The IAB Has Mapped the AI Attribution Gap. Deterministic Tracking Won't Close It.

The IAB's new AI-Era Attribution Blueprint shows 9 in 10 AI-influenced visits are credited to something else. It gets the problem right. But its evidence model still leans on the deterministic signals that created the gap.

Chris Liversidge: CEO & Founder @ Corvidae AI By Chris Liversidge · 18th November, 2026 · Attribution
As Trusted By...

Key Points

  • The IAB's AI-Era Attribution Blueprint, published November 2026, gives the industry a shared way to classify AI's role in a conversion and the evidence needed to claim credit for it.
  • Citing Similarweb data, it shows that only 8.8% of visits that follow an AI brand recommendation arrive by clicking from the AI itself. 55.9% arrive as search and 19.9% as direct.
  • The blueprint's evidence model relies on two things: deterministic trails such as referrals and citation clicks, and aggregate methods such as holdout tests and surveys.
  • That leaves out a third route that already works: reconstructing journeys from first-party data, so an earlier AI-referred visit stays attached to the later search or direct visit that converts.
  • Under a strict reject-all consent model, Corvidae tracked over 95% of RS Group PLC's active customer base and found journeys 46% longer in the UK and 68% longer in the EU than the platforms reported.

What the IAB Has Published

This month the IAB published its AI-Era Attribution Blueprint, produced by a working group led by the IAB with members from Walmart, WPP Media, Microsoft, Criteo, Tinuiti and others. It is the most serious attempt yet to answer a question every marketer now has: when AI shapes a purchase, who gets the credit, and on what evidence?

The framework asks three questions at the point of sale: who or what created the intent, who or what made the decision, and who or what executed the transaction. From the answers, every conversion is given one pathway:

  • Human-Owned, where the human held all three roles.
  • AI-Assisted, split into AI-Influenced (AI shaped intent, the human decided) and AI-Recommended (AI shaped the final choice).
  • AI-Executed, split into specific instruction (the human chose what to buy in advance) and open instruction (the AI chose and bought).

Each pathway then determines who is eligible to make a claim: a decision claim, a contributory claim, or none. The blueprint is careful to separate three steps the industry usually collapses: pathway, then evidence, then credit. As it puts it, a claim is not credit. It is the right to be evaluated for credit.

IAB & Similarweb Estimates Significant Influence

IAB & Simliarweb Nov '26: 9 In 10 AI Influenced Visits Credited To Something Else
IAB & Simliarweb June '26: 9 In 10 AI Influenced Visits Credited To Something Else

The Number That Matters: 9 in 10

The blueprint's most striking evidence comes from Similarweb's June 2026 study of US desktop visits. Consumers who received a brand recommendation from an AI assistant were 2.5 times more likely to visit that brand's site within seven days. But only 8.8% of those visits came from clicking through from the AI itself. The rest arrived wearing other labels:

  • 55.9% as search
  • 19.9% as direct
  • 13.5% as referrals
  • 1.9% as other

The report also cites Adobe Analytics data showing AI-sourced traffic to US retail sites grew 393% year on year in the first quarter of 2026, and by March converted 42% better than non-AI traffic.

The consequence is not that AI-influenced sales vanish. It is that they are credited to the wrong source. A dashboard that credits search and direct sends the next pound to search and direct, while the channel that created the demand is starved of investment.

What the Blueprint Gets Right

A great deal. Three things in particular.

First, it defines pathways by what actually happened, not by what today's tools can see. It explicitly refuses to let "AI-Assisted" quietly mean "AI-Assisted with a surviving referral signal". That is exactly the right instinct, and one the industry has rarely shown.

Second, it insists on separating claims supported by platform-reported evidence from claims supported by independently verifiable evidence, and says independent verification matters more as AI platforms take on more of the journey.

Third, it asks AI platforms for four specific disclosures: whether a product appeared through paid placement or organic recommendation, which sources informed it, what the system surfaced, recommended or executed, and consistent labelling of AI-originated traffic. Every one of those should happen.

IAB Proposed Attribution Journey Framework

IAB Assisted AI Journey Framework for Attribution
IAB Assisted AI Journey Framework for Attribution

Where It Stops Short

Look closely at the evidence the blueprint says is available today: referral and citation data, UTM tags and referrers, affiliate links, seller purchase records, agent transaction records, holdout tests and shopper surveys. It falls into two camps. Either a deterministic trail survives from the AI to the sale, or the effect is inferred in aggregate across many shoppers.

Everything in between is filed under "credit recognised but not yet measurable". That includes the scenario the blueprint itself uses to illustrate the problem: a shopper gets an AI recommendation, comes back days later through a search, and buys. In the blueprint's words, those later searches and typed-in URLs leave no trail.

That is only true if you assume the trail has to be carried by an identifier. A cookie that expired, a consent banner that was rejected, a switch from phone to laptop, a return visit three days later through Google: in deterministic tracking, each of these breaks the chain, and the earlier touchpoint is lost. But the earlier visit still happened on your site. The data is still there. It just has not been connected.

The blueprint is right that the industry should not build today's measurement limits into its categories. It should not build them into its evidence standards either. By treating a broken identifier as a missing signal, it accepts the deterministic model's limits as the limits of what can be known.

Reconstruct the Journey, Don't Chase the Identifier

There is a third route between a surviving deterministic trail and an aggregate holdout: reconstruct the journey itself.

Corvidae's patented technology rebuilds customer journeys from raw first-party eventstream data using predictive AI, with no cookies and no persistent identifiers, and is compliant with GDPR and CCPA by architecture. It does not need a referral signal to survive into the converting session. An AI-referred visit that does not convert stays attached to the search or direct visit that does, so the AI touchpoint is recorded in the journey rather than overwritten by the last click.

We know this works at scale because we have published the results. RS Group PLC, the FTSE 250 distributor, moved to a strict reject-all consent model, after which conventional analytics lost most of its raw observation and fell back on modelling. Across RS Group's properties worldwide, over a four-month test, Corvidae:

  • tracked over 95% of the active customer base, without a single cookie
  • tracked £193m more revenue than the existing analytics
  • found customer journeys 46% longer in the UK and 68% longer in the EU than the platforms had reported
  • found that 58% of all sales touched a Paid Search ad, and 72% of those also involved an Organic Search interaction

Those longer journeys are the point. The touchpoints that platforms were missing were overwhelmingly the earlier, upper-funnel ones, which is exactly where AI recommendations sit. Read against the IAB's framework, reconstruction moves a large share of AI-related claims from "not yet measurable" to "evidence available today", and does it with independently verifiable evidence rather than platform self-reporting.

A traditional conversion journey using cookies compared with Corvidae's AI conversion journeys
A traditional conversion journey compared with the same journey reconstructed by Corvidae AI.

Being Precise About the Limits

Reconstruction is not magic, and a framework this careful deserves a careful answer. Where AI influence produces no visit to your site at all, because the shopper read an answer and later typed your name into a search box, no site-side method can see the AI conversation that came before. For that slice the blueprint is right: holdouts where exposure can be withheld, shopper-reported data and modelling are the tools available, and the platform disclosures it asks for are essential.

Nor does attribution establish incremental value on its own. The blueprint says so, and we agree. That is why we test it. In RS Group's four-month incrementality pilot in Italy, feeding reconstructed journeys into automated bidding delivered 6% more Paid Search revenue on 5% less ad spend, a 12% improvement in ROAS.

But the honest position is not that AI influence is unmeasurable until the platforms decide to disclose more. It is that a large part of what is being credited to search and direct today can be reconnected now, from data brands already own.

Chris Liversidge: CEO & Founder @ Corvidae AI

The IAB has done the industry a real service by calling out the AI attribution gap: AI is shaping purchases that our dashboards credit to search and direct.

But the answer cannot be to wait for AI platforms to mark their own homework, or to accept that a broken trail is a lost one.

Lets stop chasing identifiers and reconstruct the journey from your own first-party data, and most of that trail is still there.

— Chris Liversidge: CEO @ Corvidae AI

What Brands Should Do Now

  1. Adopt the IAB's pathways. Estimate what share of your conversions are Human-Owned, AI-Assisted and AI-Executed. Even a rough split will change how you read your channel reports.
  2. Stop treating the referral signal as the edge of what you can know. If your measurement only credits AI when a referrer survives into the converting session, you are seeing roughly the 8.8%, not the whole picture.
  3. Reconstruct journeys from first-party data. Connect AI-referred visits to the later search and direct visits that convert, without depending on cookies or persistent identifiers that consent rules and browsers keep removing.
  4. Separate platform-reported credit from independently verified evidence, as the blueprint recommends, in every report you present.
  5. Test for incrementality where you can, especially on paid AI placements where exposure can be withheld, and ask every AI platform for the four disclosures the IAB has set out.

The IAB has drawn an accurate map of the AI attribution gap. The industry should use it. It should not settle for the idea that the only way across is to wait for someone else to build the bridge.

Ready to try attribution the Corvidae way?
Work with your dedicated Success Manager

Getting started on the road to accurate attribution doesn’t have to be complicated. Our team are on-hand to make the move to a new attribution tool as easy as possible.

Author
Chris Liversidge: CEO & Founder @ Corvidae AI
Chris Liversidge
Founder and CEO of Corvidae AI
Latest Insights
Our Accreditations & Partnerships

Contact Us

Newsletter Signup