OpenAI launches advertising platform, a wealthy business selling to the poor

Original author: Kaori

Original editor: Sleepy

Sam Altman once described advertising for ChatGPT as its “last resort.”

For a long time, this was a form of restraint. OpenAI still presents itself as a research company, an infrastructure company, and a company trying to make AI capabilities available to everyone. Advertising—an old-school monetization method most familiar to the internet—is treated as a backup option.

But the advertising plan was promoted to official business much faster than expected.

On May 5, OpenAI launched the self-serve advertising platform Ads Manager, beginning to let advertisers run ads on ChatGPT directly or through agencies such as Dentsu, Omnicom, Publicis, and WPP. This was less than three months after the first ad pilot kicked off on February 9.

The platform is still in testing, but the direction is already clear: ChatGPT is no longer just a conversational product—it has started becoming ad inventory, too. OpenAI’s goal is to reach $2.5 billion in ad revenue in 2026, and to push ad revenue to $100 billion by 2030.

With a user base of 900 million, ChatGPT has found that the “free” route is getting harder and harder to sustain.

Annual losses in the hundreds of billions—recharging through ads

OpenAI’s growth has been so fast that traditional internet companies have trouble finding something to compare it with.

But it burns money just as quickly.

HSBC analysts estimated that by the end of 2025, OpenAI could still face a funding gap of $207 billion before 2030. Their cloud and AI infrastructure spending from the second half of 2025 through 2030 could reach $792 billion, and long-term compute commitments by 2033 might come close to $1.4 trillion.

These numbers explain why they need to move into the advertising business.

Subscription revenue can prove users are willing to pay, but it’s hard to cover the reasoning costs of all the free users. Enterprise APIs can contribute cash flow, but they face price wars and model convergence. Capital financing can keep things going, but it dilutes equity and pushes higher valuation pressure back inside the company.

Advertising is the fastest source of non-dilutive revenue. It doesn’t require free users to pay. It doesn’t need to re-educate the market. And it’s easier to explain to investors.

According to Reuters, within six weeks, OpenAI’s ad pilot annualized revenue exceeded $100 million. The ads are only shown to users on the free and Go plans, do not affect ChatGPT’s generated responses, and do not share user data with marketers.

For now, let’s set user privacy aside. Behind this strategy lies an even more fundamental problem.

Advertising is sold to free users, but advertisers are looking for paying users

ChatGPT has 900 million weekly active users, about 50 million paying subscribers, and a free-to-paid conversion rate of less than 6%. Ads are shown only to free users, which means all of OpenAI’s ad inventory comes from the 94% who are unwilling to pay.

The real problem is that ad buyers willing to invest $50,000 to start with generally are not selling products aimed at individual consumers. Enterprise software, SaaS tools, and B2B services—these high-ticket categories are precisely the kinds of decision-makers most likely to be ChatGPT’s paying users. They spend $20 to $200 per month to get stronger models and larger context windows, and ads will never appear on their screens.

Beyond audience mismatch, there’s an even deeper question: even if the ads successfully reach free users, how much ad value can their usage scenarios actually support?

High intent doesn’t equal high conversion

OpenAI’s advertising narrative is built on a core assumption: ChatGPT users enter the conversation with real intent, so ad impressions in high-intent situations are worth more.

That assumption is only correct halfway.

Over the past 20 years, what brands most wanted to capture was the search box, because the search box represents intent. If users search for a hotel, they might be about to book. If they search for corporate tax software, they might be ready to purchase. If they search for the best noise-canceling headphones, they’re already right at the threshold of a consumer decision.

Google built an advertising empire on this. After ChatGPT appeared, users handed the decision-making process directly to AI. For advertisers, that is more tempting—and more terrifying—than search ads. The appeal is that ChatGPT sees the entire demand story: it doesn’t just know what users want to buy, it knows why they want to buy it. The danger is that if AI gives the answer directly, users might not even look at the search results page.

But “help me buy a pair of running shoes” and “help me write an email” are two completely different intents. The former is a consumption scenario; the latter is a productivity scenario. In ChatGPT’s everyday use, the latter accounts for far more than the former. Users come here to write, translate, code, draft plans, and vent their feelings—high-frequency activities that do not naturally map to product purchases.

This directly drags down advertising effectiveness metrics. Advertisers are willing to pay high prices for purchase intent with high certainty. Google search ads are expensive because users often enter the search box with explicit intent to buy, compare, reserve, or order. Meta ads can be cheaper, but Meta has social profiles and massive conversion data, enabling algorithms to repeatedly filter low-intent users into potential consumers.

ChatGPT sits in the middle. It’s more like an entry point for needs than social media, but it’s harder to judge commercial intent than search. It’s more private than search, yet harder to attribute. It can solve users’ problems, but it doesn’t necessarily generate ad clicks.

That’s also why OpenAI is shifting from CPM (cost per thousand impressions) to CPC (cost per click). This isn’t just a product upgrade—advertisers aren’t willing to keep paying based on the idea of a “next-generation search entry point” for the long term. Ultimately, they will ask: whose click is it? Where does the conversion happen? How much of the budget should be moved from Google, Meta, and TikTok to ChatGPT?

Category fit is also a challenge. Low-risk categories such as home, travel, education, and software tools can be tested first. High-profit categories are often also high-regulation categories—such as finance, healthcare, insurance, and recruitment. Once ChatGPT runs ads in these areas, the platform has to take on not only ad performance risk, but also misinformation, discrimination, and compliance risk.

Google’s approach is a mirror. In the first quarter of 2026, Google’s search ad revenue reached $77.25 billion. Even so, Google’s ad placements in AI Mode and AI Overviews remain very cautious, and the standalone Gemini app has not formally started running ads.

OpenAI’s expansion into advertising is essentially exploring broader business models for the large model race as a whole.

OpenAI needs to make users feel that AI is intimate enough, while convincing advertisers that there is sufficient commercial intent here. Once that balance goes off the rails, ChatGPT could end up losing both sides: users may feel it isn’t pure, and advertisers may feel it can’t convert.

But the changes brought by advertising go beyond that—it is also reshaping how brands compete.

The center of gravity for GEO is shifting

Over the past year, brand anxiety has been whether they will disappear from AI answers. The market has packaged this as GEO, but in essence it’s not a new concept—it’s just an updated skin for old search marketing anxiety in the AI era.

OpenAI’s Ads Manager taps into this anxiety, but it also changes where that anxiety points.

In a no-ads era, GEO’s core question is “how to get into AI’s context.” Brands compete by trying to be referenced by models through product documentation, media coverage, third-party reviews, and community discussions—competing on information quality and how well data is structured.

Once the ad platform launches, precision traffic can be bought directly, and brands no longer have to rely solely on natural mentions. But the focus of competition does not return to the traditional “buy more impressions.” Instead, it shifts from “how to enter AI’s answers” to “how AI evaluates my product.”

The reason is simple: after users see an ad, the most natural next step is to ask AI, “is this product actually any good?” AI’s response becomes the real conversion gate. Advertisers may be able to buy impressions, but they can’t buy AI’s positive reviews. If AI gives a negative evaluation based on public data, every dollar spent on ads accelerates user churn rather than driving conversions.

This means brands need to build positive reputation within AI’s evaluation system. Signals that AI can read—product quality, the density of user reviews, and third-party review coverage—will matter more for conversion than the ad spend itself.

GEO is moving from “entering the context” to “winning the evaluation.” This is also the trend worth paying attention to after OpenAI launched its new ad business platform.

Not advertising is the most expensive advertising in 2026

Since we’re talking about OpenAI, we also have to mention its nemesis, Anthropic, which is pursuing a completely different “advertising model.”

On February 4, 2026—two days before the Super Bowl—Anthropic published a blog post and said that Claude will never run ads. There were no sponsored links and no third-party placements.

That statement itself is an expensive advertisement.

Super Bowl ads are not cheap. Anthropic’s costly, high-profile promise of “no ads” is essentially buying brand recognition through advertising—an investment in a no-ads brand image.

Going ad-free has never been only a moral stance—it’s also a business positioning. It tells enterprise clients, professional users, and high-sensitivity scenario users that Claude’s answers won’t be influenced by advertisers, that Claude’s product direction won’t be optimized around ad inventory, and that its revenue comes from what you pay.

The results were immediate. Claude’s ranking in the U.S. App Store climbed from No. 42 at the start of the year to the top. On February 28, after OpenAI signed a Pentagon contract that sparked the QuitGPT movement, Claude first took the No. 1 spot in the U.S. App Store free apps, for the first time ever surpassing ChatGPT. Free active users grew by 60%, daily registrations quadrupled, and paid users doubled within a week.

Anthropic’s revenue structure is completely different from OpenAI’s: more than 80% comes from enterprise clients, and annual recurring revenue often surged from about $9 billion to $19 billion. Enterprise tools such as Claude Code and Cowork have contributed at least $1 billion. Anthropic does not need the advertising value of free users; it needs enterprise clients to trust that their data won’t be used for ads, which creates a premium for privacy.

In this context, choosing not to run ads is a precise business decision: by giving up ad revenue, Anthropic strengthens the trust barrier with enterprise clients, thereby supporting higher subscription pricing.

However, “not running ads” is not an eternal virtue.

Stanford’s AI Index shows that the cost to achieve GPT-3.5 equivalent performance dropped by 280x within two years—from $20 per million tokens in November 2022 to $0.07 in October 2024. If model capabilities continue to converge and a price war in APIs breaks out, the enterprise subscription premium that Anthropic enjoys today could gradually be eroded. When model costs fall to the point where all competitors can provide near-identical performance, why would enterprise clients keep paying more for Claude?

No conclusion exists yet, but time will answer what this choice is worth.

There’s no such thing as a free lunch

OpenAI chose advertising; Anthropic chose to turn not running ads into a premium. They look like opposite paths, but both are answering the same question: when the inference cost of AI products can’t be covered by free models long-term, who pays?

OpenAI’s Ads Manager is not just an advertising product—it’s also a signal that the AI industry is shifting from free expansion to cost recovery.

But the bleeding-control method OpenAI chose exposes the most fragile part of this business. It needs to support an ad pricing model using the user group with the least commercial intent—who cost more to reach than Meta by 3 times.

This isn’t a problem that can be solved by user scale alone. 900 million weekly active users is a nice number, but if those 900 million people come to ChatGPT to write emails rather than buy things, advertisers will eventually vote with their feet.

Ads can be a revenue source for AI products, but they shouldn’t be the only answer. Because when a product’s business model requires users to stay for as long as possible and expose as much intent as possible, that product is no longer an assistant to the user—it becomes an assistant to the advertiser.