Eric Seufert: ChatGPT's ad launch mirrors Netflix's slow start — the real question is how fast OpenAI can build a Meta-style ad stack

Summary

OpenAI's ads launch mirrors Netflix's initial approach—a deliberate but primitive MVP that signals the real question: how fast can the company evolve toward Meta's far more sophisticated ad stack.

OpenAI is charging $60 CPM on a commitment-free basis with minimal targeting or measurement. Eric Seufert argues this is textbook platform launch strategy. You cannot build conversion-optimized ads before you have conversion data. The initial phase bootstraps that data and gathers advertiser feedback. Targeting, measurement, and creative optimization happen later.

Seufert estimates six to eighteen months for OpenAI to move from this basic offering to something resembling Meta's capability. Netflix took years and stumbled because it partnered with an external platform (Xandr) rather than building its own ad tech. OpenAI is building proprietary infrastructure from the start. The company has also acquired Statsig, a Meta-alumni experimentation platform, and hired Fiji Simo, who led Facebook's mobile news feed ads—one of the most successful ad products in history.

Seufert reads the internal dynamics as investor-driven. Sam Altman said two years ago that ads were a "failure state," but investors pushed the inevitability argument. Ads are not optional if you want to monetize at scale. The hiring of Simo and acquisition of Statsig suggest a move to bring commercial discipline to a research-first organization.

Instant checkout

OpenAI quietly rolled out instant checkout on ChatGPT. Seufert reads this as a stalking horse for conversion data. The 4% fee on transactions is not viable long-term economics for most e-commerce, but it lets OpenAI collect conversion signals that feed ad targeting. Once that data engine is built, instant checkout likely fades and the real ads business emerges.

Attribution in a post-ATT world

ChatGPT has structural advantages over Netflix's CTV ads. Netflix relies on IP address matching in clean rooms, which is probabilistic and delayed. ChatGPT gets click data, CAPI (server-side conversion tracking that bypasses device limitations), and pixels. That is direct response fidelity, similar to Facebook. Netflix's measurement ceiling is much lower.

The creepy ads risk

Seufert is skeptical this will tank OpenAI's ads effort. If users truly hated targeted ads, Meta's user base would not grow monotonically each quarter, and Europeans would be stampeding toward the subscription option (no ads). Only 1% of EU users opted into Meta's paid tier despite DMA-mandated ad-free alternatives. People tolerate ads because they prefer free tools over paywalls. OpenAI can also sidestep the perception problem by decoupling ads from chat context, serving display ads based on user intent signals rather than conversation history.

Compounding effects in direct response

Seufert emphasizes a dynamic that investors have underestimated. Meta's AI improvements in ad performance produce 3.5% to 5% quarterly bumps from models like Andromeda and Lattice. They compound. A marketer hitting 110% ROAS over 120 days reinvests that 110% back into the channel. A 5% improvement quarterly gets multiplied by that reinvestment loop. That is why Meta's growth is reaccelerating into Q1 2026 and why Meta stock jumped 10% after earnings. The heavy CapEx is paying measurable dividends in ad performance.

Long-tail agents and networks

AI agents that want to monetize via ads but lack the scale or in-house expertise to build full ad stacks will need to tap existing DSPs rather than a new ad network. Seufert sees little reason The Trade Desk would not win that opportunity. The winners at scale—ChatGPT, Perplexity, Google's Gemini—will build proprietary stacks because that is where margins and control live.

Apple's AI embarrassment

Apple will use Google's Gemini for Siri and other AI features, which creates a privacy optics problem. Apple spent years positioning itself as privacy-first, insisting it would do AI the "Apple way." Then it outsourced the heavy lifting—data sourcing, model training, and ads monetization—to Google. Seufert frames this as privacy washing. Apple gets the brand benefit of "we do this privately," while Google does the actual work. Apple will make money on the back end as Google monetizes Gemini access in iOS, but Apple avoids the regulatory and reputational cost of training models on user data. Until on-device models are powerful enough to handle most use cases, Seufert expects Apple to continue outsourcing to Google for frontier AI, just as it has outsourced search for years.

Netflix-Warner Bros

Netflix valued WB Games at zero in the deal, despite the franchise being worth billions when standalone. The issue is that Netflix would need licensing agreements to keep monetizing those games, making them a liability rather than an asset. The real signal: Netflix is buying IP to boost engagement and thus ad CPMs and subscription retention. Matt Damon reportedly said Netflix now requires constant plot exposition in dialogue because viewers are on their phones. High-end IP might fix that attention problem.

TikTok's commerce gamble

TikTok is doubling down on commerce after initially walking it back. Seufert suspects this is a gap-filling move—Meta abandoned social commerce, so TikTok is moving in. But he is skeptical the model works long-term in Western markets. The brands scaling to hundreds of millions on TikTok Shop were subsidized by ByteDance for years. New ownership (Sequoia-led and operationally separate from ByteDance) may not have the appetite to fund that subsidy. The economics only work if ads, not commerce, drive the core business.