Twenty companies will sell you an AI-visibility score. They won't agree on yours.
The IAB published the first shared vocabulary for measuring how brands surface inside AI answers — and graded a whole vendor category on the way out the door.
Same brand, same question, five different needles. The IAB just drew the line where a reading starts to count. Illustration: AI in Marketing Weekly
On Monday the IAB released Measuring Visibility in the AI Era, guidance for tracking how brands and publishers surface inside ChatGPT, Gemini, Copilot, Perplexity and Claude. IAB's VP of AI, Caroline Giegerich, was careful with AdExchanger about what it is: not a standard. Standards need stability, and AI search doesn't have any yet.
Two numbers carry the story. More than 20 companies now sell AI-visibility measurement, each with its own method, each capable of returning a different answer about the same brand. And anything under 50 queries is, by the IAB's own line, “exploratory” — a rung below even “directional,” and nowhere near what you'd carry into a budget meeting. PPC Land, citing McKinsey CMO survey work, puts the share of brands tracking AI visibility systematically at 16%.
That's the gap I'd put in front of a CMO. Most brands aren't measuring this at all, and a good number who are ran thirty prompts and called it a benchmark. The IAB's four P's — presence, prominence, portrayal, persuasion — give you a hierarchy to argue from, and the one to watch is portrayal. Giegerich says the thing that keeps her up at night isn't hallucination, which is improving. It's a model calmly repeating stale or wrong facts about your product, in your category, to a buyer who has no reason to doubt it.
What it means Before you buy another AEO tool, make the vendor put query volume, prompt coverage and reproducibility in writing. If they won't, you're buying a vibe.
20+vendors selling AI-visibility scores
50queries — below this is “exploratory”
16%of brands track it systematically
Ethics & Privacy
The disclosure deadline landed
The EU AI Act's Article 50 transparency rules went live August 2. Chatbots have to tell people they're chatbots. AI-made images, audio and video that look like real people carry a label, even when nobody meant to deceive. California's AI Transparency Act took effect the same day.
What it means Sweep every customer-facing bot and AI-made asset running in the EU or California for a disclosure line. It's a compliance chore that doubles as a trust move.
Retail Media
Amazon's ad business keeps outrunning the market
Amazon posted $19.8 billion in Q2 ad revenue, up 26% from $15.7 billion a year ago. The retail-media network that also owns the shopping agent is growing faster than the ad market around it — and faster than most media plans have been rebalanced.
What it means Pull your Amazon share of digital spend and ask whether it matches Amazon's share of discovery in your category. If the two don't line up, one of them is wrong.
A brand test worth copying · the agent layer · the consumer underneath it all
Brand & Agency
Cuisinart ran the test everybody else is arguing about
Conair's Amazon team beta-tested Amazon Ads' video creative agent with a clean experiment: equal spend, same month, one video made in-house against one made by the agent. The AI video won on click-through and conversion. The interesting part is why. Amazon's engineers say the agent invents nothing — it digests every ad the brand has run, finds what worked, and makes more that's close but not identical. Shoppers go blind to creative they've seen too often; the agent stays fresh enough.
What it means The lesson isn't “use AI creative.” It's that Cuisinart got a defensible answer in a month by running a test instead of a debate. Copy the method.
Marketing Data & Media
The agent layer shipped in a cluster
PulsePoint, Simpli.fi and Rokt all pushed agentic platforms on August 4. Rokt's new Brain V4 is the one to read closely: a rebuilt decisioning engine pointed at the transaction moment, running across a network the company expects to carry more than 10 billion transactions this year. When three vendors ship the same shape of product in one day, that isn't a trend piece. That's a category forming.
What it means Name the one funnel decision you'd hand to an agent first, and write down now how you'd know it worked. Pick the metric before the demo, not after.
What it means The shopper your agent is talking to is far more careful than Amazon's 26% ad growth makes the market look. Price and message for a cautious buyer having a decent month, not a confident one.
The Marketer's Playbook
What To Do Monday Morning
Three moves, in order, with dates attached. None of them require buying anything.
1 · By FridayAudit your AEO vendor against the IAB's disclosure bar. Ask for query volume, prompt coverage and reproducibility in writing. If they dodge, you have your answer.
2 · By month endPut a disclosure line on every customer-facing bot and AI-made asset. The EU and California rules are live, and whoever discloses first sets the trust floor in the category.
3 · Before Q4 planningRun one controlled A/B test of AI creative against your in-house work. Equal spend, same window, one metric. Cuisinart got its answer in 30 days.
16%Share of brands tracking AI visibility systematically. Whatever your AEO budget is, it's being spent into a room where almost nobody is keeping score.
What it means This was a measurement week, not a model week. Every move above is about being able to defend a number you already have.
Watch List & Sources
What I'm tracking into next week
01BLS CPI for July, Wednesday, August 12. Does June's cooldown hold, or was energy a one-off? BLS Consumer Price Index
02Michigan preliminary sentiment, Friday, August 14. Two straight months up. A third makes it a trend instead of a bounce. Surveys of Consumers
03The first AEO vendor to publish IAB-aligned methodology disclosures. Whoever moves first wins the next round of RFPs, and forces everyone else to answer the same questions.
Every claim above traces to a first-hand source. Links go to the original, not to coverage of it.
Strip the acronyms out of this week and what's left is an old problem in new clothes. Twenty companies will hand you a number about your own brand. The numbers disagree. Nobody in the room can settle it by running the query again, because the machine gives a different answer the second time. So a person has to decide which evidence is good enough to spend money against.
That's what the IAB actually shipped. Not a measurement — a way of arguing about measurement. How many queries. Which prompts. Does it reproduce. Who discloses their method. It's the scientific method wearing a marketing badge, and I say that as an engineer who still trusts his gut at the end of the day. The gut works better when you feed it something honest.
Wake Forest's motto is Pro Humanitate, for humanity. Here's how it showed up this week. The most valuable person on your team isn't the one who can pull an AI-visibility score. It's the one who can look at five conflicting scores and say, out loud and in front of other people, which one we're going to believe and why. Machines got very good at generating answers. Deciding which answer to stand behind is still the job.
Human-centered AI in marketing — what actually shipped, and what to do about it Monday. Free, from Wake Forest University's School of Professional Studies.
