hallucination rate, and factual inaccuracy rate in this layer. Those aren't cosmetic metrics. They describe what the market is being told about you when your team isn't in the room.
A brand with high Presence and bad Portrayal has a reputation problem disguised as a reach problem. Publishing more content won't automatically fix it. The system may be relying on stale reviews, third-party summaries, scraped pages, or a single inaccurate description copied across the web.
This is why the old idea of “owning the narrative” is getting harder. You can control your site. You can't control every page an answer engine uses to assemble a response. Your job becomes less about producing a perfect statement and more about making the factual record clear, current, and repeated across credible places.
That connects directly to the measurement failures I wrote about in the AI attribution crisis. If the underlying representation is unstable, the conversion report built on top of it won't become trustworthy just because the chart has more decimals.
Persuasion is where the proof gets expensive
The fourth P is Persuasion. Does AI visibility change behavior?
IAB lists recommendation strength and post-citation click-through rate here, with a forthcoming attribution framework intended to connect visibility to action. This is the layer executives actually care about, which is why it will be the most abused.
Recommendation strength is not a sale. A click after a citation is not necessarily incremental demand. Someone may have already chosen your brand and used the assistant to confirm the decision. Someone else may click because the answer was wrong and they want to investigate.
The cleanest way to think about AI influence is as a set of assisted decision events, not a replacement for attribution. Track the answer exposure when you can. Compare exposed and unexposed cohorts when the platform permits it. Use controlled prompts for directional learning, then look for changes in branded search, qualified leads, conversion rate, and sales quality.
Don't ask one visibility score to carry all of that weight. It can't.
What a serious dashboard should show
The practical response isn't to ignore AI visibility. It is to stop treating every vendor score as a fact.
Start with a measurement brief that makes the hidden choices explicit:
- Which platforms are included, and why those platforms represent your audience
- Which prompts, locations, languages, and user contexts are sampled
- How often the same prompts are repeated, and how model changes are handled
- Whether results are reproducible by someone outside the vendor
- Which claims are directional, and which are strong enough for a decision
Then split the report into separate layers. Presence and Prominence belong in an awareness view. Portrayal belongs with brand, communications, and risk teams. Persuasion belongs beside conversion and revenue data, where it can be challenged by the same standards as every other assisted channel.
This is also where share of model starts to replace search rankings, but the replacement isn't one-for-one. Search rankings were already an imperfect proxy. The same problem shows up in the new AI marketing intelligence teams, where more data can still produce less clarity. AI visibility is a stack of proxies, each with its own sampling risk.
A useful internal rule is simple: no metric gets budget authority until the team can reproduce it, explain its limitations, and connect it to a behavior the business actually values.
The uncomfortable part
The IAB framework is a good development because measurement markets need shared language. Presence, Prominence, Portrayal, and Persuasion give marketers a better way to separate being mentioned from being recommended, and being recommended from creating value.
But standards don't erase uncertainty. They make uncertainty easier to name.
That distinction matters. The next wave of AI visibility vendors will use the framework as a trust signal. Some will earn that trust by showing their samples, methods, and limitations. Others will put the 4 P's on a slide and keep the black box underneath.
The buyer's job is to tell the difference before the score becomes a target.
The brands that win this category won't be the ones with the highest visibility number. They'll be the ones that know which parts of that number are real, which parts are directional, and what evidence would change their mind.