Similarweb: Over 90% of AI-Driven Sales Get Miscredited

Over 90% of AI-driven sales are miscredited as branded search, direct traffic or retail media, Similarweb says. The IAB's new attribution blueprint aims to change that.
By Marcus Bennett
4 min readAt a glance
- More than 90% of AI-driven sales are miscredited as branded search, direct traffic or retail media, per Similarweb.
- The IAB's AI-Era Attribution Blueprint defines three conversion types: human-owned, AI-assisted and AI-executed.
- Anthropic's new Claude usage policies took effect Friday, banning influence operations and undisclosed sponsored content.
- Google is testing moving citations to the bottom of AI Overviews, likely cutting referral traffic further.
- Jensen Huang resigned as Nvidia CEO; CFO Colette Kress becomes interim CEO.
More than 90% of AI-driven sales are currently miscredited as branded search, direct traffic or retail media, according to Similarweb — the starkest number yet in an industry still fumbling how to attribute purchases made through AI agents. The IAB's answer, unveiled as the AI-Era Attribution Blueprint, splits AI-attributable conversions into three types: human-owned, AI-assisted and AI-executed.
A human-owned conversion means AI didn't substantively create the purchase intent, influence the product decision or carry out the purchase — though an AI assistant may still have been involved in the decision-making process. AI-assisted conversions cover cases where the AI narrows down and suggests options. The IAB proposes two sub-types there: "AI-influenced," where the AI "shapes intent" but a human makes the final choice, and "AI-recommended," where the AI "shapes the choice itself." AI-executed conversions mean a human gave an agent permission to buy — again with two variants, one where the human specified what to buy and one where the AI made the selection.
What does the framework change for measurement?
If that taxonomy sounds fuzzy next to how programmatic measurement actually works, you're not wrong. But the IAB's goal at this stage is descriptive, not operational. Caroline Giegerich, the IAB's VP of AI, said during a webinar on agentic advertising on Monday: "We define this based on what's actually happening, not what we can measure, per se, today." For practitioners, the near-term implication is honest: agentic search marketing (AI agents performing searches and purchases on users' behalf) is live, but the plumbing to credit it correctly isn't.
'Premium' TV loses its meaning at the IAB's CTV event
At the IAB's "I Want My CTV" event in New York City, panelists took aim at the word "premium." FreeWheel product chief David Dworin said it could mean any professionally produced content with engaged audiences — but the classification matters less than finding content, or more accurately audiences, that fulfill the buyer's performance goals.
Roku Product Innovation Leader Curt Larson pushed the framing further: every marketer is a performance marketer, even those who focus on TV. "You're not spending money for the sake of spending money," Larson said. "You're spending money to create some sort of outcome. And when that outcome is far in the future or very difficult to measure, you call yourself a brand marketer."
In practice, buyers still balance attributable performance against scale and cost, which shapes what premium inventory (high-quality, high-visibility ad placements) is actually feasible. "You need premium, first-impression, low-session, very visible placements," said PMG Head of Video Mike Treon. "But you also need that efficiency to really balance it."
Anthropic's new Claude usage rules take effect
A sweeping set of Claude usage policies went into effect Friday, a month after Anthropic first announced them. One section forbids users from being "abusive or cruel toward our models" — a requirement widely panned online.
The consequential parts target manipulation. Anthropic now promises to remove state or commercial media outlets that use Claude to fabricate social accounts and news sites or run other influence operations. Using Claude to amplify a message while obscuring who's behind it is banned, as is misrepresenting sponsored content as organic and gaming AI search engines. The stated focus is cracking down on political manipulation — though striking at AI slop content sites would also help Google's search results.
In brief
- Google is testing moving citations to the very bottom of AI Overviews, which should further reduce website referral traffic. [Search Engine Roundtable]
- Mars faces a conundrum: what happens to cash register impulse buys when AI agents do the grocery shopping? [Digiday]
- OpenAI and Anthropic have big marketing budgets and want to sponsor podcasts, but listeners tend to revolt when they hear ads from these companies. [Bloomberg]
- The Washington Post's executive editor Matt Murray says the paper, after losing "hundreds and hundreds of millions of dollars" in recent years, is on the cusp of breaking even next year. [Adweek]
- Jensen Huang announced his resignation as Nvidia CEO; CFO Colette Kress will serve as interim CEO until a successor is appointed. [WSJ]
For marketers, the through-line this week is attribution: as agents quietly absorb search and commerce, the measurement gaps — and the policy crackdowns around synthetic influence — will only get harder to ignore.
Original: seroundtable.com
Marcus Bennett
Show full bio
Market editor covering money, funding and market shifts around Social platform news and policy at Reach Brief.
Related articles
ChatGPT Ads Convert at 0.37% — Six Times Worse Than Organic Answers
IAB Admits AI Attribution Cannot Be Standardized: Publishers Lack Data
Amazon Ads Agent Puts AI in Charge of Campaign Buying
B2B Buying Decisions Are Moving Where Ad Tech Can't See
CreatorIQ Rolls Out 'Q,' an Agentic AI Built for Creator Discovery and Measurement