Are AI agents making attention-based ads irrelevant for aggregators?

Rihard Jarc, writing on UncoverAlpha, argues that as AI shopping and travel agents take over the comparison step of a purchase, they strip aggregators of the one thing advertisers pay for: human attention. He writes that "the agent keeps the demand signal, and the aggregator gets reduced to a fulfillment API," and that because OpenAI's ChatGPT shopping and Google's AI Mode hotel booking rank merchants on stated criteria (availability, price, quality, primary-seller status) rather than paid placement, "when the buyer is a piece of software optimizing for your stated preferences, the ad becomes irrelevant." He points to Tripadvisor as the live case: its Hotels segment (the former metasearch business) saw revenue fall 23% to $117.9 million in Q2 2026, with media and advertising down 12%, and management guided the whole Hotels & Other segment down another 20-23% in Q3, "citing structural changes in its main SEO channel."

The mechanism he gives is architectural, not just competitive: across Muse, ChatGPT, and Google's AI Mode, "the merchant stays the merchant of record, but the agent owns the conversation" — the inspiration, the comparison, and the "why this hotel and not that one" reasoning all stay inside the agent's interface. He notes DoorDash, Uber Eats, Instacart, and Amazon built close-to-100%-margin ad businesses worth over $4 billion annualized on exactly that human-attention layer (Amazon's on-site search ads alone were roughly $60 billion of its 2025 ad revenue), and argues that layer has no equivalent once a software agent, not a scrolling human, is doing the choosing. He treats Booking Holdings' rising marketing-expense ratio (4.7% of gross bookings in Q2 2026 vs. 4.6% a year earlier, growing faster than gross bookings) as the first-order symptom — free SEO traffic turning into paid traffic — with agent-mediated booking as the second-order shift sending that paid traffic "to a new toll collector."

This sits alongside Is non-human traffic really half of all internet use?, which documents the same traffic-diversion pattern from the publisher side; Jarc's account supplies the revenue mechanism — ad irrelevance — that explains why publishers are moving to licensing instead of waiting for clicks to return. It extends Will brands succeed by building their own AI shopping agents?: where Accenture frames the risk at the brand level, Jarc applies the same decision-layer logic to aggregators and metasearch sites, with Tripadvisor as the earliest casualty. And it complicates Why do consumers delay adopting agentic search? — that model's adoption lag assumes platforms benefit from slow-walking agentic search, but Jarc's examples describe platforms already shipping agent-ranked comparison at the direct expense of their own and partners' ad revenue.

The piece is a paid Substack analysis, not a financial study: its company figures are drawn from public filings and earnings calls but assembled to support Jarc's own thesis, and it offers no accounting of how much of any single company's decline is attributable to agents specifically versus other causes — Tripadvisor cites "structural changes in its main SEO channel," a shift that predates most agent commerce. The excerpt does not establish that ad-funded aggregators cannot find a new toll-collection point inside the agent flow itself, only that their current attention-based model is shrinking. If the pattern generalizes beyond travel metasearch, it implies the sharpest near-term risk falls on businesses whose revenue depends on a human, not an agent, doing the comparing.