On January 16, OpenAI announced it would begin testing advertisements within ChatGPT — a dramatic reversal for a company whose CEO had previously expressed opposition to an ad-supported model.

On February 9, those ads went live, with major agency holding companies including Omnicom, Dentsu, and WPP among the first buyers.

Here is what we know about how the ads work:

Who sees them: Ads appear only for U.S. users on the free tier and the new US$8/month “Go” subscription plan. Premium tiers (Plus at US$20/month, Pro, Business, Enterprise, and Education) remain ad-free. Users can also opt out of ads in exchange for fewer daily free messages.

What they look like: Ads appear as clearly labelled “sponsored” products and services below ChatGPT’s organic responses. OpenAI has provided examples such as a sauce brand appearing when a user searches for recipes. The company has also teased the possibility of users engaging directly with brand agents within the chat.

How they’re targeted: Ads are served based on the topic of the active conversation, past chat history, and prior ad interactions. OpenAI says advertisers never see personal details or conversation content — they receive only aggregate performance data such as views and clicks.

Pricing: OpenAI is charging US$60 CPMs (cost per thousand impressions), a premium rate comparable to Netflix’s initial asking price when it launched ads. The company is requiring a minimum US$200,000 commitment for its early ad pilot. The CPM-based model is notable: Unlike Google’s cost-per-click search ads, OpenAI is initially positioning ChatGPT ads as brand exposure vehicles rather than direct-response instruments.

Safeguards: Ads won’t appear near sensitive topics (health, mental health, politics), won’t be shown to users predicted to be under 18, and OpenAI pledges that ads will not influence ChatGPT’s actual responses.

Omnicom has more than 30 brands participating in the pilot. As Omnicom’s chief commercial officer Justin Wroe put it, ads in ChatGPT create “a new way for brands to introduce products and services that consumers might not otherwise encounter through organic interactions alone.” Dentsu’s Will Swayne was more direct: “LLMs are the next media frontier.”

The financial imperative

OpenAI’s move to advertising is driven by straightforward economics. The company reportedly has nearly 900 million weekly active users, but only about 5% are on paid plans. OpenAI has told investors it could generate US$112 billion in revenue from free users over the next five years through advertising and in-chat commerce, with projections of US$46 billion from non-paying users by 2030 alone. One analyst model estimated ChatGPT’s ad revenue could reach US$17 billion in 2026 and grow to US$80 billion by 2030.

But there’s a geographic challenge: Nearly 90% of ChatGPT’s users are outside the U.S. and Canada, where ad rates are significantly lower. And the ad infrastructure is still basic — there is no self-service ad buying yet, and measurement capabilities lag well behind established platforms. For now, this is very much a managed pilot, not a scaled marketplace.

What the rest of the AI field is doing

Google/Gemini: Adweek reported exclusively in December that Google has begun telling ad clients it plans to bring advertising to its Gemini chatbot in 2026 — separate from ads already appearing in its AI Overviews and AI Mode search experiences. 

Google’s vice president of global ads subsequently denied the report on X, saying there are “no current plans” to change Gemini’s ad-free status. But the signals are clear: Google already has ads in its AI-powered search products, Gemini has grown to over 750 million monthly active users, and Alphabet reported US$114 billion in Q4 revenue with AI driving improvements across all areas of marketing. 

At their most recent earnings, CEO Sundar Pichai noted Google is “expanding the entire playing field that advertisers can compete on.” Google more than any other player has the blueprint for an ads-based AI model.

Perplexity: The AI search startup, which launched ad testing with brands including Indeed and Whole Foods, has paused accepting new advertisers to reassess its approach. The company generated just US$20,000 in ad revenue out of US$34 million in total revenue. Ad buyers reported struggling to measure performance metrics like click-through rates. As Sonata Insights’ Debra Aho Williamson observed, “Companies that launched with a brand awareness model like Perplexity have struggled to maintain momentum.”

Meta: Meta has announced plans to use AI chatbot conversations to personalise ads across Facebook and Instagram, leveraging the data from its Meta AI interactions to enhance targeting across its existing ad ecosystem.

Anthropic/Claude: Anthropic has taken the opposite approach, pledging Claude will remain ad-free — a position it reinforced with a US$25-US$30 million Super Bowl ad campaign (more on this below). Google DeepMind CEO Demis Hassabis also took a subtle jab at OpenAI at Davos, saying the ad launch “felt early” and suggesting OpenAI may have needed ads to bring in more revenue. 

Regulatory scrutiny begins

U.S. Senator Ed Markey (a Democrat from Massachusetts) wasted no time. On January 22, he sent letters to the CEOs of OpenAI, Anthropic, Google, Meta, Microsoft, Snap, and xAI, warning that embedding ads in AI chatbots “raises significant concerns for consumer protection, privacy, and the safety of young users.” 

Markey’s concerns centre on three risks: that users’ emotional connections with chatbots could be exploited for commercial purposes; that “blurred advertising” could make it impossible to distinguish genuine recommendations from paid placements; and that intimate conversational data could be repurposed for ad targeting. He gave all seven companies until February 12 to respond.

What this means for news media

Content monetised without publisher compensation. As Digiday reported, OpenAI will now make additional revenue from content that ChatGPT pulls from the open Web, created by publishers, without a revenue share. 

As one publishing executive put it, “They’re cutting into our business model, and they’re monetising it beyond just subscriptions … . You’re using our IP to fill out these answers that you’re then getting an advertiser to pay you because they’ll transact there. That’s a bitter pill.” 

Publishers with existing OpenAI licensing deals may want to revisit those agreements as they come up for renewal.

Competition for ad dollars. ChatGPT’s 800+ million weekly users and US$60 CPMs put it in direct competition for brand budgets. 

Scott Messer of Messer Media believes it won’t dangerously erode open Web display budgets for a few years, noting the current spending is “experimental” and “not moving the US$400 billion ad ecosystem.” 

But the longer-term trajectory matters. And as David Caswell of StoryFlow noted, ChatGPT ads could “cannibalise intent-based search advertising much more than CPM-based display advertising” because the intent signal in a conversational interface is so much stronger.

Google may be the bigger threat. Multiple sources in the Digiday report flagged Google’s AI advertising moves as potentially more concerning for publishers than OpenAI’s. As Messer put it, “Google worries me more because now they have an incentive to keep the traffic to themselves by enabling ads in AI Overviews.” 

If Google adds more advertising into AI Overviews or AI Mode, it could further crowd out organic results and accelerate the traffic decline publishers are already experiencing.

LLMs are becoming media platforms. As Bauer Media’s CPO Marcel Semmler observed, “These systems are no longer neutral utilities. They’re becoming media platforms. And once that happens, publishers need a seat at the table — not just as content providers, but as partners with clear rules.” 

This echoes what we’ve been tracking across the INMA Advertising Initiative: the need for publishers to assert their value in emerging channels before the terms are set by others.

Anthropic’s US$30 million Super Bowl sermon

Anthropic, the maker of Claude, used two Super Bowl ad slots (at an estimated US$25-US$30 million total) to make its debut with a campaign called A Time and a Place created by agency Mother and directed by Jeff Low. 

The spots each depict a person asking an AI chatbot a personal question about health, a relationship, a business idea, only for the response to be interrupted by a jarring, irrelevant sponsored ad. In one, a man asks “Can I get a six-pack quickly?” and after providing his height and weight, receives an ad for insoles that “help short kings stand tall.”

The original tagline was pointed: “Ads are coming to AI. But not to Claude.” After Sam Altman attacked the ads as “clearly dishonest,” the version that actually aired during the game used softer language: “There is a time and place for ads. Your conversations with AI should not be one of them.”

Anthropic President Daniela Amodei reinforced the position on LinkedIn: “The conversations people have with LLMs are often very personal: from discussing sensitive health topics, to working through difficult life decisions … . Using intimate details like these to serve ads didn’t feel like a respectful way to treat our users’ information.” 

The accompanying blog post was careful to note that Anthropic is “not anti-advertising” — they run ads for Claude themselves, and many of their customers use Claude for marketing.

Altman fired back on X, arguing ads help make AI accessible to people who can’t afford subscriptions: “More Texans use ChatGPT for free than total people use Claude in the U.S., so we have a differently shaped problem than they do.” OpenAI CMO Kate Rouch added: “What’s not funny: Calling ‘ads’ a betrayal when your business model is selling paid subscriptions to companies.”

The ad industry’s reaction was split. Ad Age’s Tim Nudd ranked Anthropic’s spot as the best ad of the entire Super Bowl, and it won the Super Clio. But according to an iSpot survey, the ad’s likeability score placed it in the bottom 3% of Super Bowl ads over the past five years, with viewers most commonly reacting with “WTF” — suggesting the message may have landed with industry insiders but confused the broader audience. 

And as one satirical analysis observed, “If you are not meaningfully in the consumer business, what is the point of warning consumers about the dangers of your competitors’ consumer business?”

Regardless of where you land on the debate, the Anthropic-OpenAI exchange crystallised the central tension of this moment: AI companies are simultaneously trying to build trust with users and monetise the vast audiences they’ve built. How they resolve that tension will shape the competitive landscape for years to come, and publishers should be paying very close attention.

Banner art: Adobe Stock By Amni.

If you’d like to subscribe to my bi-weekly newsletter, INMA members can do so here.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .