OpenAI starts testing ads inside ChatGPT

OpenAI is testing advertising that talks back. On 16 September the company said it is trialling “Sponsored Agents” in ChatGPT, a format where a user who clicks an ad can open a clearly labelled conversation with a business-sponsored agent, according to [1] and [2]. The test is limited to a set of selected advertisers in the United States, according to [1] and [2]. For a product that many people have come to treat as a neutral place to ask questions, that is a meaningful shift, and it is the reason this is worth reading closely rather than filing under routine product news.

What was announced

The centrepiece is Sponsored Agents. When a relevant ad appears in ChatGPT, a person can choose to start a separate, clearly labelled conversation with an agent that a business has sponsored, according to [1] and [2]. In that conversation they can spell out what matters to them and ask follow-up questions before clicking through to the advertiser’s site, according to [1]. OpenAI’s own illustration, reported in [1] and [2], is a shopper who sees a dining table in an ad and wants to know whether it fits their space, how many people it seats, or how to care for the finish — the kind of back-and-forth that a static banner cannot hold.

OpenAI frames the boundaries carefully. The sponsored conversation is kept distinct from ChatGPT’s independent answers and separate from the user’s original chat, according to [2]. In other words, the pitch is supposed to live in its own lane rather than bleed into the assistant’s normal replies. Whether that separation holds in practice is exactly what a careful reader should watch as the test widens.

Around the agents, OpenAI is also opening a set of tools for the people who buy ads. HubSpot is named as its first CRM partner and Shopify as its first ecommerce partner, according to [1] and [2]. From 16 September, businesses that manage customers in HubSpot can connect a ChatGPT Ads account and create ads, monitor performance and follow up on leads without leaving HubSpot, according to [1]. The Shopify app is free; it is available to US merchants from 16 September and reaches other markets where ChatGPT Ads runs from 23 September, according to [2].

There is a third piece aimed at the ad copy itself. An opt-in, AI-powered text feature can adjust an advertiser’s existing headlines and descriptions so they fit the context of a conversation, and it automatically translates ad copy into the reader’s preferred language, according to [1] and [2]. Advertisers can also use natural-language prompts to create, update and analyse campaigns, with the system suggesting copy and imagery drawn from a landing page and the campaign’s stated objective, according to [1]. Early advertisers named in the test include Newegg, Best Buy, Lowe’s and VistaPrint, according to [1].

What changed, and for whom

For most people the practical change is small today and larger in principle. Sponsored Agents are a test, not a general rollout, so a typical ChatGPT user in the US may or may not see one, and users elsewhere will not see them yet, according to [1] and [2]. But the direction is clear: advertising is moving from the edges of OpenAI’s business into the assistant itself, and it is doing so as a conversation rather than a box in the corner. The thing to internalise is that a chat which looks like a helpful agent may be a paid one — which is why the label OpenAI attaches to it is the single most important detail.

On data, OpenAI’s stated position is that ads are clearly identified, kept distinct from ChatGPT’s answers, and that people control how their data is used for ads, according to [1] and [2]. Securities.io reports a sharper detail: advertisers receive only aggregate performance information, such as numbers of views or clicks, rather than access to the chats themselves, according to [2]. That is a reasonable-sounding boundary, but it is a claim from the company about its own system, not an independently verified guarantee, and it should be read that way until outside testing exists.

For small businesses the calculus is different. OpenAI is deliberately meeting advertisers where they already work — inside HubSpot and Shopify — which lowers the effort of trying it, according to [1] and [2]. That convenience cuts both ways: it is easy to start, and easy to spend on a format whose reach and returns are not yet public. Nothing in the announcement discloses pricing, audience size or performance benchmarks, according to [1] and [2], so any early budget is a bet on a test.

Who it is for — and not

This matters most to two groups. The first is heavy ChatGPT users who lean on it for research and shopping decisions; for them the useful habit is simply to notice the label and keep sponsored answers at arm’s length. The second is operators of small and mid-sized businesses who already run performance marketing and have a HubSpot or Shopify account; for them it is a channel worth understanding now and testing cautiously later.

It is less relevant, for the moment, to anyone outside the United States, since the agent test is US-only, and to businesses without an existing HubSpot or Shopify setup, since the launch integrations are built around those two, according to [1] and [2]. It is also not a reason to change tools: none of this alters how ChatGPT answers ordinary questions, and OpenAI’s stated design keeps the ad lane separate, according to [2].

The honest summary is that OpenAI has crossed a line it had been approaching all year. Ads inside the assistant were widely expected; a conversational, agent-shaped ad that you can question is a more novel thing, and it puts more weight than usual on disclosure. The test is small and the guardrails are the company’s own. This page will update as the test widens, as pricing or reach becomes public, or as independent testing checks OpenAI’s claims about separation and data.

∴ Ads are arriving in the assistant people treat as neutral; the label, not the pitch, is now the thing to check.