Who Pays for Free AI? Ads Are Coming to Chatbots

Since 24 August 2026, ChatGPT users in Poland on the free plan may see an ad below an answer. Google is expanding ads in AI Mode to more countries, while Elon Musk has said advertisers will be able to pay for placement inside Grok’s answers. That changes more than the look of the screen: the business model may start shaping what you hear, and in a chat there aren’t ten other links beside the answer for you to compare.

Consider what happened before OpenAI officially sold its first ad. In early December 2025, paying ChatGPT users began posting screenshots online showing the chat suggesting Peloton and Target apps in the middle of conversations, unrelated to their questions. OpenAI said these were not ads, but a test of app suggestions for which nobody had paid. Mark Chen, the company’s head of research, nevertheless admitted that “anything that looks like an ad requires caution, and we fell short”, and the feature was turned off. The small misstep revealed something important: in a chat, users may not distinguish a suggestion from an ad. They simply see a sentence from something they trust.

From “last resort” to a billion-dollar business

In 2024, Sam Altman said at Harvard that combining advertising and AI was “uniquely unsettling” to him, and described ads as a last-resort business model. In January 2026, OpenAI published its principles for advertising in ChatGPT, promising that ads would not affect the content of answers. The US test began on 9 February, with advertisers paying 60 USD per thousand impressions and facing a minimum budget of 200,000 USD. That is a premium price, far higher than typical social media advertising.

Then things accelerated. On 18 August, OpenAI announced the expansion of ads to 31 European countries, including Poland. On 31 August, the company said its advertising business had reached an annual revenue run rate of 1 billion USD - less than 200 days after launch. But investors were told, according to Axios, that the company’s 2026 target was 2.5 billion USD. The company that recoiled from advertising in 2024 now has a 2026 sales target it has yet to reach. That is the pressure Zoë Hitzig wrote about.

Hitzig spent two years at OpenAI working on how ChatGPT is built and priced. She left in February, as the advertising test began, and wrote in The New York Times that the company sits on “an unprecedented archive of human candour”: people confide in chatbots about health fears, relationship problems and faith. She did not claim that the first version of ads would break OpenAI’s promises. Her argument was that the business model would give the company strong incentives to break them in later versions. I find that more persuasive than any declaration that answers will remain independent, because it describes a mechanism, not intentions.

An answer has no neighbours

Search advertising has its faults, but it appears in an environment that can weaken its influence. A sponsored link sits above the results, with ten others below it. You can scroll, open three tabs and check for yourself. In a chat, you get one answer written in a fluent, confident voice. In 2025, Pew Research examined what happens when Google users see an AI-generated summary: they clicked a link on 8% of visits when a summary appeared, compared with 15% of visits without one. They clicked a source cited in the summary on just 1% of visits. A ready-made answer discourages comparison. The better it works, the less you have to check what you’ve been told against.

That is why I’m not reassured by a promise that an ad will sit below an answer rather than inside it. Suppose that is entirely true. You ask about exercise bikes, get an honest comparison of three models, and see a sponsored fourth one underneath. In a search engine, that fourth model would be one result among many. In a chat, it is the only thing on the screen besides the answer, attached to it and presented in the same interface voice you have just trusted. Even if the model doesn’t lie, the layout still has an effect.

And an ad does not have to stay below the answer. In August 2025, Musk told advertisers that if a user was trying to solve a problem with Grok, “advertising a specific solution would be ideal in that moment”. On 28 October, Google announced that it was bringing ads into follow-up questions in the same conversation in the US. It is also expanding AI Mode ads to markets including Canada, Australia and India, with the UK set to follow in the fourth quarter. Philipp Schindler acknowledged to investors that AI Mode earns less than traditional search. When a company reports both rising traffic and lower earnings from that traffic, it is clear which way it will try to push the product. I wrote about Google’s 500 billion USD stock-market loss in a separate article.

Meta goes further. From 16 December 2025, conversations with Meta AI have been used to personalise ads on Facebook and Instagram, without a full opt-out. The European Union, the UK and South Korea were excluded because of privacy rules. That leaves the EU in a particular position: OpenAI ads have arrived in Poland, while Google and Meta remain on the other side of a regulatory threshold. I have also written about less personalised advertising in Europe.

Free - but who pays?

The strongest argument on the other side is one I take seriously. In February 2026, OpenAI said ChatGPT had 900 million weekly users and 50 million paying subscribers. By my calculation, that means fewer than 6% pay. The other 94% use a model that costs a fortune to run in electricity and processors. A subscription is not an option for everyone. If advertising pays for people to have access to the same tool as a professional programmer, it is hard to call advertising nothing but bad. Hitzig herself noted that AI is expensive and ads could be an important source of revenue. Subscriptions, as a comparison of AI subscription prices illustrates, do not scale to a billion people.

I agree with the arithmetic, but not with the conclusion that ads are the only way. Perplexity, which had tested sponsored questions since autumn 2024, pulled back from advertising in February 2026, even though the ads were labelled and the company said they did not affect answers. The reason was simple: users start to suspect everything. On 4 February, Anthropic announced that Claude would remain ad-free and be supported by businesses and subscriptions. It immediately turned that position into a Super Bowl ad campaign, itself evidence that neutrality can be sold. Hitzig suggested other options: using revenue from enterprise contracts to subsidise the free plan, and independent oversight of data use. Free access could be funded by the business, rather than by the attention of its poorest users.

There is a third option, one OpenAI itself has shown in Europe. The company has also kept an ad-free free plan with a lower daily message limit and no image generation. That is more honest than pretending ads change nothing, because it names the trade-off: you pay with money, attention or limits. The problem is that people who cannot afford Plus may not be able to afford fewer answers either.

Who writes the answer?

Here is my argument: in search, the business model decides the order of results; in chat, it can shape the answer itself, even if nobody adds a word to it. Ads below an answer change the product’s goal. A chat that earns money from clicks will eventually be rewarded for questions that end in a purchase, rather than questions that leave users better informed and empty-handed. No statement of principles can stop that on its own. At most, it can be checked by regulators, ad-free competitors and users who notice the difference.

I would rather see people discuss chat ads as they discuss product placement on television: with a clear boundary and a clear account of who paid for what. Until then, I follow a simple rule: if something important is at stake - your health, money or a major purchase - ask the chat, then check two sources it did not choose for you. Someone at Perplexity, quoted by the Financial Times when the company pulled back from ads, put it well: users have to believe they are getting the best possible answer, or they will stop using the product and paying for it. The next test of that trust in Europe will come when Google expands AI Mode ads from the UK to the EU.