The central risk in Alphabet's AI search transition is not a collapse in query volume; it is an attribution gap. Google is answering more questions itself, paying a small pilot to publishers for the content behind those answers, and expanding AI Overviews and agents while the click-based evidence remains mixed and the ad-tech exchange is being repriced by courts. The history of advertising is a history of unmeasured waste, and Alphabet is now running the largest unmeasured experiment in search monetization since the ad auction itself.
The publisher pilot is tiny, variable, and legally ambiguous
Three independent accounts report that Google is paying roughly 100 publishers for content used in AI Overviews 31,57,58, a scale described elsewhere as a pilot paying around 100 publishers for AI search results 26,27 and as Google paying a small number of publishers for AI-generated overviews 38. The sums are not a revenue replacement. Some payments are under $1,000 27, and one Ars Technica article titles the program 'Google AI Pilot Pays Some Sites Under 0.1% Of Ad Revenue' 32,33, with many sites receiving one-tenth of one percent of advertising revenue 34. The same reporting says the initiative is struggling 34,67, while a Digiday headline frames it as a 'pay-per-value AI licensing program' 35. The post notes this could also be legal maneuvering, but no evidence about Google's motives is supplied 35.
That ambiguity matters because the product change is real. Google increasingly answers questions with an AI-generated summary at the top instead of sending users elsewhere 78, and AI Overviews are described as siphoning traffic from websites 24. The source attributes the broader shift to users getting answers inside Google AI Overviews, ChatGPT, Claude, and Perplexity instead of clicking through 72. Publishers report declining traffic and strategy changes 52, but Reuters cannot determine how much of the decline in organic-search referrals was caused by AI rather than other changes 73. That is precisely a measurement failure: the parties harmed cannot isolate the effect of the new interface, and Google is paying sums too small to settle the value exchange.
The click-based evidence is mixed and recent. During Q2 2026, desktop click-through rates for top organic results fell while mobile rates rose 43, according to Advanced Web Ranking 43. AI Overviews are now shown for many brand-name searches, with tests claiming 93% of major brands received an Overview for a branded search 36. Ads appear above AI Overviews 17, and one- to two-word searches now drive most AI Overview ad views 23. At the same time, ChatGPT advertisers gained 65% as AI Overview queries shrank 23. Google Gemini has added UTM parameters to provide site owners clearer attribution for traffic from Google's AI platform 30, but that is a measurement patch, not a proof of incrementality.
The shopping surface shows the same pattern of scale claims against thinner product evidence. Google's Heiko Hotz pitches 60 billion product listings as fuel for shopping agents 41,42, yet Productrise found AI Mode showed 95% fewer products than standard search in July 41,42, framed as a potential gap for retailers 42. AI Mode monitoring is now rolling out globally 49 after being limited to Pro and Ultra users 51, and YouTube's AI answer tool has grown more than 500% since December 25, but nobody disclosed how usage was counted or which screens were included 25.
Europe makes the attribution problem a product problem
Google has moved from negotiating over AI answers to litigating the rules of access. It is fighting EU orders to share search data and open Android to rival AI assistants 37,45,46,48, with the dispute concerning access to search infrastructure by AI chatbot providers 53. Google warns of the biggest reduction in search quality and privacy risks in Europe 47. The product changes are already visible: Google removed live prices and trip dates from EU Search 39,40, and trip-date search is no longer available in the EEA according to Google's search chief 39. The company attributes the removals to Brussels 40. A Washington Post opinion piece argues the EU has made Google less useful for Europeans 60, while posts citing the Initiative for Neutral Search note Google's own Shopping box and Travel carousel remain unchanged after September 21 44.
Separately, Google was fined 403 million euros for violating users' location-data privacy 56, equivalent to about 9-10 hours of revenue 28, adding legal and financial exposure 29. A lawyer argues the EDPB letter on Search data conflicts with its own guidance 54,55, while Google argues the required data is insufficiently anonymised 50. The European fight is not only about legal principle; it is about whether the AI Search product remains useful and measurable enough to keep ad utility intact.
The core advertising machine still funds the AI build, but concentration is the risk
Despite the AI noise, the advertising base remains the most corroborated fact in the material. Most of Alphabet's revenue comes from search advertising 93, with online advertising more than 80% of revenue in 2020 18 and more than 75% of total revenue in FY2024 14. Full-year 2025 advertising revenue exceeded $294 billion 88. Search itself is still growing 84,90, with core search growing faster than it has in over a decade excluding the COVID period 65. The most recent hard read is Q2 Search & other revenue of $63.3 billion, up 17% 69,76,92. YouTube adds a second pillar: Q2 YouTube advertising reached $11.1 billion, up 13% 3,7,10,11,12,13,16,77,86,87,90, and combined YouTube advertising and subscription revenue exceeded $60 billion in 2025 1,2,10,14. That cash flow funds the AI build-out, with advertising profits supporting infrastructure investment 75,80 and search and advertising generating more than $100 billion in annual cash flow 15.
But the base is narrower than it appears. Fewer than 25% of Google searches carried ads 17, leaving the engine dependent on intent-rich queries. The AI transition is concentrating monetization on short queries: one- to two-word searches now drive most AI Overview ad views 23, and an Ahrefs study associated AI Overviews with a 58% reduction in click-through rate to top-ranking search results 66. If those short answers satisfy intent without a click, the ad-bearing query set shrinks. Google's UTM parameters 30 and AI Max/Performance Max reporting a 15% conversion lift at similar return on ad spend 87 are attempts to prove the new surface, but they do not yet answer whether AI answers incrementally grow total search revenue or shift spend within it.
Meta provides the counterfactual with cleaner numbers. Meta's AI ranking models increased Facebook ad clicks by 8.3% 82,94 and conversions by 15.7% 82, while impressions rose 14% 8,9,68,84,85 and price per ad rose 12% 4,5,6,61,68,84,85. That is the standard Alphabet needs to match: measured lift at scale, not just engagement. Google's own tests show Hellmann's gained 24% unique reach and cost per individual viewer fell 25% in early Unilever/WPP testing 22, and Sponsored Prompt clicks convert 48% more often 91. Those are promising, but they remain vendor-supplied case studies. The post argues branded search advertising can be incremental 81, yet the exchange itself is being repriced through U.S. ad-tech litigation: First Look could win impressions even when rival exchanges offered higher bids 19, Dynamic Revenue Share adjusted AdX's take rate based on competition 19, and an internal memo stated the Bernanke algorithm increased AdX's win rate from 37% to 44% 19.
The deeper issue is that standard ad measurement may be substantially inaccurate. Across 663 experiments, machine-learning estimates of upper-funnel advertising lift had a median of 83%, while randomized experiments measured only 29% 20. Facebook's own tests indicate standard ad measurement is often off by a factor of 3x 21. If the industry baseline is that unreliable, then Alphabet's AI search transition is taking place on a foundation that cannot yet support its claims. Apple privacy changes have also affected ad-targeting effectiveness for Meta 63,83,89, which raises the same measurement problem for anyone relying on cross-app attribution.
The moat question has moved from query share to the agent starting point
The strategic debate is whether assistants and agents defend or dilute the moat. Competition between assistants and traditional search is increasing 79, with Meta's Muse suggested as a potential threat to Search as the primary gateway 71 and downloads cited as a reason to watch diversion 70. AI could disrupt Search itself 74, and a bearish thread argues AI agents could disrupt the search advertising model 64. The bull case is that if agents become the starting point for discovery, leverage could shift toward Google at the operating-system level 59. That is unresolved prediction, not measured effect.
The material collectively points to Alphabet trying to convert a traffic-liability narrative into a licensing and agent-commerce narrative before regulation hardens the rules of access. The publisher pilot 31,57,58 is too small to offset traffic siphoning 24, and the European feature removals 39 show the product cost of losing data-control fights. The market is not treating search as dead, with one Reddit discussion titled 'Alphabet price target raised to 450 - yes, search is not dead' 62, but the moat question has moved from query share to who controls the agent starting point and whether monetization follows.
The history of advertising is a history of unmeasured waste. Alphabet's AI search bet will succeed only if its new surfaces show incrementality, not just impressions, clicks, or branded Overview coverage, but net new advertiser value that survives court scrutiny and EU feature removals. Until then, the question is not whether AI Overviews work, but how Alphabet knows they work.