IAB's AI attribution blueprint names the gaps, not the fixes

Only 8.8% of visits from consumers who received an AI brand recommendation came directly from AI referrals. Similarweb data cited in the Interactive Advertising Bureau's newly published "AI-Era Attribution Blueprint" found those consumers were 2.5 times more likely to visit the brand's website within seven days, yet 55.9% of those visits are credited to search and 19.9% to direct. The IAB spent months trying to close that gap. It could not, because the underlying data stays inside AI platforms.

What did the IAB try to standardize?

Digiday first reported in August that the IAB was developing an AI attribution framework to standardize how the industry measures and credits AI's role in conversions. The working group combined tech companies, publishers, agencies, measurement vendors and brands: People Inc., Walmart, WPP Media, Emberos, Criteo, Bluefish, Microsoft, Brandlight, Tinuiti and Prudential Financial. The blueprint landed this week after roughly four months of work.

Caroline Giegerich, the IAB's vice president of AI, spoke on a virtual IAB roundtable on Monday. “Going in, I thought that we were going to help figure out how credit was assigned through the process. I realized actually what this document was starting to become was more of a call to action to the industry,” she said. “There are evidence layers that are missing,” she added, and the document “started in one place, and over the course of four months, ended in another.”

Why does the measurement gap persist?

Recommendations, purchase decisions and transactions now happen inside AI platforms, and advertisers cannot see that activity. Impressions, clicks and referral data are not always available when an assistant guides a consumer from product discovery to checkout. Even when an AI recommendation does lead to a sale, that conversion can be miscredited to search, direct traffic or another channel, the IAB reported. The distortion reaches budgets: spend flows to channels that capture conversions, not those that influenced them.

The blueprint sorts the journey into three parts. It classifies who or what created the intent, made the decision and executed the transaction. It defines the additional claims AI involvement creates, plus the evidence each claim requires. And it names the transparency and measurement gaps that make those claims hard to verify.

Why publisher content is the hardest case

Ask an answer engine for the best option in a category, and the answer may draw on a publisher's review the consumer never clicks. The retailer then logs the sale as search or direct. The IAB's report reads: “The review shaped the sale, and the publisher is eligible for a contributory claim. The evidence differs depending on how publisher content informed the answer.” Training content is generally not traceable from outside the model; grounding and citation are more observable.

What the IAB called the missing piece: “What remains largely missing is the connection between those sourcing signals and downstream conversions. Until that link exists, the contributory claim cannot be fully substantiated.”

What must platforms, brands and publishers disclose?

Platforms should disclose whether recommendations are paid or organic, which sources inform them, and label AI-originated traffic consistently. Brands and agencies should apply the framework's claim-and-evidence standards across platforms. Publishers should identify claims about their content's influence and the evidence needed to support them.

Jaime Schultheis, head of global data partnerships at Bombora, put the imbalance plainly: “Publishers are feeding the answers but they don't have the receipts.” Jon Roberts, chief innovation officer at People Inc., said the framework still lets publishers compare measurement vendors: “Everyone's pitching you on their custom proprietary measurement solution, and they're all using the same terms, and the terms all mean something different.”

Where first-party visibility comes from

The U.K.'s Competition and Markets Authority requires Google to provide U.K. publishers with clickthrough data for its AI search features by Dec. 3. OpenAI is testing an analytics dashboard that counts how often a business appears in ChatGPT answers and how many clicks those appearances send to a site. Graham Howes, senior vice president of distribution and AI intelligence at BDG, said he is “waiting [with] bated breath” for that data, which he called incredibly valuable for BDG's GEO offering.

Roberts said publishers remain, in his words, “flying blind” without it, while adding that third-party measurement still matters for verifying platforms' own numbers. Brands feel the same squeeze. Ross Bucholc, associate media director at Prudential, said the lack of first-party data from AI platforms is holding back his ability to unlock more ad budgets going to LLMs.

Can AI visibility be tied to sales?

Not yet, according to Josh Blyska, head of insights at Profound. His platform measures AI visibility, citations and brand appearances in AI answers, but cannot reliably track whether those interactions produce purchases. In an analysis of browser-session data, Profound found one in three consumers had a relevant AI conversation about a purchase within a month that went unattributed. The dataset excluded Google's AI Mode and AI Overviews. “It's a really tricky problem right now,” he said.

John Sills, head of sales, North America at The Guardian, said the industry lacks a consistent way to understand how AI patterns surface answers: “We're not measuring CTRs here. There's no way of really understanding where the attribution is coming from.” He added that the LLMs are now in the ad business and need the same hands-on support as Facebook and YouTube campaigns.

Publishers are testing partial fixes. Time is working with Mobian to measure the performance of ads served to AI agents. Bustle Digital Group is working with Profound to measure GEO campaign performance for brands. Neither closes the link to sales.

The compensation fight

Matt Murray, executive editor of The Washington Post, defended the paper's licensing deal with OpenAI at Web Summit in Lisbon. “Ninety-five percent of the AI referrals we get now come through OpenAI. So to me, that's a win for them and a win for us if that's how people want to access the news,” he said.

Numbers to know

• $330 million: the reported price under consideration for The New York Times Company's potential acquisition of Letterboxd, an online community for film enthusiasts.
• 2.7 billion: the number of times Ziff Davis content was cited in Google's AI Overviews and AI Mode in Q3, nearly 10 times the clicks those sites received from Google Search.
• 37%: Ziff Davis' year-over-year decline in Google referrals.
• 44%: BuzzFeed's total year-over-year revenue decline in Q3 2026.
• 4,000: new subscribers The Washington Post gained across U.S. midterms election day and the following day, per Semafor.

Also on our radar

• Google added opt-out controls for AI search features under CMA pressure, while making publisher links more prominent in AI Overviews and AI Mode.
• Anthropic, Google, Microsoft and OpenAI backed a proposed robots.txt “policy token” for AI training opt-outs. It is not an official standard and cannot be enforced.
• Meta indefinitely paused terms of service changes that would allow it to scrape publisher sites; Yahoo is expanding its AI search engine Scout across Yahoo News, Finance and Sports.
• AppLovin's Q3 revenue rose 55% to $2.19 billion, ahead of Unity at 32% and Magnite at 14%, while Criteo, Taboola and Teads fell 11%, 6.9% and 26%.
• Press Forward has invested nearly $600 million in news organizations, two years ahead of schedule, and Anthropic struck its first news licensing deal with the Associated Press for real-time midterms data in Claude.

Dec. 3 is the near-term test. If Google's clickthrough data and OpenAI's dashboard arrive with enough detail to verify claims, publishers may start collecting the receipts they currently lack.