Do AI shopping agents destroy search ad revenue while leaving influencer marketing intact?

Emil Rustamli argues that as AI agents take over shopping — Perplexity's shopping feature, ChatGPT's shopping mode, Mastercard's "Agent Pay," Visa's "Intelligent Commerce" partnership with OpenAI, Perplexity, Stripe, and Microsoft — they compress "the discovery-to-decision process," cutting into the ad revenue tied to search result pages and product listings. He cites Klarna CEO Sebastian Siemiatkowski predicting shoppers will "stop visiting retailer websites altogether" and instead just "send" their AI to buy. Rustamli calls the resulting pattern, following a Bain & Company brief, "zero-click" search: the user gets an answer without the clicks that generate ad impressions.

The mechanism he gives is twofold. First, an unnamed analysis found that when an AI answer topped Google results, paid-search click-through fell "from ~21% to under 10%," with organic results seeing "a ~70% drop in clicks" — concerning because Google search ads brought in "$175 billion" in 2023, "more than half of Alphabet's total revenue." He extends the same logic to Amazon, arguing that an AI assistant querying Amazon's database and placing an order directly turns the retailer into "a transaction back-end rather than a digital mall with aisles of ads," even though Amazon keeps the sale. Second, he argues this erosion has a boundary: "AI agents excel at fulfilling needs, but they're trash at creating wants." A need ("best bagless vacuum under $300") is a solved query; a want (the lip gloss a TikToker swears by, the home decor piece seen on Instagram) is "created through discovery, vibes, and social influence" — territory he says AI "simply can't compete" in against "the dopamine hit of scrolling." He points to influencer marketing budgets "projected to hit $9.3 billion in 2025 (up 14% from 2024)" as evidence that want-creation value is migrating toward, not away from, human social platforms.

This gives the needs/wants distinction as the limiting condition on Why do consumers delay adopting agentic search?: that note's formal model treats agentic search adoption as a platform-revenue-ranking question across the board, while Rustamli's argument implies the ranking differs by query type — need-driven queries are where agents substitute for ad-funded discovery, want-driven queries are where they can't. It also sits alongside Will brands succeed by building their own AI shopping agents?: Accenture frames the threat as brands needing to win agent-mediated decisions, but Rustamli's account suggests that threat is confined to need-fulfillment purchases, leaving want-creation as a channel brands can still own directly through social and influencer content. And where Can an AI agent serve both merchant and user interests fairly? identifies a loyalty conflict inside the agent's own business model, Rustamli's piece describes the adjacent structural question of where ad dollars go once the agent, loyal or not, closes the need-based transaction. The zero-click mechanism he describes for shopping search parallels what Will AI Overviews reduce search referral traffic to publishers? documents for news publishers — the same AI-answer-suppresses-click pattern, applied to a different ad-revenue-dependent industry.

The piece offers no sourced study for its own central numbers — the Google CTR figures are attributed only to "one analysis," unnamed, and the needs/wants claim is Rustamli's own framing rather than a tested finding. It is also written from mid-2025, before OpenAI's and Visa's later protocol announcements and before any agentic-commerce ad model shipped, so its prediction that "social media and influencer marketing might actually become more valuable" is a forecast, not an observed shift. The implication it supports is narrower than AI kills ad revenue: the erosion is specific to discovery-funnel advertising around pre-formed needs, and the piece's own logic predicts a compensating migration of ad spend toward want-creation channels rather than a net decline in advertising.