The Interactive Advertising Bureau on November 12, 2026 released the AI-Era Attribution Blueprint, a framework that classifies every sale according to whether a person or an AI system created the intent, made the decision and completed the purchase. The 25-page document, marked "Published November 2026" and developed by a working group led by Caroline Giegerich, the IAB's vice president for AI, sorts conversions into three pathways: Human-Owned, AI-Assisted and AI-Executed. It assigns no percentages of credit. Instead it defines which parties may claim a role in a sale, what evidence each claim requires and, at some length, where that evidence does not yet exist.

The problem it addresses is a reporting one. According to the IAB, a purchase shaped by an AI recommendation often arrives in analytics as search or direct traffic, so dashboards credit the channel they can see rather than the one that created the demand. The announcement cites the IAB's 2026 Outlook Study: September Update, saying 45% of ad buyers name comparing AI-driven and traditional customer journeys as their top measurement challenge and 44% rely on branded search and direct traffic as proxies for AI influence they cannot track.

In Short

The IAB has published a framework that labels each sale as Human-Owned, AI-Assisted or AI-Executed, depending on whether a person or an AI system sparked the interest, picked the product and paid for it. Advertisers, agencies and publishers are affected because sales that started with an AI answer often show up in analytics as search or direct visits, which can send the next budget to channels that did not create the demand. The framework hands out no percentages of credit: it sets out who may make a claim and what proof each claim needs, and it concedes that much of that proof sits inside AI platforms that do not yet disclose it.

What the IAB published

The blueprint has six sections: an executive summary, a list of contributors, a statement of the attribution problem, a taxonomy the IAB calls AI Influence Pathways, an AI Credit Claim Framework and a closing section on the measurement gap. It is published on the IAB's guidelines site.

"As AI takes on more of the purchase journey, from sparking interest to completing the sale, the clicks and referrals that attribution depends on are thinning out or disappearing," Giegerich said in the announcement. "AI-influenced sales are sometimes reported as search or direct traffic, which can send the next marketing investment to the wrong channel. The AI-Era Attribution Blueprint works alongside existing attribution systems, defining the additional credit claims AI creates and the evidence each one requires."

The document builds on the IAB's Measuring Visibility in the AI Era playbook, released on August 3, 2026, which found that only 16% of brands systematically track AI visibility. According to the IAB, the new work extends that effort "from how brands show up in AI platforms to how that presence connects to sales."

The opening pages set the tone. A panel headed "Maximize this Blueprint" suggests readers put questions to an AI platform of their choice, and offers an example: "A platform says AI drove 20% of my sales. What data should I ask for to validate it?" Much of what follows is an answer to that question.

Where the attribution trail breaks

The blueprint starts from the mechanics of conventional digital attribution, which in "its dominant deterministic form" was built on three observable signals: exposure to an ad, a trackable click or interaction, and a sale. An ad server logged the impression, a click carried a signal forward, and the conversion landed somewhere that could identify how the visitor arrived.

It does not pretend that system was ever complete. "Attribution has never provided a complete or causal view of the customer journey," the document states, listing signal loss, walled environments, offline behavior and the gap between touchpoints and incremental impact as problems that "all predate AI." What AI adds, according to the blueprint, is a new journey of "prompt, recommendation, action," in which research and comparison that once spread across many trackable pages happen inside a single conversation.

The document identifies five ways the trail weakens in that journey:

  • No required click. A consumer can leave an AI conversation informed and decided without clicking anything. Television, display and zero-click search also influenced without clicks, the blueprint notes, but they worked at the awareness end, and the research that followed "still generally produced clicks."
  • Fewer referral signals. Even when a shopper clicks through, the signal identifying the AI source "may not survive the handoff."
  • No conventional impression. An organic recommendation inside an AI answer is not a served-media impression, though "some emerging AI metrics now count brand or link appearances as impressions."
  • Walled transactions. Platforms keeping transaction data to themselves is an old problem in marketplaces, social checkout and retail media. "What is new is what sits behind the wall: whether a product was recommended and why, so a brand can be recommended and sold without ever knowing it."
  • Agent autonomy. An agent acting on standing instructions can create a purchase with no browsing session at all.

The consequence, according to the IAB, is not that AI-influenced sales vanish. "They are often reported under the wrong source of influence." A sale shaped by an AI recommendation may arrive as branded search, direct traffic or an ordinary retail media conversion.

The figures behind the argument

Two outside data sets carry the case for scale. The first comes from Adobe Analytics: traffic from AI sources to US retail sites grew 393% year over year in the first quarter of 2026, and by March it converted 42% better than non-AI traffic, according to the blueprint. The IAB attaches its own caveat: "This reflects only the AI activity measurement can currently see."

The second is Similarweb's June 2026 study, The Downstream Impact of AI Visibility, based on US desktop visits, which found that only 8.8% of AI-influenced visits came from the AI itself. Consumers who received a brand recommendation from an AI assistant were 2.5 times more likely to visit the brand's site within seven days. Search delivered 55.9% of those visits, direct traffic 19.9%, referrals 13.5% and other sources 1.9%. The blueprint charts the split under the heading "9 in 10 AI-influenced visits are credited to something else."

The mechanism is mundane. A consumer hears about a brand in an AI answer, searches for it by name days later and arrives from the results, "where web analytics logged a typical search visit." A consumer who types the address arrives as direct traffic, "which leaves no trail at all."

Two scenarios

The blueprint illustrates the problem with two cases. In the first, a consumer asks an AI assistant for the best running shoe under $150 and is pointed to a specific model from a hypothetical Brand X. Three days later the consumer searches for Brand X and buys the shoe at a retailer. Analytics credit search, and "the AI recommendation that initiated the purchase is absent from the measurement record."

In the second, a shopper researches in an AI assistant, reaches a retailer with a short list, then sees retailer search and display ads before buying. The retail media trail supports a claim that the ads contributed. "But the AI may have determined which brands entered consideration in the first place. Did the media create demand, influence the final choice, or simply capture demand that already existed? Current attribution systems cannot reliably distinguish among those roles."

That second case carries weight for retail media networks, whose pitch to advertisers rests on closed-loop sales data. In April, an IAB white paper co-authored by Collin Colburn, the IAB's vice president for commerce and retail media and now a member of the blueprint working group, argued that legacy measurement undervalues retail media. The new document raises the possibility that some retail media conversions capture demand an AI system had already shaped.

Three questions, one pathway

The core of the framework is three questions asked at the point of sale. Who or what created the intent or interest? Who or what made the decision? Who or what executed the transaction? Each conversion receives exactly one pathway. Earlier AI interactions in a journey "may provide evidence for that classification, but they do not create separate or competing claims on the same conversion."

The IAB is careful about causation. "Intent rarely has a single source," the document states, since brand experience, advertising, word of mouth and prior consideration may all build up before an AI interaction. The framework "does not attempt to identify the original cause of demand."

One design choice runs through the whole document: pathways describe what actually happened, not what measurement tools can observe. "A conversion that AI shaped is AI-Assisted even when no observable signal survives to show it." Defining categories by what can be tracked, the blueprint argues, would bake today's limits into the taxonomy, so that "'AI-Assisted' would quietly come to mean 'AI-Assisted with a surviving referral signal,' and every conversion without one would default to search or direct." The safeguard runs the other way too. Classification and evidence are separate steps, and for a single conversion "a party should assert a pathway only where evidence supports it." Across many conversions, the mix of pathways can be estimated through shopper surveys, lift tests and modeling.

Human-Owned

A conversion is Human-Owned when the person created the intent, made the decision and executed the purchase, even if AI appeared somewhere along the way. The document's example is a shopper replacing running shoes with the same model, who asks an AI assistant whether the newest version runs true to size, compares prices at two retailers and buys from one. "The AI answered a question the shopper brought to it, but it did not shape what they wanted or what they chose."

AI-Assisted

A conversion is AI-Assisted when the human completes the purchase but AI shaped the intent, the decision or both, by "creating awareness, framing the available options, narrowing the choice, or recommending a specific product." Clicking through from an AI answer does not change the label: "A clickthrough does not erase AI's role in surfacing the product or source."

The pathway splits in two. In an AI-Influenced conversion, AI contributes to intent, including when an agent raises a need unprompted, such as noticing a product is running low, but the human weighs the options and decides. The document likens the credit logic to upper-funnel media. In an AI-Recommended conversion, AI "materially shapes the final choice," narrowing the field to a specific product, brand, seller or course of action that the human then accepts.

Where the purchase happens matters for evidence. When an AI-Assisted sale happens off-platform, the AI interaction often leaves "little or no usable signal at the destination." When it happens on-platform, the handoff is easier to track, but "the platform controls much of the available information."

AI-Executed

A conversion is AI-Executed when a person gave an AI system permission to buy in advance and the system completes the purchase in the moment. Agent-to-agent transactions, such as a buyer's agent dealing directly with a seller's agent, are included. Where one agent hands part of the job to another, "claims follow the role each agent actually played, not simply the last agent to touch the transaction." An agent that suggests a purchase and waits for approval does not qualify; the in-the-moment approval makes the sale AI-Assisted.

The IAB then draws a second line. Under a specific instruction, the human chose the product in advance and the agent buys when a condition is met, as with an instruction to reorder the same dog food whenever supply falls below a set level. Credit follows Human-Owned logic. Under an open instruction, the human sets a broader objective and the AI selects the product, so credit follows AI-Recommended logic, "with AI treated as the decision-maker."

The resulting grid is compact:

Pathway Intent Decision Execution
Human-Owned Human Human Human
AI-Assisted: AI-Influenced AI Human Human
AI-Assisted: AI-Recommended Human or AI AI Human
AI-Executed: specific instruction Human Human AI
AI-Executed: open instruction Human AI AI

The blueprint concedes that its most consequential line is also the hardest to see. "The hardest distinction in this framework to observe is the line between AI shaping intent only and AI shaping the decision." A retailer may know a shopper arrived after an AI interaction without knowing whether the AI introduced the product, recommended that exact item or confirmed a choice the shopper had already made.

What counts as AI

The framework defines AI by what it does, not by what it is called. "An AI system is one that generates a recommendation or takes an action that replaces part of the normal search-and-click journey," whether through a chatbot, an AI search experience, a conversational assistant or an autonomous shopping agent. The same chatbot may answer a factual question in one journey, recommend a product in a second and complete a purchase in a third.

Conventional retailer features are excluded. Product carousels, "customers also bought" modules and personalized shelves operate inside the traditional search-and-browse journey and are already covered by retail media and site attribution. The exclusion is "behavioral, not locational," so an assistant inside a retailer's own app that narrows options or recommends a product is classified like any other.

Two further distinctions add context without creating new pathways. Transaction location is either on-platform, inside the environment where the AI operates, or off-platform. The on-platform examples in the document are ChatGPT, Microsoft's Copilot, Amazon's Rufus and Walmart's Sparky, along with third-party AI embedded in another company's site or app. Location is judged relative to the AI being classified: a shopper who leaves one AI surface and buys at a retailer has transacted off-platform, even if that retailer runs an assistant of its own.

Commercial relationship is either paid, where the product's appearance is tied to a paid placement or sponsored recommendation, or organic. The blueprint adds a clarification with consequences for content licensing deals: a commercial relationship that provides access to grounding content or data "does not make a recommendation paid unless that relationship influences the product's appearance or selection."

A claim is not credit

The second half of the blueprint turns classification into eligibility. Every claim is one of three types.

A decision claim applies when an AI surface or agent determines the final purchase choice. The IAB spells out its limits: it "does not mean AI caused the sale, had the greatest overall influence, or deserves the most credit." The example is a shopper who already plans to buy sunscreen and asks an AI assistant which brand to choose. If the shopper accepts the recommendation, AI may hold the decision claim "even though the underlying demand for sunscreen came from elsewhere."

A contributory claim applies when an AI answer moved the shopper toward the purchase without determining the final choice. Eligibility attaches to the individual sale, but proof does not. It comes from comparing shoppers who saw the AI's answer with shoppers who did not. "That difference only shows up across many shoppers. It cannot be seen in any single sale."

That comparison works most cleanly for paid AI placements, where exposure can be withheld from a control group. Organic answers are harder, because they are "dynamic, personalized, and generally cannot be withheld for a controlled group." Surveys, before-and-after trends and modeling can help, but such evidence is "generally directional rather than causal." The document puts AI visibility measures, among them presence, rank, citation frequency and share of prompt, in the same bracket. They can feed trend analysis and marketing mix modelling, but they are "directional indicators rather than standalone proof of a claim."

No claim means AI's involvement did not rise to either role. Existing credit under search, retail media or affiliate attribution is left as it was.

The eligibility table that follows sets several rules. More than one party can be eligible on the same conversion, but only one can hold the decision claim. In AI-Influenced sales, the AI surface and the parties whose content or data informed its answer are eligible for contributory claims. In AI-Recommended sales, the AI surface holds the decision claim, and parties whose "content, data, or paid inputs" informed the recommendation may hold contributory ones. An agent acting on an open instruction holds the decision claim on the same terms. An agent acting on a specific instruction holds nothing: "The execution is a service, not an influence."

Three steps that are often merged are kept apart. The pathway sets who is eligible to claim; evidence decides whether a claim is substantiated; and the parties evaluating it, "such as a brand, an agency, or an auditor," decide what credit follows. Where parties disagree about the pathway, the party asserting the claim carries the burden of proof. If the evidence shows AI influence but cannot separate AI-Influenced from AI-Recommended, it supports only a contributory claim.

Fixed shares are rejected outright. "There is no defensible basis for assigning fixed shares of credit across parties in journeys that leave partial or no trails, and inventing one would undermine the classification work the framework rests on." Nor does the framework measure value. Establishing the incremental economic value of a party's influence "requires experimentation, modeling, or other causal measurement."

Independence counts as well. Where the party making a claim also controls the recommendation, the transaction or the evidence, its claim should be distinguished from one backed by independently verifiable evidence. "As AI platforms take on more roles across the journey, independent verification becomes more important."

For a precedent, the IAB reaches back to the industry's definition of viewability. "Viewability standards did not measure attention; they defined what counts as a viewable impression and what evidence substantiates it," the document states. AI attribution is harder, it concedes, because much of the relevant evidence sits inside AI platforms and depends on what they choose to disclose.

Evidence available now

The blueprint calls the split between evidence that exists and evidence that does not "the central, practical finding of this framework." Five types of evidence can support claims today, though "no single item below proves an AI claim on its own":

  • Referral and citation data. UTM tags, referrers showing an AI domain, clicks on source links inside AI answers and affiliate links. Where they exist, "the AI path is demonstrably in the trail, and the evidence is deterministic."
  • Seller-side purchase records. Whoever runs the checkout holds a record of every sale. "That record proves the sale, not how the shopper arrived." It anchors the rest.
  • Agent transaction records. Agentic checkout standards produce a record in which the agent identifies itself, the merchant confirms the order and the payment carries the shopper's advance permission. That proves an agent completed the purchase, but it cannot answer the two questions that decide credit: what the human told the agent to do, and who picked the product.
  • Incrementality and holdout tests. Comparisons of shoppers who saw an AI answer with similar shoppers who did not are the standard evidence for AI-Influenced claims, because that pathway "plays the same role as awareness advertising." Television has always been judged the same way: "A TV campaign is never proven to have caused one sale. It is proven by markets that saw it outselling markets that did not."
  • Shopper-reported data. Post-purchase surveys and panels are directional rather than causal, but they are available now and cover journeys that leave no signal.

What cannot be proven yet

The second category, which the IAB labels "credit recognized but not yet measurable," covers influence that is real but cannot be substantiated, "either because the evidence exists and is not disclosed or because it is not captured at all."

The clearest case concerns publishers. A shopper asks an AI system for the best option in a category. The answer draws on a publisher's review. The shopper never clicks the citation and later buys at a retailer, where the sale reports as search or direct. Under the framework, the publisher is eligible for a contributory claim. Proving it is another matter. Content used in model training "is generally not traceable from outside the model." Grounding and citation are easier to observe, since publishers may see when their content is cited, but a citation does not show how much the content shaped the answer. "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."

Three other claims sit in the same category:

  1. Whether AI shaped the decision or only the intent, which often cannot be observed without platform disclosure or shopper-reported data.
  2. Whether a paid relationship influenced a recommendation. Platforms may disclose advertising, but "the basis for a recommendation inside an AI answer is not independently verifiable today." Commercial influence can also hide in cited content, such as sponsored listings or affiliate relationships "that the AI system cannot distinguish from editorial content."
  3. Conversions that leave no signal, which can be estimated in aggregate but not proven one sale at a time.

The categories are not fixed, according to the IAB. Evidence moves into the first group as platforms disclose more, checkout protocols spread and measurement methods mature. "Some gaps close through disclosure alone, because the evidence already exists inside AI platforms." Others need capabilities "the ecosystem has not yet built."

Two practical rules close the section. Transaction location "does not change the claim. It changes the evidence available to support it." On-platform, the handoff is trackable but the platform controls what is reported; off-platform, a claim usually rests on surviving referral data or, failing that, aggregate evidence. Commercial relationship does not change the claim either, but it changes what must be disclosed. A party asserting credit for a paid recommendation should declare the payment, because evaluators "cannot weigh it honestly without knowing whether money shaped the recommendation."

What the IAB asks of each party

The closing section reduces any evaluation to three questions: which pathway, what claim type and what evidence category. It then hands out work.

Brands are asked to measure or estimate the share of their conversions in each pathway, which the document says shows "whether the brand is increasingly influencing people, AI systems, or both." When a platform reports AI-driven conversions, brands should ask which pathway they came from, what credit is being claimed and what evidence supports it.

AI platforms, which "hold much of the evidence," are asked for four disclosures:

  1. Whether a product surfaced through paid placement or organic recommendation
  2. What sources informed the recommendation
  3. What the AI system surfaced, recommended or executed, and what influence is observed versus inferred
  4. Consistent, durable labeling of AI-originated traffic and referrals

Platforms reporting AI-driven conversions should also separate claims backed by observable evidence from claims "that rest on their own reporting."

Agencies can apply the same standard across platforms, with the key line drawn "between platform-reported credit and credit supported by independently verifiable evidence such as lift."

Publishers can identify the contributory claims they are eligible to make. "Citation reporting is a start. Connecting the sources that informed an AI recommendation to downstream outcomes is what would make those claims measurable."

David Cohen, the IAB's chief executive, presented the release as an invitation. "Attribution has never given us a complete view of how influence becomes an outcome, and AI adds new dimensions to the challenge," he said in the announcement. "The AI-Era Attribution Blueprint is honest about what the industry can't yet substantiate, and we're inviting everyone in the ecosystem to apply it and help make AI's role in commerce more observable, verifiable, and measurable." According to the IAB, it will bring brands, agencies, AI platforms, retailers and publishers together to share what they learn as more data becomes available.

Who wrote it, and who is missing

The IAB describes the working group as a cross-industry team of brand, agency, publisher, AI platform and measurement leaders. The blueprint names 16 people, Giegerich among them. WPP Media supplies four, the largest group, including Luke MacLean, its US total search lead, and Edward (Teddie) Cowell, its global vice president for generative and search optimisation. Microsoft supplies two: Ihab Rizk, a senior product manager at Microsoft Clarity, and Priya Bhasin, a principal product manager for AI. The IAB's Colburn is joined by Breeze Zhao of Walmart, Floriana Nicastro of Criteo, Ross Bucholc of Prudential Financial and Simon Poulton of Tinuiti.

The remaining four come from three startups: Brandlight, whose chief executive Imri Marcus is listed; Bluefish, represented by co-founder Jing Feng; and emberos, with founder and chief executive Justin Inman and partnerships head Dave Lewis. All three sell software for tracking or managing how brands appear in AI-generated answers, the category whose metrics the blueprint classes as directional rather than proof.

The list has gaps worth noting. None of OpenAI, Google, Amazon, Meta, Perplexity or Anthropic appears among the named contributors, even though the four disclosures fall mainly on companies of that kind and the document uses ChatGPT and Amazon's Rufus as examples. Microsoft, which runs Copilot, and Walmart, which runs Sparky, are the two contributors whose assistants the document names. No contributor on the printed list is identified with a news or content publisher, although publishers are one of the four groups the blueprint addresses and the publisher citation case is its clearest example of unmeasurable credit.

Why it matters for marketers

Paid placements already run inside AI answers, which gives the disclosure questions commercial weight. In August, Similarweb estimated that 26% of ChatGPT responses already carried a sponsored ad for Free and Go users. The same week, ChatGPT Ads added view-through conversions with a window fixed at one day, reported separately and kept out of the main conversions total, bidding and billing. Conversion figures of that sort fall into what the blueprint calls platform-reported evidence, which it asks agencies to keep apart from independently verified results.

Agent purchases raise a different problem. The agent transaction records the blueprint lists as evidence depend on checkout standards that are still taking shape. OpenAI introduced Instant Checkout in ChatGPT on September 29, 2025, built on the Agentic Commerce Protocol it developed with Stripe, and Google in January 2026 launched the Universal Commerce Protocol for agent shopping. Walmart, a working group member, tested ad formats inside its Sparky agent in fall 2025. The line between specific and open instructions requires a record of what the shopper told the agent and how the agent chose, information the blueprint says "emerging agentic commerce protocols may increasingly surface."

The fourth disclosure, consistent labeling of AI-originated traffic, touches a problem analytics vendors have tried to solve from the receiving end. Microsoft Clarity, whose Rizk sits on the working group, added AI Platform and Paid AI Platform channel groups in August 2025, splitting organic visits from AI chat tools from sessions that start with ads inside AI experiences. Tools of that kind depend on a referral signal surviving the trip. Similarweb's figures suggest that for most AI-influenced visits, none does.

The blueprint is also the latest in a run of IAB measurement documents. On October 20, the trade body opened a 15-day comment period on a draft Cross-Channel Measurement Decision Guide, developed under its Project Eidos program, on what to do when marketing mix models, incrementality tests and attribution disagree. The AI blueprint leans on the same tools, lift tests and modeling, for the claims that cannot be proven one sale at a time.

The survey figures in the announcement come from the IAB's September Outlook update, which PPC Land reported at the time found 45% of buyers naming the comparison of AI-driven and traditional customer journeys as a measurement challenge. That coverage tied a 44% figure to a different question, the share of buyers naming adaptation to AI-driven search as their top media investment challenge, and gave no number for buyers using branded search and direct traffic as proxies. The full study sits behind a registration wall on the IAB site.

What the blueprint leaves open

The document is a voluntary framework. It sets no numeric threshold for when evidence substantiates a claim, names no auditor or certification process, and gives no timetable for the platform disclosures it requests. The IAB's materials do not say whether any AI platform has agreed to provide them. Its stated aim is narrower: "to make the gap explicit: to establish how AI can influence a conversion, what evidence would be needed to account for that influence within attribution, and where the ecosystem cannot yet provide it."

The call to action is to apply the framework to live categories, find where evidence breaks down and build the transparency and measurement needed to close the gaps. How quickly that happens depends largely on companies that, by the blueprint's own account, already hold much of the missing evidence.

Timeline

Summary

Who: The Interactive Advertising Bureau, through its AI-Era Attribution Blueprint Working Group led by Caroline Giegerich, the IAB's vice president for AI. The 16 named contributors come from the IAB, WPP Media, Microsoft, Walmart, Criteo, Prudential Financial, Tinuiti, Brandlight, Bluefish and emberos. IAB chief executive David Cohen is quoted in the announcement. The framework addresses brands, agencies, AI platforms and publishers.

What: A 25-page framework that classifies each conversion by who or what created the intent, made the decision and executed the purchase. Every sale falls into one of three pathways: Human-Owned, AI-Assisted (split into AI-Influenced and AI-Recommended) or AI-Executed (split into specific and open instructions). The pathway determines eligibility for a decision claim, a contributory claim or no claim; execution alone earns no claim, and no percentages of credit are assigned. Evidence is divided into what is available today and credit that is recognized but not yet measurable, and AI platforms are asked for four disclosures.

When: The IAB released the blueprint on November 12, 2026. The document is marked as published in November 2026.

Where: The IAB is headquartered in New York, and the framework's examples and cited data concern the US market, including Adobe Analytics figures for US retail sites and Similarweb data on US desktop visits.

Why: AI assistants and agents increasingly shape or complete purchases without leaving the clicks and referrals attribution depends on, so sales they influence often appear as search or direct traffic. According to the IAB, that can send the next marketing investment to the wrong channel. The blueprint aims to give the industry a shared way to state AI's role in a sale and to make explicit which claims cannot yet be substantiated because the evidence is not captured or remains inside AI platforms.