IAB Developing Framework to Measure Ads Served to AI Agents

Full analysis

The IAB is building a framework, due November 12, to measure ads that get served to AI agents instead of humans, and to figure out who gets credit when a bot browses, evaluates, and buys on someone's behalf. The real question for operators: who ends up owning the trust layer inside the AI black box, and does the independent measurement stack survive it?

What's actually being decided: not "will the IAB ship a framework" (it will), but whether measurement of AI-mediated buying gets standardized in the open or coded inside each AI platform the way Facebook coded its own. Type 1, hard to reverse. Once OpenAI ships its own attribution integration and vendors rubber-stamp it, that schema is the standard for a decade. Forcing function: the November 12 deadline is soft. The hard forcing function is whenever AI-driven commerce becomes a material share of conversions, and that clock is not the IAB's to set.

The Market Analyst. Michael Bishop, co-founder of OpenAds, laid out the whole plot: Facebook coded its own measurement integrations and vendors just audited the homework. That's the base case here, and it's a slow grind down for anyone whose value depends on owning the tag. DoubleVerify and IAS both survived the walled-garden era, but as auditors, not sources of truth, and their influence compressed. Expect the same. In plain terms for the non-specialist: the company that browses on your behalf also grades whether the ad worked, and marks its own paper. The overlooked winner is the identity and clean-room layer. Whoever becomes the handoff between publisher content signals and AI attribution APIs owns the moat. Watch for an AI platform buying an attribution startup. That's the tell.

The Skeptic. How much ad spend actually flows through AI agents today? A rounding error. OpenAI's business is subscriptions, not ads, so its incentive to open a measurement API to third parties is near zero. You open your books when advertiser confidence funds your revenue. It doesn't, yet. IAB frameworks for nascent behavior arrive too early or too late, and November 12 is too early. Plain version: the industry is building a ruler for a room almost nobody is standing in. And the publisher claim that their content "informs AI responses" is technically and legally unverifiable right now. The measurement problem is real. It won't bite operationally until AI-driven commerce is a real share of conversions, which is a 2027 story at the earliest.

The Operator. Here's what breaks Tuesday morning, well before November. UTM tags and referral strings don't survive the AI-mediated journey. That means last-click attribution, still the default at most mid-market agencies, quietly starts lying. In plain terms: the tracking breadcrumbs an ad leaves behind get eaten by the AI in the middle, so your reports credit the wrong thing. Analysts ship monthly pacing decks showing ROI that won't reconcile with actual revenue, and nobody knows why until the Q3 audit. Publisher ad ops hit the mirror image: impression data they can't tie to any outcome, so they can't defend CPM floors to buyers demanding proof. The framework lands in November. Frameworks don't install themselves. Budget Q4 to audit which of your integrations are already AI-blind.

The Customer / End User (the brand advertiser). From the buy side, this is the awareness-versus-purchase split Caroline Giegerich's working group is drafting: did the AI just surface my content, or did it actually drive the buy? Brands need both answers to set budgets. The danger is obvious. If the platform serves the ad, mediates the journey, and reports the result, the brand is trusting a scorecard written by the seller. We lived this with Facebook and paid for it in years of unaudited "success." Plain version: you're being asked to buy on numbers the store owner calculated and won't let you check. The brands with leverage will demand log-level data or clean-room verification as a condition of spend. Most won't have that leverage, and will take the black box's word.

Where the council splits.

First tension: timing. The Skeptic says the room is empty and the framework is premature. The Operator says the plumbing is already breaking on live campaigns through broken UTMs, regardless of how much agent-commerce exists. Both can be true. The reporting breaks before the spend arrives, which is the worst order.

Second tension: who wins. The Market Analyst says the independent measurement vendors get demoted to auditors and the identity layer inherits the value. The Skeptic says none of it matters until 2027. The disagreement is really about how fast OpenAI decides advertising funds its P&L.

Third tension: leverage. Publishers are pushing to be in the attribution conversation, arguing their content feeds the AI answers. The Market Analyst thinks that's the right noise and the wrong ask. A seat at an IAB working group is not a data licensing agreement, and only one of those pays.

What it hinges on. Two beliefs. One, whether AI platforms ever depend on advertiser confidence enough to open measurement. If they stay subscription-first, there's no pressure to open the box and the Skeptic wins on timing. Two, whether the standard gets set in the open or coded per-platform. The Facebook precedent says per-platform, and once it's set it's set. The council leans toward the Facebook replay on structure, and toward the Skeptic on timing. So: the outcome is close to predetermined, but slower than the November deadline implies.

What to de-risk now. Audit which of your attribution integrations already go blind through an AI intermediary, because that's a today problem, not a 2027 one. If you're a publisher, chase a data licensing conversation with an AI platform directly, not a seat on a framework committee. If you're a brand, make log-level or clean-room verification a line in the next renewal, before the black box closes and the ask gets harder.

Prediction: No major AI platform (OpenAI, Perplexity, Google's AI answers, Anthropic) will commit to sharing log-level, independently verifiable ad-attribution data with third-party measurement vendors as a result of the IAB framework, and the version they eventually ship will be a platform-coded integration that vendors audit rather than run themselves, confirmed by the shape of the first AI-platform ad measurement integration announced before the 2027 upfront (roughly May 2027).

Confidence: Medium. The incentive is clear and the Facebook precedent is direct, but timing depends on when AI ad revenue becomes material enough to force the question.

Why: The platforms that would have to open up make their money on subscriptions today, so they have no revenue reason to hand independent auditors the raw signal, and every reason to control it, which is exactly the Facebook pattern Michael Bishop of OpenAds described, where Facebook coded its own measurement and vendors rubber-stamped it. An IAB framework is a schema recommendation with no enforcement power, so it can't compel a black box to open. The opposite outcome, a platform voluntarily exposing log-level data to competitors' auditors before advertisers force it, would require them to give up control they currently have no reason to surrender.

Revisit by 2027-05-15: We're right if the first ad-measurement integration shipped by any major AI platform is platform-coded with vendors in an audit-only role, or if no platform has opened log-level data to third-party measurement. We're wrong if any of OpenAI, Perplexity, Google, or Anthropic ships an integration that lets an independent vendor run its own measurement on raw, unfiltered attribution signal.

The IAB framework is table-stakes lobbying. The moat fight is over the schema, and the platforms hold the pen.