OpenAI just proved AI ads are real. It also proved they're building a walled garden.
Tal โ Founder at Lulu
August 24, 2026
Six months after its US pilot, OpenAI is expanding ChatGPT Ads into 31 European markets next week โ 40 markets total. Tens of thousands of marketers have already run campaigns. The platform now ships an Ads Manager, an OpenAI Pixel, a Conversions API, geo-targeting, custom audiences, and bidding that optimizes for real outcomes, not just clicks. This is a serious, fully-built ad platform, and it's real infrastructure, not a beta.
Take that at face value: it's the single biggest market signal the agent-ads category has gotten so far. The company with the most-used AI product on earth just spent six months proving that people transacting inside an AI conversation โ planning a trip, picking software, furnishing a home โ is a targetable, monetizable, high-intent moment. That validates the entire premise we built Lulu Ads on a year ahead of it.
It also tells you exactly where OpenAI thinks this ends: inside ChatGPT. The Pixel fires on OpenAI's surface. The Conversions API reports back to OpenAI's Ads Manager. Every measurement tool they shipped assumes the conversation is happening in their app, on their infrastructure, under their roof. That's not a criticism โ it's the correct move for a company that owns one enormous, single surface. But it's a bet on a specific shape of the future: one dominant chat app, monetized like a search engine, with the ad stack built to match.
We're betting on the other shape
The agent economy we're watching build up under Lulu MCPs โ 75,000+ servers and growing โ doesn't look like one app. It looks like the App Store explosion happening one layer up the stack: thousands of independent MCP servers, developer tools, coding agents, and autonomous frameworks, each one a separate place where an agent actually does something. Claude Code calling a flights server. A CI pipeline calling a compliance checker. A Telegram bot calling a weather tool. None of that traffic ever passes through ChatGPT, and OpenAI's Pixel has no way to see it, target it, or monetize it.
If the agentic future is genuinely decentralized โ and every count we track says it is โ then a walled-garden ad stack can only ever monetize the fraction of it that happens to route through one company's chat window. Everything else stays unmonetized, the same way it's unmonetized today. That gap is the whole opportunity, and it's structural, not temporary: OpenAI can't build a Pixel for a server they don't operate.
What "built for everywhere" actually means
We're not building a smaller ChatGPT Ads. We're building the layer that works precisely because it doesn't assume a chat window:
Agent-native, not app-native. One line of middleware โ mcp.add_middleware(LuluAdsMiddleware()) โ turns any MCP server's tool response into a slot the host model judges on its own, whichever host that is. No Pixel to embed, because there's often no page to put one on.
Reach OpenAI structurally can't have. Every server on Lulu MCPs that flips on monetization earns from calls made inside Claude, Cursor, a CI job, or a custom agent framework โ surfaces a single-app ad network never touches, by construction.
An SDK instead of an ad manager. Publisher setup is a few minutes for a developer, not a campaign console for a marketer. The unit we monetize is a tool call, not a chat turn โ data, never directives, and priced only on real conversions, because a tool call has no impression to sell in the first place.
Open by default. Every MCP server, on every host, gets the same one-line path to earning. It's not gated behind being big enough to get a sales call from a platform.
Two bets on the same validated market
OpenAI just spent six months and a real product proving AI-native advertising works. We take that as confirmation, not competition โ closed and open aren't fighting for the same inventory. OpenAI monetizes the surface it owns. We're building the layer for the tens of thousands of surfaces nobody owns, which is most of where the agent economy will actually run.
If you run an MCP server, that's the inventory this is built for โ check your listing. If you're a developer shipping tools into the open half of this market: getlulu.dev/publishers takes a few minutes.
Your tools already have the traffic.
One line of code makes them earn โ 70% goes to you.
Join the publisher beta ->