Skip to content

Voices from the open internet

Opinion

IAB’s AI attribution blueprint: A 25-Page confession that the industry has no idea who sold the shoe

The IAB’s new framework sorts AI-influenced sales into three buckets, but the evidence needed to prove any claim sits with platforms that haven’t signed on.

A pair of shoes sits on a bar graph winning podium

Credit Illustration by Terrence Wilson / Getty / The Current

By

Principal and Chief Analyst, W Media Research

November 18

Imagine a shopper asking a chatbot for the best running shoe under $150, getting a confident answer, and going to bed. Three days later, the shopper searches for the brand and buys the shoe. The brand’s dashboard logs a branded search conversion, and the search team probably gets a bigger budget next quarter. The problem: the chatbot sold the product, but search gets the credit. 

The IAB’s AI-Era Attribution Blueprint, released November 12, attempts to fix this mess. The diagnosis is correct, and the proposed cure makes sense in theory. In practice, it requires cooperation from companies with little incentive to surrender their measurement secrets. 

The crime scene 

The numbers are damning. Similarweb found that shoppers receiving an AI brand recommendation are 2.5 times more likely to visit the brand’s website within a week, yet only 8.8% of those visits arrive through a click from the assistant. More than half show up as search traffic and another fifth as direct traffic. Adobe reported AI-sourced traffic to U.S. retail sites growing 393% year over year in Q1 2026, with those visitors converting 42% better than other traffic by March. 

So, roughly nine in ten visits following an AI recommendation arrive wearing somebody else’s name tag. Marketers celebrating booming branded search traffic should feel a chill, retail media buyers most of all. A shopper may arrive at a retailer with a shortlist the chatbot already wrote, then encounter a sponsored product ad that gets the applause. Retail media networks have graded their own homework for years. Now an assistant may have done much of the selling before the shopper even saw the ad. 

Three buckets and a claim

The framework sorts conversions into three buckets: human-owned, AI-assisted, or AI-executed, depending on who shaped the intent, made the decision, and completed the purchase.

An agent that merely reorders dog food you already selected earns no attribution claim, a ruling I especially like. Execution is a service, like the guys carrying your couch upstairs. They don’t deserve credit for your taste in furniture.

Depending on its role, a party may earn a “claim,” which the IAB stresses is merely the right to be evaluated for credit. Moreover, the document refuses to divide sales into arbitrary percentages because there is no defensible way to do so. The hard part is left to whoever evaluates the claim, meaning a brand, an agency or an auditor.

The evidence lives in somebody else’s house

My favorite phrase in the blueprint is referring to the second evidence category, “credit recognized but not yet measurable.” So far, publishers get the rawest deal. Their reviews may inform a chatbot’s recommendation, making them eligible for a contributory claim, but the connection between their content and the eventual sale generally cannot be demonstrated today.

That’s because closing that gap requires AI platforms to disclose whether recommendations were paid or organic, which sources informed them, and what their systems recommended or executed. It also requires consistent labeling of AI-generated traffic.

All of that is voluntary, with no auditor, no certification and no deadline. The blueprint’s contributors count illustrious names among them, such as Microsoft, Walmart, WPP and Prudential, ad-tech firm Criteo and startups that sell AI tracking. OpenAI, Google, Meta, Anthropic and X are conspicuously absent, yet those companies hold the receipts. Still, the IAB is not discouraged (or so they say): viewability took years to take off as a viable measurement category. With OpenAI now selling ads inside ChatGPT, the incentive to self-report should be growing.

What marketers should do tomorrow

Treat the three pathways as a diagnostic tool rather than a new attribution model. Estimate the share of conversions in each bucket using surveys, holdout tests, and modeling.  When a platform reports “AI-driven” sales, ask which pathway, which claim type, and what evidence supports the claim. Separate platform-reported numbers from independently verifiable lift.

Similarly, publishers should begin documenting the contributory claims their content creates, even if the connection to individual sales remains unmeasurable.

The blueprint identifies the missing evidence and who needs to provide it. For an industry that spent a decade arguing over multi-touch attribution while signals kept disappearing, that’s progress. The next challenge is persuading the companies holding the evidence to share it.


This op-ed represents the views and opinions of the author and not of The Current, a division of The Trade Desk, or The Trade Desk. The appearance of the op-ed on The Current does not constitute an endorsement by The Current or The Trade Desk.

Subscribe to The Current

Industry news, expert analysis, behind-the-scenes podcasts and more—straight to your inbox.