Sam Altman once referred to advertising as ChatGPT's "last resort."

For a long time, this statement was one of restraint. OpenAI continued to position itself as a research company, an infrastructure company, one striving to democratize AI capabilities for everyone. Advertising—the oldest and most familiar monetization method on the internet—was treated as a backup option.

But the ad campaign was quickly approved.

On May 5, OpenAI launched its self-service advertising platform, Ads Manager, allowing advertisers to place ads on ChatGPT directly or through agencies such as Dentsu, Omnicom, Publicis, and WPP—less than three months after initiating its first ad pilot on February 9.

The platform is still in testing, but the direction is clear: ChatGPT is no longer just a conversational product—it’s also becoming an advertising inventory. OpenAI aims to generate $2.5 billion in advertising revenue by 2026 and scale advertising income to $100 billion by 2030.

ChatGPT, with a user base of 900 million, has found that the free model is becoming increasingly difficult to sustain.

Lost billions in a year, relying on advertising to recover

OpenAI is growing so rapidly that it's hard to find a comparable traditional internet company.

But it also burns through money quickly.

HSBC analysts project that by the end of 2025, OpenAI may still face a $207 billion funding gap before 2030. Its cloud and AI infrastructure spending between the second half of 2025 and 2030 could reach $792 billion, with long-term compute commitments approaching $1.4 trillion by 2033.

This set of numbers explains why they entered the advertising business.

Subscription revenue demonstrates that users are willing to pay, but it struggles to cover the cost of serving all free users. Enterprise APIs can contribute cash flow, yet face price wars and model convergence. Equity financing can extend runway, but dilutes ownership and pushes higher valuation pressures back onto the company.

Advertising is the fastest non-dilutive revenue source. It doesn’t require free users to pay, doesn’t need re-educating the market, and is easier to explain to investors.

According to Reuters, OpenAI’s advertising pilot generated annualized revenue of over $100 million in six weeks. The ads are shown only to free and Go plan users, do not affect ChatGPT’s generated responses, and do not share user data with advertisers.

Setting user privacy aside for now, this strategy hides a more fundamental issue.

Ads are sold to free users, but advertisers are looking for paying users.

ChatGPT has 900 million weekly active users and approximately 50 million paid subscribers, with a free-to-paid conversion rate of less than 6%. Advertising is targeted exclusively at free users, meaning OpenAI’s entire ad inventory comes from the 94% of users unwilling to pay.

The issue is that advertisers willing to invest $50,000 or more typically aren’t selling consumer-facing products. Decision-makers for enterprise software, SaaS tools, and B2B services—high-ticket categories—are precisely the most likely to be paying users of ChatGPT. They spend $20 to $200 per month on stronger models and larger context windows, yet ads never appear on their screens.

Beyond audience mismatch, there is a deeper issue: even if the ads successfully reach free users, can their usage scenarios support high advertising value?

High intent does not equal high conversion

OpenAI's advertising narrative is built on a core assumption: ChatGPT users enter the chatbox with genuine intent, and ad exposure in these high-intent scenarios commands a higher price.

This assumption is only half correct.

Over the past two decades, brands have most wanted to secure the search box, because it represents intent. When a user searches for a hotel, they may be planning to book a room; when they search for business tax software, they may be ready to make a purchase; when they search for the best noise-canceling headphones, they are already at the doorstep of a purchasing decision.

Google built its advertising empire on this. After ChatGPT emerged, users began handing the decision-making process directly to AI. This is both more appealing and more frightening for advertisers. It’s appealing because ChatGPT sees the entire context of a need—it doesn’t just know what the user wants to buy, but why they want to buy it. It’s frightening because if the AI gives a direct answer, users may never even look at a search results page.

But “help me buy a pair of running shoes” and “help me write an email” represent two entirely different intents—the former is a consumption scenario, while the latter is a productivity scenario. In ChatGPT’s everyday usage, the latter far outweighs the former. Users come here to write, translate, debug code, develop plans, and organize their thoughts—frequent activities that do not naturally correspond to product purchases.

This directly lowers advertising performance metrics. Advertisers are willing to pay a premium for high-confidence purchase intent. Google search ads are expensive because users often enter search queries with clear intentions to buy, compare, book, or order. Meta ads are somewhat cheaper, but they leverage social profiles and vast conversion data to use algorithms to repeatedly filter low-intent users into potential customers.

ChatGPT is caught in the middle. It functions more like a demand entry point than social media, yet it’s harder to discern commercial intent than search. It’s more private than search, but even harder to attribute results to. It can solve user problems, but doesn’t necessarily generate ad clicks.

This is also why OpenAI shifted from CPM (cost per thousand impressions) to CPC (cost per click)—it wasn’t just a product upgrade, because advertisers were unwilling to pay long-term based on the vision of “the next-generation search interface.” Ultimately, they want to know: Who drove this click? Where did the conversion happen? And how much of the budget should be reallocated from Google, Meta, and TikTok to ChatGPT?

Category compatibility is also an issue. Low-risk categories such as home goods, travel, education, and software tools can be tested first. High-profit categories, such as finance, healthcare, insurance, and recruitment, are often highly regulated. If ChatGPT runs ads in these areas, the platform will face not only advertising performance concerns but also risks of misinformation, discrimination, and non-compliance.

Google's approach is a mirror. In the first quarter of 2026, Google's search advertising revenue reached $77.25 billion. Yet even so, Google remains highly cautious about integrating ads into AI Mode and AI Overviews, and its standalone Gemini app has still not officially introduced advertisements.

OpenAI's expansion into advertising is exploring broader business models for the large model industry as a whole.

OpenAI must make users feel that AI is sufficiently intimate while convincing advertisers that there is sufficient commercial intent. If this balance is lost, ChatGPT will alienate both sides: users will perceive it as inauthentic, and advertisers will see it as ineffective at driving conversions.

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

GEO's focus is shifting.

Over the past year, brands have been anxious about disappearing from AI-generated responses. The market has framed this as GEO, but it’s not a new concept—merely the old search marketing anxiety repackaged for the AI era.

OpenAI launched Ads Manager, which precisely taps into this anxiety but also shifts its focus.

In the ad-free era, GEO’s core challenge is “how to enter AI’s context.” Brands compete by earning citations from models through product documentation, media coverage, third-party reviews, and community discussions—measured by the quality of information and the degree of data structuring.

After the advertising platform launched, targeted traffic can be purchased directly, so brands are no longer solely reliant on organic referrals. However, the focus of competition has not reverted to the traditional approach of “buying more exposure”; instead, it has shifted from “how to get into AI’s answers” to “how AI evaluates my product.”

The reason is simple: after seeing an ad, the most natural next step for users is to ask AI, “Is this product actually any good?” AI’s response becomes the true conversion gatekeeper. Advertisers can buy impressions, but they cannot buy positive reviews from AI. If AI provides negative feedback based on public data, every dollar spent on advertising accelerates user attrition rather than driving conversions.

This means brands need to build a positive reputation within AI evaluation systems. Signals that AI can detect—such as product quality, density of user reviews, and coverage by third-party evaluations—will have a greater impact on conversion rates than advertising spend itself.

GEO is moving from "entering the context" to "earning evaluation," a trend worth noting following OpenAI's launch of its new advertising platform.

Not advertising is the most expensive advertisement in 2026.

After discussing OpenAI, we must mention its rival Anthropic, which is pursuing a completely different advertising model.

On February 4, 2026, two days before the Super Bowl, Anthropic published a blog post stating that Claude will never display advertisements—no sponsored links, no third-party integrations.

This statement itself is an expensive advertisement.

Super Bowl ads aren't cheap; Anthropic spent heavily to tell users it doesn't sell ads, essentially using advertising to buy brand awareness without ads.

No ads has never been just a moral stance—it’s also a business positioning. It tells enterprise customers, professional users, and those in highly sensitive scenarios that Claude’s responses are not influenced by advertisers, Claude’s product direction is not driven by ad inventory optimization, and Claude’s revenue comes from what you pay.

The impact was immediate. Claude’s ranking on the U.S. App Store rose steadily from 42nd at the start of the year. On February 28, following OpenAI’s signing of a Pentagon contract that sparked the QuitGPT movement, Claude topped the U.S. App Store’s free apps chart for the first time ever, surpassing ChatGPT. Free active users increased by 60%, daily registrations quadrupled, and paying users doubled within a week.

Anthropic’s revenue model differs entirely from OpenAI’s: over 80% comes from enterprise customers, with annualized recurring revenue surging from approximately $9 billion to $19 billion. Enterprise tools like Claude Code and Cowork have already generated at least $1 billion in revenue. Anthropic does not rely on the advertising value of free users; instead, it requires the trust premium from enterprise clients who expect their data will not be used for advertising.

Choosing not to advertise is a precise business decision: by forgoing advertising revenue, the company strengthens trust among enterprise clients, enabling higher subscription pricing.

However, "not advertising" is not an eternal virtue.

Data from the Stanford AI Index shows that the cost to achieve performance equivalent to GPT-3.5 has decreased 280-fold over two years, dropping from $20 per million tokens in November 2022 to $0.07 in October 2024. If model capabilities continue to converge and an API price war ensues, the enterprise subscription premium Anthropic currently enjoys may gradually erode. When model costs fall to the point where all competitors can offer similar performance, what reason will enterprise customers have to pay more for Claude?

There is currently no conclusion on this issue, but time will provide the answer.

There's no such thing as a free lunch.

OpenAI chooses advertising; Anthropic chooses to make not advertising a premium feature. These appear to be two opposing paths, but both are answering the same question: when the cost of AI product inference can no longer be sustainably covered by a free model, who pays?

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

But the way OpenAI chose to stem the bleeding exposed the most vulnerable aspect of this business: it needs to rely on the least commercially motivated user base to support an advertising price three times higher than Meta's.

This is not a problem that can be solved by user scale. Nine hundred million weekly active users is an impressive number, but if those 900 million people are coming to ChatGPT to write emails rather than to shop, advertisers will eventually vote with their feet.

Advertising can be a revenue source for AI products, but it shouldn't be seen as the only solution. When a product’s business model requires users to stay as long as possible and reveal their intentions as much as possible, it is no longer serving the user—it’s serving the advertiser.