After years of calling advertising a “last resort,” OpenAI is now testing ads inside ChatGPT. The shift is drawing criticism from users, skepticism from privacy advocates, and mockery from rivals.
From “Last Resort” to “Let’s Test It”
For years, OpenAI CEO Sam Altman was clear about his feelings on advertising. In May 2024, he told an audience at Harvard that using ads in ChatGPT would be a “last resort” and that “ads plus AI is sort of uniquely unsettling” . Just six months later, his tone had shifted. “I personally hate ads,” he said, “but not completely opposed, nor would I say never consider them” .
By January 2026, that consideration became action. OpenAI announced it would begin testing impression-based advertising in ChatGPT, starting with free users and subscribers to the newly launched ChatGPT Go tier ($8/month) . The company officially launched the test on February 9, 2026, in the United States .
What changed? The short answer is money. OpenAI reportedly lost around $8 billion in operating expenses during the first half of 2025 . Despite having approximately 800 million weekly active users, only about 5% pay for subscriptions . Each ChatGPT conversation carries significant computing costs—estimated at roughly $0.01 per interaction, which means 10 billion daily interactions would burn $10 million .
“Our mission is artificial general intelligence for the benefit of humanity,” OpenAI CFO Sarah Friar said at the World Economic Forum in Davos, “not for the benefit of humanity who can pay” . Advertising, in this framing, is a way to keep the service accessible to non-paying users while generating revenue to cover ballooning infrastructure costs.
How ChatGPT Ads Actually Work
The ad format OpenAI is testing differs significantly from traditional digital advertising. Rather than banner ads or pop-ups, ChatGPT ads appear as sponsored content placed below the chatbot’s response, clearly labeled and visually separated from the answer .
Here’s an example: A user asks for “simple but authentic Mexican dish ideas for a dinner party.” ChatGPT provides a recipe. Below that response, a sponsored carousel might appear showing ingredients or related products from a grocery delivery service, complete with pricing and delivery estimates .
OpenAI has made several key promises about how the system operates:
- Ads do not influence answers. The company insists that responses are driven entirely by what’s objectively useful, not by commercial relationships .
- Targeting is contextual, not personal. Ads are triggered by the topic of conversation rather than user profiles or browsing history. If you ask about travel, you might see hotel recommendations—but OpenAI says it won’t sell your conversation data to advertisers .
- Users have control. Free users can opt out of ad personalization, though doing so may reduce their daily token limits . They can also dismiss individual ads and provide feedback.
- Premium tiers remain ad-free. ChatGPT Plus ($20/month), Pro ($200/month), Business, and Enterprise subscribers will not see ads .
The ad inventory is currently being sold through multiple channels. OpenAI sells placements directly at a reported $60 CPM (cost per thousand impressions) with a minimum spend of $200,000. But third-party demand-side platform StackAdapt has been offering access through a pilot program at CPMs ranging from $15 to $60, with a $50,000 minimum . The sharp drop in pricing—75% below OpenAI’s direct rate—suggests that advertiser demand hasn’t yet met the company’s initial expectations .
The Privacy Questions No One Has Fully Answered
The shift to advertising raises concerns that go beyond traditional digital marketing. ChatGPT isn’t a search engine or a social media feed—it’s a conversational tool where users routinely share personal, emotional, or even sensitive information .
OpenAI says ad targeting will be contextual, not behavioral. That means ads are triggered by the topic of conversation rather than a user profile built from browsing history. But even contextual targeting inside a chat interface is new territory.
“Conversational AI doesn’t know what users are interested in, like traditional digital advertising, but it often knows why,” noted one analysis. “This makes advertising decisions far more sensitive and potentially powerful” .
The company has promised not to share user conversations with advertisers. But critics point out that OpenAI hasn’t fully explained what internal data it will use to determine ad relevance, how long that data is retained, or who has access to it .
There are also concerns about how the system might evolve. “Once their advertising turns into a serious income stream, the platform may shift to looser protocols,” warned one tech publication. “The internal use of data to optimize the relevance of ads and measure their effectiveness could expand, even if OpenAI never sells the data externally” .
The User Backlash Has Been Immediate
When OpenAI announced the test, the response from users was swift and largely negative. Comments on social media and forums echoed a common theme: distrust.
“Google once said ads would always be clearly labeled and separate, but over time the distinction blurred,” one user wrote on OpenAI’s social media post about the announcement .
Another put it more bluntly: “OpenAI is harming user interests. They can decide what you say, what you do, what you know—and none of it is transparent” .
Even rival Anthropic got in on the criticism. During the 2026 Super Bowl, the company ran ads that ended with the tagline: “Ads are coming to AI. But not to Claude” . It was a direct jab at OpenAI’s decision to monetize through advertising.
Internal OpenAI data suggests the company is aware of the challenge. According to ad measurement firm iSpot, ChatGPT’s television ads have struggled to resonate with consumers. Of twelve ads tested between the 2025 Super Bowl and February 2026, only one (“Family Farm”) outperformed category norms. The rest underperformed, often by wide margins .
The issue appears to be the abstract nature of AI marketing. Many of ChatGPT’s earlier ads featured long blocks of text scrolling across the screen—a visual representation of how the AI works. Viewers consistently reacted poorly, finding the approach “lazy, unoriginal, or simply frustrating” . The more successful “Family Farm” spot took a different approach, telling a grounded story about a multigenerational farm using ChatGPT to solve everyday problems.
Even that ad drew criticism from sophisticated viewers who noted the irony: “I think it’s ironic how she asked how much water she is using on her farm,” one viewer commented. “Meanwhile AI data centers use millions of gallons of water just to answer that question” .
The Financial Imperative
Despite the backlash, OpenAI’s move toward advertising appears driven by necessity rather than choice.
The company’s financial picture is challenging. Annualized revenue reportedly surpassed $20 billion in 2025, but operating losses remain substantial . The company has spending commitments tied to data centers and chips that some reports estimate at $14 trillion over the long term .
Subscription growth has also plateaued. With only 5% of users paying, simply raising prices or converting more free users isn’t a viable path to profitability—especially in price-sensitive markets outside the US.
Advertising, by contrast, offers a scalable revenue stream that could support free users while generating hundreds of billions in projected long-term income. OpenAI reportedly forecasts that ad revenue could exceed $10 billion by 2027, with a 2030 target of $110 billion from non-paying users .
Those numbers may seem optimistic, but they reflect the scale of ChatGPT’s user base. The platform generates an estimated 80 million product-related conversations per week—roughly 4 billion product queries annually . If even a fraction of those convert to sales, the advertising opportunity is enormous.
What Happens Next
The ad test is currently limited to US users on free and Go tiers. OpenAI has said it will use the pilot to refine the experience before any broader rollout.
But the company is already looking beyond simple display ads. Future iterations could include more agentic advertising—where AI proactively finds deals and recommendations based on long-term user preferences, rather than simply responding to immediate prompts .
“In the next 10 years, the advertising model will evolve into a more Agentic interaction,” OpenAI has suggested. “AI can actively find the best discounts and deals based on the user’s long-term preferences, transforming from passive search to active discovery” .
Whether users will accept that vision remains an open question. For now, the ads are here—clearly labeled, separated from answers, and theoretically harmless. But as one analyst put it: “Whether users continue to believe that as ads become more familiar is an open question” .