Two numbers landed about OpenAI's advertising business within the same stretch of July, and they tell completely opposite stories. Which would be fine — contradictory data is basically my love language at this point — except that both of them appear to be correct.

The first number: the advertiser count on ChatGPT has climbed from around 300 in April to over 820 by mid-July, with roughly 160 joining in a single month. Users on the mobile app are now seeing about twice as many ads per hour. And these aren't scrappy growth-hack startups anymore — we're talking Booking, Intuit, Home Depot, L.L.Bean. The kind of names that make media buyers sit up and stop pretending to read the deck.

The second number: eMarketer reckons OpenAI is on course to miss its own ad revenue forecast by roughly 90 per cent.

Ninety. Per cent.

OpenAI projected $2.5 billion in ad revenue this year, growing to $100 billion by 2030. eMarketer's estimate for the entire US chatbot advertising market — ChatGPT, Copilot, Google's AI Mode, and Alexa for Shopping combined — tops out at $5.41 billion by 2030. OpenAI's single-company target is about twenty times larger than the ceiling an independent forecaster has placed over the whole category.

So the platform is genuinely growing. And the target is genuinely unhinged. Both things, simultaneously, are true. (Welcome to 2026, where cognitive dissonance is a lifestyle choice.)

Whatever you think about that revenue number — and I think it belongs in the same genre as "we'll be profitable by Q3" pitch decks — the product buildout is not something you can wave away.

Since we last looked at ChatGPT's ad suite in detail, OpenAI has shipped basically the entire standard toolkit of performance advertising. Conversion-optimised campaigns. Average daily budgets paced across a seven-day window (the same flexible-spend logic every mature platform uses). Geo-exclusion. A bulk API for creating campaigns at scale. Mobile measurement integrations with AppsFlyer and Adjust. Refreshed product cards with price and star ratings. Custom audiences where you upload your own customer lists by email or phone, apply inclusion or exclusion rules, and set bid multipliers for high-value segments.

None of this is exotic. That's exactly the point. It's the standard performance advertising starter pack, assembled at a speed that's almost suspicious. Throw in the promo credits OpenAI is now dangling in front of new advertisers — the $50-for-$50 and $100 offers that constitute the "coupon phase" every platform from Google to Meta to TikTok Shop has run — and you're looking at a company executing an extremely well-worn playbook with unusual urgency.

The question is whether executing someone else's playbook very fast is the same thing as building something new. (Spoiler: it isn't, but it might not need to be.)

Here's where it gets genuinely interesting — and by "interesting" I mean "slightly awkward for OpenAI's branding team."

Look closely at the toolkit and a quiet admission surfaces. OpenAI isn't inventing an advertising business. It's reconstructing Meta's.

The conversion pixel uses syntax close enough to Meta's that a Facebook advertiser can port across with minimal friction. The custom-audience upload — hashed emails, minimum match sizes, bid multipliers — is the list-based targeting that retargeters have been running for a decade. The promo credits are the oldest customer-acquisition trick in digital media. Even the budget pacing mirrors logic advertisers already know from literally every other buying platform.

Each of these choices is sensible. Together, they make ChatGPT feel like somewhere advertisers have worked before. Which is smart! Lowering the barrier to entry is how you get 820 advertisers in a few months.

But familiarity is a trade. At Cannes, OpenAI pitched a grand shift from the "attention economy" to the "intelligence economy" — a channel judged by whether ads help users complete tasks rather than by impressions or dwell time. Very inspiring. Standing ovation energy.

The machinery now shipping tells a more prosaic story. Conversion objectives, list-based audiences, and a Meta-shaped pixel are the instruments of performance marketing, and they measure precisely the things the Cannes framing said were the wrong things to measure. It's like announcing you've invented a revolutionary new diet and then opening a Domino's franchise. (The Domino's might do very well. But nobody's calling it a revolution.)

What's being built is, operationally, performance advertising with a conversational wrapper. And the wrapper — for all its behavioural novelty — is being monetised through the same levers as the feed.

That's not a failure. It's arguably the fastest route to revenue available. But it reframes what the buildout actually demonstrates: OpenAI can execute a known playbook, not that it's found a new one.

For OpenAI's forecast to hold, eMarketer argues, the company would need to pull search ad budgets across from traditional search en masse, dominate a fully mature chatbot ad market, and outperform essentially every ad format in history — not in sequence, but simultaneously. Any one of those would be a stretch. Requiring all three makes the projection read less like a plan and more like something someone said at a board meeting when things were going well and nobody wanted to be the person who asked "but... how?"

For investors, this is the load-bearing crack. Advertising is expected to supply a substantial share of the revenue base underwriting OpenAI's multi-year compute commitments, so a shortfall in the ad pillar doesn't just dent one company's income statement — it unsettles the foundations beneath the wider AI trade.

But here's the thing: that's a capital-markets problem. It's not our problem.

A seller deciding whether to test ChatGPT ads is not underwriting OpenAI's 2030 revenue. You're asking whether the channel converts today. And here the forecast miss actually inverts into something useful. A category that stays smaller than the hype implied is a category that stays cheaper to enter early. The advertisers arriving now are buying into a surface before the auction fills, before CPMs get bid up by the full weight of search budgets that eMarketer doubts will ever fully migrate.

If the ceiling is lower than OpenAI claims, the ground floor is quieter than it will ever be again. The number that frightens investors is, read from a seller's chair, close to an invitation.

Provided — and this is a significant "provided" — the channel is actually reaching the right buyers.

This is where it gets awkward for ecommerce specifically, and I wish I could sugarcoat it, but sugarcoating is for people who don't read data.

Retail remains ChatGPT's largest advertising vertical by spend. Good news, right? Except retail's share of that spend is falling even as the total climbs. Financial services has recorded the sharpest rise — from about 2 per cent of ChatGPT ad spend in April to roughly 12 per cent by July, making it the third-largest category. Travel, tourism, jobs, and education have all muscled in alongside it, each eating into the slice that commerce once held more comfortably.

This is the awkward inversion beneath the growth story. ChatGPT's commercial promise was always framed around shopping intent — the conversation that drifts from research into purchase, the upper-funnel query that turns into a buy. Yet the ad dollars now accelerating onto the platform are arriving disproportionately from lenders, aggregators, and recruiters. Higher-margin verticals that can absorb a premium CPM more easily than a consumer-goods seller working on thin retail margins.

The platform built on the assumption that its edge was commerce is being colonised — at the level that matters most to its revenue — by categories that aren't commerce at all.

It's a bit like building the world's best farmers' market and then discovering the stalls are filling up with insurance brokers and recruitment consultants. The farmers are still there. They're just getting progressively less space.

None of this settles whether ChatGPT becomes a channel you need to plan around or one you monitor from a distance while focusing your budget elsewhere. The buildout is genuine. The revenue target is fantastical. Both of those remain true after you've accounted for the other.

The question worth carrying forward is narrower and sharper than the headline forecast: if the shoppers are on ChatGPT but the ad economics increasingly favour everyone except the people selling to them, whose platform is it actually becoming?

For sellers, the practical calculus is this: the early-mover window is real, the CPMs are still relatively cheap, and the targeting toolkit is now robust enough to test properly. But "test" is the operative word. This is not "redirect your Meta budget" territory — it's "allocate a meaningful but non-catastrophic test budget and watch what happens" territory.

Because right now, OpenAI is building a platform that talks about commerce, runs on commerce's data, and increasingly charges commerce's competitors less to reach the same users.

And that, if you're selling products online, is a dynamic worth watching very closely indeed.

📹 Google VideoPoet - An LLM for zero-shot video generation

🖼️ Visual Electric - An image generator built for the creative process.

🫟 Rizzle AI - An AI-driven video creation platform.

🏜️ LoudFame - Stylize your video with creative AI filters.

🌐 Lummi - Free stock photos made and curated by AI artists.

😍It was great, all I needed

🙂Good, not great

🙄It sucked