IAB's Gabilan: Agentic AI Breaks Core Ad Measurement Assumptions

When the Buyer Is a Bot, the Impression Means Nothing

Cintia Gabilan, who runs product development at the IAB, is saying plainly what most of the industry has been stepping around: when an AI agent does the browsing, comparing, and buying, the impression is dead. No human eye on the ad. Viewability scores, click-through rates, and last-touch attribution (crediting the final ad seen before a purchase) all stop meaning anything. Her argument is that 2026 is the year the industry agrees on new measurement definitions together, or the AI platforms write those definitions for us.

That second outcome is the more likely one.

The measurement break is real; the timing is not

Agent-driven purchases are a rounding error in today's campaign data. The CFOs demanding ROI proof are focused on internal AI tools, not agent ad metrics. The breakage is genuine in theory and invisible in your reports, and it will stay invisible until agent-driven buying is large enough to show up in an A/B test. Probably 2027 and later before it touches your P&L directly.

The definitional fight, though, is happening now. Whoever writes the vocabulary owns the auditing standard, and auditing is a recurring-revenue business. The platforms that own the agent, the shopping surface, and the checkout data can define and enforce a measurement spec unilaterally the moment it helps them sell ads. The IAB needs member consensus across competitors who benefit from delay. Those two move at very different speeds.

Gabilan's own framing gives the game away. She wants to lead "instead of letting it be defined by each LLM themselves," which means she already sees OpenAI and Google moving first. The same pattern played out when Google and Meta pulled audience data inside their walls and left neutral data management platforms debating standards nobody adopted.

Who loses when the numbers go quietly wrong

The verification vendors, DoubleVerify and Integral Ad Science (the two dominant third-party ad measurement firms), are the obvious beneficiaries in the lazy read. More complexity, more verification. That logic is backwards. Their moats are built on a decade of human-behavior signal libraries. Agent traffic has no eye to measure for viewability and no behavior to score for fraud. Those libraries don't error out cleanly when the actor is a software process. They report nonsense.

The fee migrates to whoever owns the transaction layer: Amazon, Shopify, LiveRamp (which holds the identity graph that could resolve a non-human actor). When the thing you measure stops being a person, the company closest to the purchase inherits the measurement business.

The silent risk for publishers is CPM compression. If buyers start distrusting impression-based reporting before a new standard lands, they discount inventory near AI-shopping surfaces for uncertainty. That bill doesn't arrive as a line item you choose to pay. It arrives as softer prices.

What to do now, cheaply

Audit which of your KPIs return garbage when the actor is not human. Get your trading-desk team trained to distinguish agent traffic from fraud before the false-positive wave hits. Agency desks will see strange click-through patterns on inventory touching AI-shopping surfaces, flag it as fraud, and pause spend. That is a triage problem the verification vendors are not staffed to handle. The audit costs little and pays off whenever the volume arrives.

Our call: Before the IAB publishes a finalized, industry-adopted agentic measurement standard, at least one of OpenAI or Google ships its own agent-commerce measurement spec inside its platform, on its own terms, by the Q4 2026 earnings calls in February 2027. Medium confidence. Standards bodies move at committee speed. Platforms move at product-ship speed, and they have no incentive to wait for shared vocabulary that constrains them.

Cintia Gabilan, who runs product development at the IAB, says the thing most of the industry has been politely not saying out loud: when an AI agent does the browsing, comparing, and buying, the impression is dead. No human eye on the ad. No human hand on the click. Viewability, click-through rate, and last-touch attribution (crediting the last ad someone saw before buying) all stop meaning anything. Her pitch is that 2026 is the year the industry agrees on new measurement words and numbers together, or each big AI platform writes those definitions for us.

So the question for an operator: is this a real break in how we get paid, or a standards body making noise about a trend that is still years from your P&L? And if it is real, who captures the new measurement money?

How hard is this to undo? Easy to undo on your side. Nobody is forcing you to rip out your measurement stack this quarter. The expensive, hard-to-undo decision belongs to whoever sets the standard, because standards calcify and recurring audit revenue flows to whoever owns the definition. For a publisher or DSP, the real exposure is slower: budgets quietly repricing as buyers lose faith in metrics they can no longer trust.

What's actually being decided. Not "do we believe agents are coming." It's who owns the measurement language when they arrive: a neutral body like the IAB, or the AI platforms (OpenAI, Google) that sit on the transaction.

What sets the deadline. Nothing hard yet. Gabilan's "critical year" framing is her deadline, not the market's. Agent purchase volume sets the real clock, and it isn't ticking loudly.

---

The Skeptic. Agentic commerce at advertising scale is a 2027-and-later problem, and the IAB is doing what the IAB always does: convene a working group around a real-but-early trend so members can compete on vocabulary while nothing is at stake yet. The $68.25 billion eMarketer figure for 2030 is a forecast draped over a loose definition of "AI ad spending" that already folds in boring algorithmic bidding. Actual autonomous agent purchases are a rounding error today. The CFOs demanding ROI proof are demanding it on their own internal AI tools, not on agent ad metrics. Plain version: the crisis is genuine in theory and invisible in your campaign reports, and will stay invisible until agent-driven buying is big enough to show up in an A/B test.

The Market Analyst. The lazy trade is long the verifiers, DoubleVerify and Integral Ad Science, on the logic that more complexity means more verification. That has it backwards. Their moats are methodological: a decade of human-behavior signal libraries. Agent traffic makes those libraries misfire, because a bot process has no eye to measure for viewability and no behavior to score for fraud. The better-positioned names own the transaction layer where agents actually transact, or the identity graph that could resolve a non-human actor, Amazon, Shopify, LiveRamp. In plain terms: when the thing you measure stops being a person, the company closest to the purchase, not the company grading the ad, inherits the measurement fee.

The Strategist. Whoever writes the vocabulary wins the auditing standard, and auditing is a recurring-revenue business. The three-year fight is whether that authority sits with a neutral body (IAB, MRC) or gets baked into platform APIs, OpenAI's shopping-agent logs, Google's agentic ad layer. If the platforms define presence and persuasion in their own proprietary terms, the verifiers get squeezed the same way data management platforms did when Google and Meta pulled audience data inside their walls. The opening is for agent-native measurement firms spun out of AI infrastructure rather than ad-tech. Plain version: control of the measurement dictionary is worth more than any single product, and the platforms have every reason to grab the pen.

The Operator. The first thing that breaks is brand safety and viewability scoring returning null when the "viewer" is a software process. Your IAS and DoubleVerify dashboards built on human behavioral proxies will not error out cleanly, they will quietly report nonsense. At roughly 90 days, agency trading desks start seeing weird click-through patterns on any inventory touching AI-shopping surfaces, flag it as fraud, and pause spend. That is a false-positive problem the verification vendors are not staffed to triage. Attribution teams at Criteo and the retail media networks have the harder job: last-touch logic falls apart when an agent comparison-shops five platforms in 200 milliseconds and buys on the sixth. Plain version: nobody gets a crash alert, they get numbers that look fine and are wrong.

The CFO. The money question is not "is agentic measurement coming," it's "when does it touch my revenue." Today, near zero. So spending to replatform measurement now is paying for a problem that hasn't arrived. The real cost is defensive: if buyers start distrusting impression-based reporting before the new standard lands, CPMs on anything near AI-shopping surfaces get discounted for uncertainty, and you eat that whether or not you spent a dollar on new tooling. Plain version: the bill doesn't come as a line item you choose to pay, it comes as softer prices on inventory buyers can no longer grade.

---

Where they split.

The Skeptic versus the Strategist on timing and stakes. The Skeptic says there's no fire this year, so IAB's urgency is theater. The Strategist says the fight over who writes the definitions is happening now even though the transactions aren't, and losing the definition is permanent. Both can be right: the measurement break is years out, and the standard that governs it gets set before the volume shows up.

The Market Analyst versus conventional wisdom on the verifiers. Everyone assumes DoubleVerify and IAS adapt because they always have. The Analyst says agent traffic breaks their core method faster than privacy regulation did, and the fee migrates to whoever owns the transaction.

The Operator versus the CFO on urgency. The Operator wants measurement teams auditing dashboards now because the breakage is silent. The CFO says spending ahead of real volume is burning money on a hypothetical.

---

What this hinges on. Two things. One, how fast real agent-driven purchasing shows up in actual campaign data, which is slow. Two, whether the IAB can set a standard before OpenAI and Google embed their own definitions in their platforms, which the platforms have strong incentive to do first and unilaterally.

The council leans skeptical on timing and bearish on the standards body winning the definitional fight. Standards bodies move at committee speed. Platforms that own the agent and the checkout move at product-ship speed, and they have no reason to wait for a shared vocabulary that constrains them. Gabilan's own quote gives the game away: she'd rather lead than "adapt and fix the gaps." That's someone who already senses the platforms are ahead.

What to de-risk now, cheaply: audit which of your KPIs return garbage when the actor isn't human, and get your trading-desk team ready to tell agent traffic apart from fraud before the false-positive wave hits. That costs little and pays off whenever the volume arrives.

---

Prediction: Before the IAB publishes a finalized, industry-adopted agentic measurement standard, at least one of OpenAI or Google will ship its own agent-commerce measurement or reporting spec inside its platform, defined on its own terms, by the Q4 2026 earnings calls in February 2027.

Confidence: Medium — platforms ship on product cycles; standards bodies ship on committee cycles.

Why: Gabilan is openly racing to define measurement "instead of letting it be defined by each LLM themselves," which means she already sees the platforms moving first. OpenAI and Google both own the agent, the shopping surface, and the checkout data, so they can define and enforce a measurement spec unilaterally the moment it helps them sell ads, while the IAB needs member consensus across competitors who benefit from delay. The same pattern played out when Google and Meta pulled audience data inside their walls and left the neutral data management platforms arguing about standards nobody adopted. The opposite outcome, a neutral body getting there first, would require the platforms to voluntarily wait for a shared vocabulary that limits them, and they have no reason to.

Revisit by 2027-02-28: We're right if OpenAI or Google has published or rolled out an agent-commerce measurement, visibility, or ad-reporting spec defined on its own platform terms before the IAB's standard is finalized and adopted. We're wrong if the IAB (or MRC) publishes an industry-adopted agentic measurement standard first, with no competing platform-proprietary spec shipped.