Generated by Codex with GPT-5

This summary covers The Economist’s April 25th, 2026 Business article listed in the contents as Chatbot ads and published under the headline The salesman in the machine.

The article argues that artificial-intelligence chatbots are becoming a new advertising medium, not just a new software interface. As more people use large language models for search, writing, coding, shopping and personal advice, the companies running those models are starting to insert sponsored messages into the conversation. That shift matters because it could reshape both the economics of AI and the structure of digital advertising.

For AI companies, ads offer a path to fund products that are expensive to run and often free to use. For advertisers, chatbots may become the next place to reach consumers if conventional search loses attention. The result is an early but important experiment: can a chatbot sell things without destroying the trust that makes users willing to talk to it in the first place?

Why AI Companies Want Ads

The article presents advertising as an obvious answer to a hard revenue problem. OpenAI has hundreds of millions of weekly ChatGPT users, but many of them do not pay for subscriptions. At the same time, running and improving frontier AI systems consumes huge amounts of cash. The article notes that OpenAI is expected to burn around \$25bn this year and far more next year. Ads would let it earn revenue from free users while keeping them inside the product.

Other companies have similar incentives, even when their balance sheets are stronger. Google is already testing ads in its AI search mode, Microsoft has put them into Copilot and Amazon lets brands sponsor replies in Rufus, its shopping assistant. Meta has not yet put ads directly inside its chatbot, but it is already using information from AI chats to improve targeting on Facebook and Instagram.

This is not just a scramble for incremental revenue. If users increasingly ask chatbots questions they once typed into Google, the search-ad business itself becomes vulnerable. Google has reason to bring ads into AI conversations before those conversations replace some of the queries that made search so profitable.

The article’s most interesting point is that chatbot advertising may work differently from search advertising. Search ads usually appear immediately, because a query often reveals intent in one line. A person searching for flights, mortgage rates or running shoes is already close to a commercial decision.

Chatbot conversations are more gradual. The user’s intent may emerge over several turns as they revise a document, compare options or ask follow-up questions. The article describes early evidence that Google and OpenAI are taking different approaches. Google’s AI ads mostly appear in response to the first query, much like traditional search. ChatGPT appears more willing to wait, often introducing ads later in the conversation after it has inferred what the user may want.

That creates a more personal kind of advertising. A chatbot can observe not only what someone asks, but how the conversation develops. It may know whether the user is anxious before a job interview, choosing a gift, debugging code, researching a medical symptom or drafting a complaint. That context could make ads more relevant than search ads. It could also make them feel more intrusive.

The early results are uneven. The article gives examples of ads that seem well matched to a user’s task, such as job-interview coaching offered during professional-writing help. But it also describes crude mistakes, such as treating a question about cryptographic private keys as a chance to sell physical safety boxes. The medium may be new, but the basic advertising problem remains: relevance is valuable only when the system understands the user’s actual intent.

The Trust And Measurement Problem

The article suggests that users have not yet rebelled against chatbot ads. Some rivals have avoided sponsored answers because they fear damaging trust, and Anthropic has mocked the idea of AI assistants suddenly becoming pitchmen. Yet early data cited in the article indicates that many users keep chatting after seeing an ad, and ad-supported conversations appear to last about as long as conversations without ads.

That is encouraging for chatbot-makers, but it does not solve the harder problem for advertisers. In search, measurement is relatively direct: users click an ad, land on a website and perhaps buy something. In a chatbot, users may keep talking instead of clicking. That makes attribution murkier. Brands will have to estimate whether seeing an ad inside a conversation later influenced a purchase somewhere else.

There is also a brand-safety risk. If advertisers allow AI systems to create personalized ad copy on the fly, the message may become more persuasive but also harder to control. A chatbot could make an exaggerated promise, invent a discount or place a brand next to sensitive conversation topics. The same flexibility that makes AI advertising attractive also makes it dangerous for companies that care about reputation and legal exposure.

The Takeaway

The article frames chatbot ads as the beginning of a new commercial layer inside AI products. The opportunity is clear: AI platforms need money, advertisers need new channels and conversational interfaces may offer richer context than search engines ever did.

But the trade-off is just as clear. Chatbots work because users treat them less like billboards and more like assistants. If ads are relevant, restrained and clearly marked, they may become a normal part of free AI services. If they are clumsy, manipulative or too intimate, they could corrode the trust that makes people use chatbots in the first place.

That makes the next phase of AI advertising a test of judgment as much as technology. The companies that win will not merely be the ones that insert ads into conversations. They will be the ones that learn when the conversation should not be interrupted at all.