Rory Sutherland, vice chairman of Ogilvy UK, this week issued a pointed warning about the trajectory of advertising-funded artificial intelligence, arguing that financial incentives will systematically corrupt AI responses the same way they degraded Google Search over the past 15 years - a prediction that lands as the industry's two largest AI platforms diverge sharply on whether to accept advertiser money.

The remarks came during the MAD//Masters May 2026 Livestream, a monthly online session tied to MAD//Masters By Rory Sutherland, a 12-week hybrid professional development course run in partnership with MAD//Fest London. The video was published on June 11, 2026 on the MAD//Masters By Rory Sutherland YouTube channel, and had received 5,870 views by the time the excerpt was clipped and shared. Sutherland, who joined Ogilvy as a graduate trainee in 1988, was responding to a question from a B2B sales professional named Arthur about the branding and leadership differences between ChatGPT and Claude.

Two platforms, two business models

The contrast Sutherland drew was precise. According to him, ChatGPT is "pro-ads, pro-consumer but also bland branding-wise and chaotic leadership-wise," while Claude is "anti-ads, pro-enterprise and slightly fear-mongering but also warm branding-wise and stable in terms of leadership." That framing captures a real and widening strategic divide.

OpenAI confirmed plans to test advertising within ChatGPT on January 16, 2026, targeting logged-in adults in the United States across its free and Go subscription tiers. The company set a baseline CPM of approximately $60 and an initial minimum advertiser commitment reported at $200,000. By May 5, 2026, OpenAI had dropped that minimum entirely and opened a self-serve Ads Manager to all US businesses, introducing cost-per-click bidding at a recommended $3 to $5 per click. The pilot had crossed $100 million in annualised revenue within six weeks of launch.

Anthropic, by contrast, announced on February 4, 2026 that Claude would remain advertisement-free across all subscription tiers, publishing detailed principles establishing that sponsored content would never influence Claude's responses. The company then ran Super Bowl commercials in February explicitly mocking the prospect of ads appearing inside AI conversations, with a creative execution depicting an AI assistant interrupting useful advice to promote fictional products. Claude's traffic share has since tripled, with ChatGPT's worldwide generative AI website traffic share falling from 76.4% in June 2025 to 52.7% by late May 2026, according to Similarweb data published on June 11, 2026.

The Google precedent

Sutherland's concern is not with advertising at launch. He acknowledged that Google's early model was, in his own words, "totally unobjectionable." The platform gave users the results they wanted and placed clearly labelled ads alongside. The problem, he argued, was what he called "financial gravity" - a term he attributed to Eric Ries - which he described as the cumulative pressure of growth targets and profit expectations pulling the platform away from its original purpose.

"In 2010, when you search for a hotel in Google," Sutherland said during the livestream, "the top search result was the hotel with its phone number and address and stuff like this. Now the actual hotel itself appears at about number seven and there's a load of extraneous [material] above it from people who've just outbid the hotel."

The mechanism he described is structural. A competing hotel or online travel agent will always outbid the property itself for placement against a search for that property's own name, because the intermediary's margin per booking - capturing the customer from a direct relationship - exceeds what the hotel can afford to spend. "Irrelevant, sorry, tangential players will always outbid deeply relevant players," Sutherland said. "And incidentally, dishonest actors will always outbid honest actors because the dishonest actors are by definition more profitable than the honest ones."

That dynamic is traceable in Google's own revenue data. Google Network revenue - the portion flowing to external publishers through AdSense, AdMob, and Ad Manager - fell 1% to $7.4 billion in the second quarter of 2025, even as total advertising revenue grew. The divergence reflects what PPC Land has reported as a structural shift: traffic and monetisation are concentrating inside Google's own properties, with third-party publishers receiving less of both. Google AI Overviews now serve more than two billion monthly users, generating answers that reduce the need to click through to source websites.

The protection racket argument

Sutherland took the analysis further, drawing a distinction between the advertising model of an earlier era - paying a media owner to become more famous - and what dominant platforms now offer. "What Amazon can say and what Facebook can say now is the opposite of that," he said. "It's unless you give us some money, we'll make you disappear. That's not Mad Men. That's actually The Sopranos."

The specific mechanism he outlined is one in which platform scale transforms advertising from an optional amplification tool into an existential requirement. A business that does not buy ads on a dominant platform does not merely forgo growth - it risks being effectively invisible. Sutherland described advising clients including Laurel and Ogilvy to invest in influencer marketing as a contingency. "You may not like it," he said. "It may not work as well as your Facebook stuff, but you've got to have a plan B."

The warning is grounded in a concrete anecdote. Sutherland described an acquaintance whose entire business ran on LinkedIn. The account was banned without explanation. It was only through a personal contact at LinkedIn that the account was eventually reinstated - a process that took five days and, according to Sutherland, left the business owner genuinely afraid for its survival. Hotels.com, he noted, reportedly conducted a SWOT analysis that identified Google's ability to destroy its business at will as the single largest threat to the company.

The protection racket framing is not hypothetical in the current ad-tech environment. PPC Land has tracked website operators reporting traffic declines ranging from 70% to 85% following Google algorithm updates during late 2025, with some experiencing a 98% drop in Google Discover impressions. Those figures represent direct monetisation consequences for publishers who depend on platform distribution, not merely audience inconvenience. Google's own AI search changes have been documented decimating individual publisher revenues, with one example - Charleston Crafted, a do-it-yourself home improvement site - reporting a 70% traffic decline and a 65% drop in display advertising revenue after Google introduced AI features.

What this means for AI advertising

Sutherland's argument implies that the structural pressures on ad-funded AI will not be visible at first. The opening phases will look like clearly labelled, relevant advertising that causes little friction. The distortion arrives gradually, driven by what he called the two "enormous temptations" facing companies that have spent what he characterised as "an insane amount of money without any clear idea of how it's going to be paid for."

The first is selling AI purely as a cost-reduction tool - replacing headcount rather than creating new capability. The second is the gradual introduction of paid placement into AI responses. "The first form of advertising," Sutherland said, "is yes, don't worry mate, when anybody asks this question on AI we'll make sure you get a mention - that's pretty dubious, okay, but it gets worse."

The "gets worse" scenario is the protection racket endpoint: a dominant AI platform where visibility without payment becomes impossible. OpenAI itself has committed that "ads will not influence the answers ChatGPT gives you," a principle its CEO Sam Altman and Applications CEO Fidji Simo both reiterated publicly on January 16, 2026. But Sutherland's point is that the pressure to maintain that separation compounds over time as revenue expectations grow.

The IAB published research in January 2026 - covering a survey of 505 US Gen Z and Millennial consumers alongside 104 advertising industry executives conducted between October 2025 and January 2026 - that documented a 19-point decline in Gen Z trust toward AI advertising. Gen Z consumers were nearly twice as likely as Millennials to feel negatively toward AI ads, at 39% versus 20%. Twenty percent of consumers described brands using AI as "manipulative," compared to 10% of advertising executives, suggesting a significant gap between industry perception and public sentiment.

The B2B shift and enterprise positioning

Sutherland's read of Anthropic's strategic direction is that the company is "basically making itself a B2B company." He argued that while consumers use Claude and are "totally happy," Anthropic's primary objective is serving corporate clients. The observation aligns with Anthropic's product development trajectory - the company introduced an enterprise integration layer, expanded its API capabilities, and positioned Claude Code as a developer and engineering tool throughout 2025 and into 2026.

For the marketing community, the B2B versus consumer distinction carries practical consequences. An AI tool used internally by professional services firms or financial institutions for research and decision-support operates under different incentive structures than one generating consumer traffic that can be monetised through advertising. Sutherland flagged this explicitly, noting that Goldman Sachs allowing its employees to use Google hundreds of times daily without concern about third-party influence on that platform's results represents a genuine, unaddressed business risk. "Be very easy to go to Google and say, when anybody from Goldman Sachs searches for a law firm, make us come up first," he said.

Subscription-based models, he suggested, align the tool's incentives with the user's needs rather than advertisers' goals. He cited Kagi, a subscription search engine, as a personal example of a platform he uses precisely because it is designed to surface what the user wants to know rather than what advertisers want to promote.

Claude's 130% surge in US desktop unique visitors between February and March 2026 - reaching 2.66 million desktop unique visitors according to Comscore data published on May 18, 2026 - suggests the enterprise-and-subscription positioning is generating measurable traction. The broader AI chatbot category reached 44.4 million combined US desktop unique visitors in March 2026, a 21.3% month-over-month increase, indicating continued category-level growth rather than redistribution of a fixed audience pool.

Behavioural science and the existing structure problem

The second half of Sutherland's MAD//Masters session - responding to a question from a South Africa-based innovation consultant named Angus - addressed how to apply behavioural science frameworks within existing organisations rather than at a clean-sheet starting point.

His central example was a campaign for American Express from the late 1980s and early 1990s. According to Sutherland, a client named Paul - whose surname he could not recall - identified that a significant proportion of people who wanted an American Express Gold card were not applying because of fear of rejection, not lack of desire. The card cost 85 pounds per year at the time, which Sutherland described as "quite expensive," but most people in the upper quartile of earnings would have wanted one.

The intervention was mechanical, not persuasive. A colleague, Steve Harrison, reframed the simplified application - which no longer required filling in a full A4 form with salary and bank details - from "applying to American Express" to "American Express wants you to apply." According to Sutherland, that framing change resolved a psychological bottleneck. "They didn't feel they'd kind of sat too high up the table and been asked to move down," he said. "They didn't feel rejected because you'd told them to apply."

He described the framework he uses for this type of problem as COM-B: Capability, Opportunity, Motivation. In the American Express case, the capability and opportunity to apply existed. The bottleneck was motivational - specifically, a psychological barrier that rational persuasion could not address because it was not a knowledge problem. The implication for AI-driven marketing is direct. Optimisation tools that work from declared intent or conversion data - the type of measurement infrastructure that OpenAI's ChatGPT Ads Manager now supports, including a Conversions API and pixel-based measurement that went live on May 5, 2026 - will systematically miss the segment of the market that wants to act but does not.

Why this matters for the marketing industry

The video excerpt was published on June 11, 2026, the day before the current date, making the timing resonant. The advertising industry is actively navigating the consequences of decisions that are still being made: whether to build AI on advertising revenue, whether to integrate ads into AI-generated responses, and how to interpret the performance data that will emerge from these early experiments.

Google's AI Mode already has ads blending into answers seen by more than one billion users, as PPC Land has reported. Sponsored store listings were observed inside AI Mode product panels as early as March 25, 2026. The company formally announced shopping ads for AI Mode on February 11, 2026, when the surface already had more than 75 million daily active users. The speed of that rollout - from experimental to billion-user reach in a matter of weeks - is exactly the kind of trajectory that makes Sutherland's concern about financial gravity relevant now rather than hypothetical.

The MAD//Masters course that produced this discussion is available through the MAD//Fest London platform at https://www.madfestlondon.com/masters/ and is described as a 12-week hybrid programme covering behavioural science, creativity, and strategic problem-solving for marketing professionals.

Timeline

  • Late 1980s - early 1990s: Rory Sutherland works on American Express account at Ogilvy, developing what becomes a case study in psychological bottleneck removal applied to the Gold card application process.
  • 2010: Sutherland cites as the year Google search results for a hotel query began returning the property itself at approximately position seven, with competing advertisers filling the top positions.
  • November 2022: OpenAI launches ChatGPT, establishing the consumer AI chatbot category.
  • December 2, 2025: OpenAI issues an internal "code red" directive pausing advertising development to focus on ChatGPT core improvements.
  • January 16, 2026: OpenAI formally confirms plans to test advertising in ChatGPT's free and Go tiers in the United States, at a reported $60 CPM and a $200,000 minimum advertiser commitment.
  • January 2026: IAB publishes research documenting a 19-point drop in Gen Z trust toward AI advertising, covering a survey of 505 consumers and 104 advertising executives.
  • February 4, 2026: Anthropic announces Claude will remain ad-free across all subscription tiers and launches Super Bowl commercials targeting OpenAI's advertising strategy.
  • February 8, 2026: Anthropic's Super Bowl commercials air during Super Bowl LX at Levi's Stadium, Santa Clara.
  • February 11, 2026: Google formally announces shopping ads for AI Mode, a surface already used by more than 75 million daily active users.
  • February-March 2026: Claude records a 130% month-over-month increase in US desktop unique visitors, reaching 2.66 million according to Comscore data.
  • March 26, 2026: OpenAI discloses the ChatGPT ad pilot has crossed $100 million in annualised revenue within six weeks of launch.
  • May 5, 2026: OpenAI opens its self-serve Ads Manager to all US businesses, drops the minimum spend requirement, and launches Conversions API and pixel-based measurement tools.
  • May 26, 2026: Similarweb measurement date showing ChatGPT's worldwide generative AI traffic share at 52.7%, down from 76.4% twelve months prior.
  • June 11, 2026: MAD//Masters By Rory Sutherland publishes the YouTube excerpt of Sutherland's May 2026 livestream discussion covering AI business models, advertising incentives, and the Google precedent.
  • June 13, 2026: ChatGPT's traffic share falls to 52.7% as Claude triples its AI traffic share, according to Similarweb data published June 11, 2026.

Summary

Who: Rory Sutherland, vice chairman of Ogilvy UK, advertising executive, author of "Alchemy: The Power of Ideas That Don't Make Sense," and the instructor behind the MAD//Masters course run in partnership with MAD//Fest London.

What: Sutherland delivered a warning during the MAD//Masters May 2026 Livestream that advertising-funded AI models will follow the same degradation path as Google Search - where financial pressure causes relevant results to be displaced by the highest-paying advertisers, ultimately transforming a useful tool into what he characterised as a protection racket.

When: The livestream took place in May 2026. The video excerpt was published on the MAD//Masters By Rory Sutherland YouTube channel on June 11, 2026.

Where: The remarks were made during an online livestream associated with the MAD//Masters By Rory Sutherland course, a collaboration between Sutherland and MAD//Fest London. The full course is accessible at https://www.madfestlondon.com/masters/.

Why: The discussion is directly relevant to a live commercial decision the AI industry is currently navigating. OpenAI has introduced advertising into ChatGPT, beginning January 2026, while Anthropic has publicly committed to keeping Claude ad-free. Sutherland's argument - rooted in 15 years of observable decline in Google's search utility under advertising pressure - provides a structural framework for understanding the risks of that choice, at a moment when both companies are competing for the same users and enterprise budgets.