OpenAI Puts Talking Ads Inside ChatGPT
3 min readOpenAI started testing Sponsored Agents on September 16, letting advertisers drop a bot into ChatGPT that users can hold an actual conversation with. Ads inside the assistant are no longer something you scroll past. They answer back.
How ChatGPT Ads Worked Before
OpenAI switched on advertising in ChatGPT in February 2026, and the first version looked familiar. A sponsored placement surfaced alongside a relevant answer, and interested users clicked through to the advertiser. It was a conventional format bolted onto an unconventional product, and it left the most valuable part of the assistant, the conversation itself, entirely off limits to brands.
What OpenAI Announced
Sponsored Agents change that. When a relevant ad appears, a user can choose to open a clearly labeled chat with a business-sponsored agent, ask follow-up questions, spell out what they actually need, and click through to the company website when they are ready. OpenAI says those conversations stay separate from the user original thread and from ChatGPT independent answers, a firewall the company is leaning on hard.
The same release handed advertisers a set of natural-language tools. Through an Ads Manager plugin in ChatGPT Work, a marketer can create, update and analyze campaigns by describing them in plain English, and OpenAI will suggest ad copy and imagery pulled from a landing page and a stated objective. The company also named HubSpot as its first CRM partner and Shopify as its first ecommerce partner. HubSpot customers can connect ChatGPT Ads accounts as of September 16, and US Shopify merchants can install the new ChatGPT Ads app the same day.
Sponsored Agents themselves are limited for now to a test group of advertisers in the United States.
Why It Matters
Advertising is the most reliable way OpenAI has to earn money from the hundreds of millions of people who will never pay for a subscription, and a conversation converts far better than a banner ever did. A shopper who can interrogate a brand for three minutes inside the tool they already trust is worth a great deal more than a click.
The open question is that trust. ChatGPT value rests on users believing the answer is not for sale, and the labeling and separation OpenAI built here will be tested the first time a sponsored agent gives advice that sounds like the assistant own voice.
Watch for whether the test widens beyond the United States, and whether Google responds with something similar inside its own assistant.