The marketing industry has spent a year turning AI search into a new reporting category. The charts are multiplying. The confidence isn't.
That gap is the problem with AI visibility measurement right now. A brand can show up in an answer, collect a healthy share of voice score, and still fail to influence a buyer, a click, or a dollar of revenue.
The Interactive Advertising Bureau's new AI Visibility Measurement Framework is a useful attempt to bring order to the mess. The framework gives marketers four useful lenses: presence, prominence, portrayal, and persuasion. It also separates directional measurement from decision-grade measurement.
That is a good start. It isn't a budget system yet.
Mentioned is not the same as preferred
The old search journey was relatively legible. Someone searched, saw a results page, clicked a result, and eventually did something that analytics could connect to a session. It was never perfect, but the trail had footprints.
AI discovery compresses that journey into a response. A model can summarize ten sources, recommend three companies, cite one page, and answer a follow-up without sending the user anywhere. The brand may be present without owning the decision.
That distinction matters because the easiest metric to collect is mention rate. It is also the easiest metric to overvalue.
A mention can be a warning. It might show up in a comparison where the brand loses. It might appear in a hallucinated product description. It might be cited as an old source even though the company has changed its offer. A dashboard that records only presence turns all of those outcomes into the same green square.
Marketers have seen this movie before. The industry spent years treating impressions as evidence of influence, then spent more years trying to prove that attention was real. AI visibility is about to repeat the same mistake with better graphics.
The IAB's four-P structure helps because it gives the conversation more resolution. Presence asks whether the brand appears. Prominence asks where it appears and how often. Portrayal examines the framing, sentiment, and accuracy. Persuasion asks whether the answer actually recommends the brand and creates a next step.
The order is important. Persuasion is the only one that sounds like a business outcome, but it can't be trusted if the first three are noisy.
The four Ps need a fifth question
The framework's missing question is simple: what was the user trying to do?
A brand should not expect the same visibility pattern for a discovery prompt, a shortlist prompt, and a purchase prompt. Presence may be valuable early in the journey. Prominence and portrayal matter more when a buyer is comparing options. Persuasion becomes meaningful when the request contains a decision, a constraint, or a transaction.
Without intent classification, one blended visibility score hides the useful differences.
Imagine a regional healthcare brand that appears in 40 percent of broad educational prompts but only 8 percent of local provider comparisons. Its aggregate score may look healthy. Its growth team should be worried. The brand is visible in the room where people learn, then disappears when they choose.
Now imagine a software company that appears less often but is consistently recommended for one high-value use case, with accurate pricing and a link to the right product page. A lower share of voice could mask stronger commercial value.
This is the same reason I argued in AI search measurement is becoming a measurement crisis that raw citation counts can't stand in for demand. The unit of analysis has to move from “did the model mention us?” to “did the model help the right buyer make progress?”
That shift also changes what teams should sample. A serious measurement program needs prompt sets organized by intent, audience, geography, product, and stage. It should track repeated runs over time because the same prompt can produce different responses. It should preserve the exact answer, not just the score extracted from it.
Otherwise, the dashboard is a weather app that deletes the weather.
Directional data has a job
The IAB's distinction between directional and decision-grade measurement is one of the most practical parts of the proposal. Directional data is still valuable. It can reveal a competitor's sudden rise, expose a recurring factual error, or show that a new product category is being defined without you.
But directional data should be treated like reconnaissance, not a media invoice.
The temptation will be to take an early visibility score and attach a revenue forecast to it. Vendors will make that temptation easy. A neat line chart invites a budget conversation even when the underlying sample is small, the prompt mix is narrow, or the system being measured is changing under the test.
Decision-grade measurement needs more friction. It should disclose prompt volume, query selection, model coverage, testing cadence, reproducibility, and validation rules. It should show confidence intervals or at least make uncertainty visible. If a vendor can't explain how a score changes when prompts are translated, localized, or re-run, the score isn't ready to steer spend.
This is where the AI attribution problem becomes operational. Attribution drifts when the thing being measured changes faster than the reporting model. AI answers make that drift worse because the response is probabilistic, the interface is evolving, and the final conversion may happen days later through a channel that gets credit by default.
The answer isn't to throw away measurement. Google's own people-first content guidance points in the same direction: evaluate whether the work helps someone, not whether it flatters the reporting system. Stop asking one metric to do four jobs.
Use directional reporting for awareness and issue detection. Use decision-grade reporting for changes that affect content investment, agency evaluation, or budget allocation. Keep the two outputs visibly separate so an early signal doesn't dress up as a financial fact.
The brand has to measure the answer
Most AI visibility tools are built around the query. That makes sense for collection, but it leaves out the part a human actually experiences: the answer.
An answer-level review should capture at least five things:
- Whether the brand appeared at all.
- Whether it appeared in a useful position or merely as a footnote.
- Whether the description was accurate, current, and specific.
- Whether the response gave the buyer a reason to prefer it.
- Whether the citation or next step led somewhere that could support the decision.
That last point is uncomfortable because it brings the website back into the picture. A brand can improve its model visibility and still lose the click if the cited page is slow, vague, outdated, or impossible to act on. AI search doesn't remove experience design. It makes the handoff less obvious.
A candid test is to hand the captured answers to someone who wasn't involved in the campaign and ask them to choose. Which company would they call? Which one seems safest? Which one appears to understand the problem? If the answer doesn't change a decision, a higher mention rate is mostly decoration.
Budgets need a harder standard
The first AI visibility budgets should not be built around a promise to “rank in ChatGPT.” That phrase is already too blunt to be useful.
A better brief would name the buyer, the decision, the prompt family, the desired portrayal, and the action that should follow. It would distinguish between an answer that creates trust and one that simply includes a brand name. It would fund correction work when the model gets the facts wrong, not just content production when the score dips.
The practical scorecard might include visibility by intent, recommendation quality, factual accuracy, citation usefulness, assisted sessions, branded demand, and qualified conversions. None of those metrics is perfect. Together, they create a chain that can be challenged.
The chain is the point. It also connects to the problem I covered in why voice search is eating SEO: discovery only matters if the experience can carry the user forward. A metric that survives scrutiny is more valuable than a metric that looks impressive in a quarterly deck.
The IAB framework gives the industry a shared vocabulary. That alone is progress. But standards don't create truth by themselves. Teams still have to decide what counts as meaningful, collect enough evidence to support the claim, and admit when a result is only directional.
That is also why share of model is a vanity metric when it isn't tied to a buyer or a business outcome. AI visibility measurement will become a real budget discipline when the report can answer one unfashionable question: did this visibility change what the right person did? Until then, keep the chart, label it honestly, and don't let it buy the media plan.