On February 9th, 2026, OpenAI began serving advertisements inside ChatGPT. The rollout was quiet: limited to free and low-cost tier users in the United States, with ads appearing as small labeled cards at the bottom of the AI's responses. Most coverage treated it as a predictable monetization milestone. A company needs revenue, it has 800 million users, ads were inevitable. Move on.

I think that framing misses something important. What happened this week is potentially the opening move in the largest restructuring of advertising economics since Google borrowed (stole, depending on who you ask) the pay-per-click auction model from GoTo.com and turned it into the most profitable business in the history of the internet. To understand why, we need to look at a single idea that explains three centuries of advertising history. Then we need to understand why ChatGPT breaks it.

At its core, advertising is a subsidy mechanism. It transfers costs from content consumers to product sellers. The newspaper reader gets cheap news because the advertiser pays for access to the reader's attention. The YouTube viewer gets free video because brands pay to interrupt it. Every innovation in the history of advertising is a variation on making that subsidy more efficient, more targeted, and more measurable. And the way you make it more efficient is by closing the gap between a person's attention and their intent to buy.

For most of history, the gap was enormous. A town crier shouting about a merchant's wares reached everyone within earshot, regardless of whether they cared. A newspaper ad reached readers sorted loosely by class and literacy. Radio and television reached millions simultaneously, which was powerful for building desire but comically imprecise for targeting it. You paid for a vast audience and hoped the right people were in it. And because measurement barely existed (the magazine coupon was about as good as it got), the entire industry ran on faith and gut feeling. This, by the way, is why the creative quality of advertising peaked during the Mad Men era. When you can't measure whether an ad works, you'd better make sure it's convincing enough that you believe it works.

Two developments changed everything. The first was digital measurement: every interaction suddenly became trackable, from impression to click to purchase. The second, and more important, was intent signals. Instead of guessing who might want your product, you could observe behavior that revealed it.

Google captured declared intent. When someone types "best running shoes under $150," they are telling you, in their own words, exactly what they want. The gap between attention and intent collapses to nearly zero. This is why search advertising became the most profitable advertising product ever built. Google didn't create a better ad. They created a better surface: one where the user's own query did all the targeting work.

Meta captured inferred intent. Facebook and Instagram can't see what you're searching for, but they can build an eerily accurate picture of who you are: what you like, who your friends are, what content you engage with, what life events you're experiencing. The gap between attention and intent is wider than search, but Meta compensates with something Google cannot do. It creates desire where none existed. The "I didn't know I needed this" reaction to a well-targeted Instagram ad is demand creation, not demand capture. Google catches people at the bottom of the purchase funnel. Meta operates at the top.

Together, they built a duopoly that has dominated digital advertising for nearly two decades, powered by data network effects that have proven essentially impossible to challenge. More users generate more data, which improves targeting, which attracts more ad spend, which funds better products, which attract more users. The flywheel spins. Billions of dollars have been thrown at breaking it. None of it has worked.

The obvious analysis is that ChatGPT occupies the middle of the purchase funnel: the consideration phase where someone is actively weighing options. Google owns the bottom (declared intent, ready to buy). Meta owns the top (brand discovery, desire creation). ChatGPT slots neatly into the gap between them, where people deliberate. That analysis is clean, intuitive, and undersells what is actually happening by an order of magnitude.

What ChatGPT is building is not a new position in the funnel. It is the collapse of the entire funnel into a single surface.

Start at the top, where Meta has reigned for over a decade. Hundreds of millions of people already use ChatGPT for research, exploration, and learning. They come to discover things they didn't know about. Look at the example on OpenAI's own announcement page: a user is planning a potluck dinner and asking for easy entree ideas. The response includes recipes, and at the bottom, a sponsored card for a grocery store. Now imagine the same mechanic applied to someone asking how to dress better. A sponsored card for an emerging designer label the user has never heard of is functionally identical to the Instagram ad that makes you want something you didn't know existed. That is demand creation. That is Meta's territory. And ChatGPT can do it inside a conversation where the user has already told the system exactly what they care about, which is something Meta has to infer from behavioral signals.

The middle is where the new value is most obvious. When someone has a multi-turn conversation weighing the pros and cons of two laptops, explaining their budget constraints, mentioning that they travel for work, asking about durability because their last one broke, the system doesn't just see a purchase signal. It sees the full deliberation: constraints, hesitations, trade-offs, priorities. Not the output of a decision, but the decision itself, in progress. No advertising platform has ever had access to this. Until now, that process happened inside people's heads, in conversations with friends, or scattered across dozens of review sites with no single owner.

But then there's the bottom of the funnel, which is where this gets genuinely disruptive. People come to Google with three types of intent: informational ("what is X"), navigational ("take me to X"), and transactional ("I want to buy X"). ChatGPT already handles all three. Users research, navigate, and increasingly, transact. OpenAI has built a product called shopping mode, where the AI asks you clarifying questions, you select from multiple-choice responses, and it narrows you to a specific product. They've built Instant Checkout, powered by something called the Agentic Commerce Protocol, co-developed with Stripe. Etsy sellers are already live. Over a million Shopify merchants, including brands like Glossier, SKIMS, Spanx, and Vuori, are coming online. The user describes what they want, the AI recommends the most relevant products, and they can buy without ever leaving the conversation.

Read that again. Google search ads work by sending you away from Google to the merchant's site. Meta ads work by sending you away from Instagram to the product page. ChatGPT is building toward a model where discovery, research, comparison, decision, and purchase all happen within the same conversation. The user never leaves. The entire purchase funnel, which the advertising industry has spent a century mapping, segmenting, and optimizing piece by piece, collapses into a single interface. That is not a new ad format. That is a structural threat to the two most profitable advertising businesses ever built, simultaneously.

Credit where it's due: OpenAI has clearly thought hard about this. The controls they've put in place are more sophisticated than what Google or Meta offered at equivalent stages of their advertising businesses.

Paid subscribers (Plus, Pro, Business, Enterprise, Education) will never see ads. Only users on the free and $8/month Go tiers are exposed. Even within that group, users get granular control over personalization. They can decide whether ads can reference their memories, their chat history, or any conversational context at all. They can disable personalization entirely and still use the product. They can clear their ad data with a single tap. Ads will not appear near sensitive topics like health, mental health, or politics. Users predicted to be under 18 won't see ads at all.

Most interesting is the architectural decision. The model itself does not know that an advertisement is being served. The ad system reads the conversation context and selects a relevant placement, but the information flow is one-directional. The model cannot be influenced by the ad because the model cannot see the ad. This is a structural guarantee, not just a policy promise. It's the difference between a bank saying "we promise not to steal your money" and a bank putting your money in a vault that its own employees can't access.

But then there's a small button on each ad: "Ask ChatGPT." Click it, and the ad enters the conversation. Now the model does know about the product. The user can ask questions, request comparisons, evaluate whether it's worth buying. This is elegant in theory: the user voluntarily recruits the AI to help them assess a commercial offering. In practice, though, it creates something more subtle. If someone is talking to ChatGPT about hiking gear, sees an ad for boots, taps "Ask ChatGPT," and the AI then helps them evaluate those boots within the conversation, the experience is functionally indistinguishable from the AI recommending the product. The architecture says otherwise. The user won't care about the architecture. If it feels like the AI is recommending boots, that is the reality that matters for trust.

It is worth understanding who is running this operation. OpenAI's CEO of Applications is Fidji Simo, who previously spent over a decade at Meta, where she did something that everyone at the time said was impossible: she put ads in the Facebook News Feed without destroying engagement. In fact, those News Feed ads became the most profitable advertising product since Google search. She then went to Instacart and built a retail advertising platform where every inch of screen real estate became a potential revenue source. OpenAI did not hire Fidji Simo for her restraint.

Then there is Sam Altman's own evolution on the subject. In May 2024, he told an audience at Harvard Business School that advertising would be "a last resort" for ChatGPT, adding that "ads-plus-AI is sort of uniquely unsettling to me." By June 2025, speaking on The OpenAI Podcast, his tone had shifted noticeably: "I'm not totally against it. I think ads on Instagram are kind of cool. I've bought a bunch of stuff from them." By January 2026, ads were announced. That's an 18-month journey from "uniquely unsettling" to "kind of cool" to "here they are." The rate of that reversal is the tell. When the CEO of a company publicly signals that he finds his own product's most profitable business model cool, by referencing the exact platform his head of applications built it at, you are not looking at a cautious experiment. You are looking at the early infrastructure of something much larger.

Every platform follows the same arc. Facebook launched with small, unobtrusive sidebar ads and declared the News Feed sacred. Then the News Feed got ads. Then the ads got bigger. Then they got video. Then they became the primary revenue engine. YouTube pre-roll ads used to be skippable after five seconds. Google search results used to have clear visual separation between organic and paid listings; try telling them apart now. The guardrails are always most generous at launch, when the platform needs user trust and has no advertiser ecosystem demanding more. The loosening happens slowly, invisibly, and always in the direction of more revenue.

OpenAI's current ad revenue per free user is projected at roughly $2 annually. Meta generates around $50 per user. That gap is a pressure gradient. It does not close by being polite.

This is not a story about a new ad format. It is a story about who controls the recommendation layer of the internet.

Google became the gateway for commercial intent: if you wanted to buy something, you started there, and Google extracted rent from the traffic that flowed through. But Google only ever controlled one stage of the journey. You searched on Google, but you discovered on Instagram and you bought on the merchant's site. The stages were distributed, and the value was split between the platforms that owned each piece. What ChatGPT is building is the unification of all those stages under one roof. If discovery, evaluation, and purchase all happen inside the same conversation, there is no traffic to split. There is one gateway, and it captures value at every step. We are watching the formation of the most complete commercial chokepoint in the history of the internet, and most people are treating it as a banner ad story.

The competitive responses are already crystallizing. Anthropic, which makes Claude, ran Super Bowl ads this week explicitly mocking the idea of advertising in AI, with actors playing chatbots who deliver helpful advice alongside clumsy product pitches. Sam Altman called the commercials "dishonest" and Anthropic "an authoritarian company," which is roughly the reaction you'd expect from someone whose most vulnerable strategic decision just got lampooned in front of 120 million people. Google has announced plans to bring ads to Gemini in 2026. The AI advertising wars are not theoretical. They are happening.

And here is what makes this different from every previous advertising platform shift. People share things with AI assistants that they would never share with a search engine, a social feed, or sometimes even a friend. Medical fears. Financial anxieties. Relationship problems. Career doubts. Late-night spirals about whether their life is going in the right direction. The intimacy of that relationship is categorically different from typing a query into Google or scrolling a Facebook feed. The question of whether conversational AI is more like social media (where users learned to tolerate ads) or more like a therapist's office (where a product placement would be a violation) is not academic. It is worth hundreds of billions of dollars, and nobody knows the answer yet.

What we do know is this: every previous time we've faced this choice, between a trust-based utility and a commercially optimized surface, we chose ads. Search, social media, streaming video, ride-sharing, messaging. Every single time. The pattern is so consistent it might as well be a law of consumer technology. The product launches clean, earns trust, reaches scale, introduces ads carefully, and then the economics take over. By the time users notice the shift, it's already the new normal. The interesting question is not whether this will happen with AI. It is whether we'll notice it happening this time.