727 FinanceWritten by Cobo
AI is shaking the profit logic of Google Search.
On October 29, shares of Alphabet, Google's parent company, plunged about 13%, wiping nearly $500 billion off its market value in a single day. What worries the market is that artificial intelligence (AI) is eroding the search advertising model that underpins Google's revenue. The latest earnings report shows that although Google's search query volume grew at its fastest pace in five years, advertising revenue rose only 12% year over year, an increase of just about $79 million from the previous quarter. Google management acknowledged that the AI mode is still less capable of monetization than traditional search.
This means search frequency is higher than ever, but Wall Street is beginning to doubt whether Google can convert that activity into real money.
Search advertising is extremely valuable because it can influence people's purchasing decisions. Brands pay for prominent placements in search results, users click, and Google profits from it. Once AI begins to intervene in search, price comparison, and even placing orders directly on users' behalf, users will no longer pass through those ad slots. Even if people maintain their search and consumption habits, Google may find it hard to earn the same returns as in the past.
But can't AI even complete an ordinary purchase?
Ironically, these AI agents that are capable of threatening the business models of internet giants sometimes cannot even successfully complete an ordinary purchase.
X (formerly Twitter) user Maia recently tested three AI shopping agents, and none of them completed the task successfully: Instinct was blocked by Sephora's bot detection mechanism; Poke could not complete an order on Uber Eats and could not even log in to a DoorDash account; Dash did complete checkout on AutoZone, but the order was canceled about 17 hours later. In the end, all three tools failed to complete the task and suggested that the user buy manually.
These three tools are a microcosm of the rapidly rising wave of "Personal Agents." As digital stand-ins built specifically for consumers, their ambition is not only to provide advice but also to directly act on users' will, taking over the entire process of logging in, selecting products, booking, and even paying on their behalf. From ambitious startup teams to tech giants such as Meta, everyone is rushing into this track. Shopping scenarios, which are high-frequency, easy to quantify, and directly involve real money, have naturally become an excellent touchstone for testing the real-world capabilities of these AI stand-ins.
Take Dash as an example: it uses one-time virtual cards to protect the user's real card number; it checks out as a "guest," so it does not need to obtain the user's account password; and it allows the shipping address to differ from the cardholder information. From a security perspective, these designs are very reasonable. However, in the eyes of traditional payment fraud prevention systems, this combination of features looks very much like credit card fraud. An unfamiliar device, a virtual card, and a shipping address that does not match the cardholder information—any one of these is enough to trigger an alert.
Dash's setback precisely reflects a fatal flaw in AI shopping today: although AI agents have learned to "run errands and place orders" for humans, the entire commercial world is not yet ready to verify their identities. From Cloudflare's bot defense line to the tightly guarded 3D Secure verification on the payment side, existing anti-fraud firewalls are certainly strong, but they are "blind." They simply cannot tell whether the entity on the other side is a malicious crawler with bad intentions or a compliant agent carrying real money to make purchases on its owner's behalf.
Of course, merchants are entirely justified in being cautious. If artificial intelligence (AI) books the wrong flight, enters the wrong address, or makes a purchase without proper authorization, who should bear the resulting refund, chargeback, and customer service costs? Because responsibility is unclear, accepting an unverifiable order may be riskier than simply blocking it. Even if an agent (AI agent) brings in a customer who is ready to place an order, many merchants will still consider blocking the order the safer choice.
Visa, Mastercard, and the Agentic Commerce Protocol jointly launched by OpenAI and Stripe are all currently working on how to identify these agents and authorize their payment behavior. Before these standards are widely adopted, merchants can only continue to use existing anti-fraud rules, even if that means potentially missing some legitimate orders.
Why doesn't Amazon welcome shopping agents?
However, security and liability issues explain only part of the resistance. For e-commerce giants such as Amazon, allowing external AI agents to shop freely may also affect the advertising business on which it depends for profit.
Previously, Amazon did not hesitate to take Perplexity to court, forcing its "shopping agent" tool Comet to completely withdraw from the Amazon platform, and successfully obtained a preliminary injunction; in September this year, Amazon also blocked Meta's shopping agent Muse. Although Amazon officially cited authorization, privacy, and security reasons, the challenges posed by external shopping agents are clearly not limited to these.
Advertising is one of Amazon's core revenue sources. Major brands bid for promotional placements in product search results, hoping consumers will see these ads while browsing, comparing, and filtering products. If artificial intelligence (AI) takes over all of this, and can even independently select products with better value for money, then consumers will simply not see these ads. Every transaction completed directly by bypassing the Amazon platform weakens Amazon's control over users' purchasing behavior.
Similar pressure is also appearing at Google. As more and more consumers begin to use AI search and shopping, traditional search advertising may no longer be an essential step in purchasing decisions. Through protocols such as AP2 and UCP, Google hopes users can complete product discovery, decision-making, and checkout directly within Google Search and Gemini. Although the monetization efficiency of search advertising may decline somewhat, as long as shopping behavior takes place on Google's "turf," it can retain users and have the opportunity to earn revenue from the transactions themselves.
Advertising giants begin betting on transaction business
Also reliant on advertising revenue, Google and Meta are experiencing vastly different AI shocks. The former faces pressure on search advertising monetization, while the latter has already begun to gain substantive returns from AI investment. Meta's advertising revenue in the third quarter grew 26% year over year, more than twice the growth rate of Google's search business. Chief Financial Officer Susan Li said that advertisers who adopted suggestions from Meta's AI business assistant had conversion rates nearly 5% higher than similar advertisers who did not adopt the suggestions, while cost per conversion remained basically flat, and these advertisers also increased their advertising spending by 9%. AI not only improved the effectiveness of Meta's advertising but also attracted more advertising budgets.
However, just as its advertising business continues to grow, this company, which derives about 98% of its revenue from advertising, has launched a shopping agent called Muse that contains no ads at all. About six weeks after launch, Muse's user count in the United States and Canada exceeded 10 million. More than a quarter of its weekly active users shop through it, with an average order value of more than $200. Zuckerberg said that in the future it may generate revenue through transaction commissions or subscription fees.
While using AI to increase advertising revenue, Meta is also looking for new ways to monetize the commercial value that advertising may lose. As consumers gradually become accustomed to letting AI search, compare prices, and place orders, platforms that control the transaction entry point may not only sell advertising to merchants but also take a cut directly from each transaction.
This change is beginning to transmit downstream to the payment segment, opening up new growth space for payment service providers such as Stripe.
Meta's Muse, Grok Bot, and Instinct have all integrated Stripe's Link agent wallet. Data released by Stripe at the end of September shows that over the past month, AI agent purchases completed through Link grew 38-fold. Because Stripe serves a vast merchant base, it can enable these shopping agents to make payments and assist merchants with payment collection processing. Consumers can shop through a variety of different agents, while the underlying payment process is always carried by the same company.As shopping entry points become increasingly fragmented, payment service providers like Stripe are expected to become the "default payment layer" connecting agent tools and merchants. No matter which agent tool ultimately wins, every transaction must go through identity verification, payment authorization, and settlement, and each of these steps carries commercial value.
When shopping entry points shift, who gets to make money from transactions?
Google wants to keep shopping within Search and Gemini, Meta has begun exploring transaction revenue beyond advertising, and Stripe is providing payment services for different shopping agents. As AI gradually participates in consumers' search, price comparison, and purchase decisions, revenue originally captured by search engines and e-commerce platforms may also begin to flow to new players.
For years, search engines and e-commerce platforms have controlled consumers' "entry points," converting user attention into advertising revenue. Brands pay for impressions and clicks, and platforms earn revenue before consumers place an order. In a shopping process dominated by "intelligent agents," consumers may only need to state their needs, and the agent handles the rest of the searching, filtering, and ordering. The "discovery" and "recommendation" functions once performed by advertising are gradually shifting to these intelligent agents.
This change will give rise to new advertising formats. Brands still need to compete for consumers' choices, and intelligent agents may also establish their own "paid promotion" mechanisms or charge recommendation fees. However, as purchase decisions shift to AI interfaces, at least some of the value originally captured by search results pages and product listings will flow to the links that can guide decisions, facilitate transactions, or process payments.
Looking back at the divergence between search volume and advertising revenue growth in Google's financial reports, for a search giant built on advertising, nothing is more unsettling than this: users are still searching and spending, but the money is beginning to flow elsewhere.