The IAB Just Told You Which AI Visibility Numbers to Ignore

Caroline Giegerich spent this year assembling a rulebook nobody had written. On August 3, the IAB published Measuring Visibility in the AI Era. It’s the trade body’s first attempt to say what counts as proof that your brand turns up when a buyer asks ChatGPT or Gemini what to get. Giegerich, the IAB’s vice president of AI, runs its AI Center of Excellence and chairs a committee of more than 400 member companies.

Her framing of the job is modest.

“We are trying to give companies a good idea of what good looks like,” she told MediaPost.

Read the document and the modesty falls away. It’s a disqualification test.

The framework sorts every AI visibility number into two tiers. Directional data is good enough for spotting a trend or briefing your team internally, and explicitly not good enough for moving budget. Decision-grade data has to clear higher bars for sample size, reproducibility, and documented method. Below both sits a category the trade body won’t dignify with a tier. Run fewer than 50 queries and your measurement is “exploratory,” the polite way of saying it can’t characterize your category at all.

That threshold does more work than the rest of the document. More than 20 companies now sell AI visibility tracking, each with its own query set, platform coverage, and scoring rubric. A tool that samples a couple dozen prompts and hands you a score has been the category norm. As of Monday, a published standard calls that number exploratory.

Why the same brand gets different scores

The mechanism is in how these systems answer. Search returns a ranked list that’s roughly stable between two identical queries. AI assistants don’t.

“If you’re a brand or agency and familiar with doing SEO or SEM, you are used to more of a straight line of data. ChatGPT or Gemini are probabilistic models, which means they predict next works, making them variable,” Giegerich said.

Probabilistic means the model picks likely next words rather than looking up a stored answer. Ask twice, get two answers, and neither is a bug. Measuring something that varies takes a bigger sample than measuring something that doesn’t. That’s the reason for the 50-query floor, and it’s why two vendors watching the same brand in the same week report different results without either of them lying.

Operators worked this out first. Andrii Byzov, a fractional CMO for B2B tech companies, put the caveat into his own state-of-play writeup in June.

“These figures come from different studies with different methods and definitions, so they disagree at the edges, and any single number should be read as directional.”

He wrote that six weeks before a standards body reached the same conclusion and gave it a name.

The number the standard doesn’t measure

Here’s the finding that should reorder your priorities. Semrush ran 126 million US AI search prompts between January and April 2026, across ChatGPT, Gemini, Google AI Mode, and AI Overviews. It tracked more than 1,200 brands in 22 verticals. Only 36 of them held a top-100 spot on every platform in every month.

The 36 are YouTube, Google, Reddit, Amazon, Apple, Walmart, LinkedIn, Netflix, Disney, Nintendo, and a couple dozen others of that size. Not one is a company you’d call a peer. Semrush also measured how often the brands an engine names match the sources it cites. That ran from 64% on AI Overviews down to 30% on Gemini, a 34-point spread across four products your buyers use interchangeably.

Consistent visibility across every AI surface is currently a property of being one of the largest platforms on earth. For everyone else the honest goal is narrower: show up for the handful of questions your buyers ask, on the one or two engines they use.

The IAB says only 16% of brands systematically track AI visibility today. Worth noting where that figure comes from: the framework document itself, with no study, sample, or window attached. A playbook arguing that numbers need disclosed methods carries a headline number without one. We made the case in July that this gap wasn’t closing because AI search lacked the two things that fixed viewability: a shared definition and an audit regime. The first has now arrived. The second hasn’t, which is why the document can grade your vendor’s rigor while nobody grades the document’s.

What to do with this on Monday

Send your AI visibility vendor three questions, taken straight from the framework’s disclosure list.

1. How many queries is my score built on, and what are they? Under 50 and the IAB’s own standard calls it exploratory. Ask to see the prompt set, not the count.
2. Which platforms, and measured how often? A score blended across four engines hides the 34-point spread between them. You want per-engine numbers, because visibility and referral traffic are different things and so is every engine.
3. Is this directional or decision-grade, in your words? Any vendor who won’t pick one is telling you the answer.

The framework doesn’t make your AI visibility better. It makes the claims about it checkable, which has to come first. Marketers spent a decade learning to ask what a traffic number was counting. That habit is now due on a second set of dashboards, and the split between search and AI reporting is where the confusion lives.

Ask which tier your number is. If the vendor hasn’t heard of the tiers, you’ve learned something either way.