← BLOG
Industry7 min

OpenAI's Sponsored Agents Put a Salesperson Inside Every ChatGPT Conversation

SolidAtoms Team
OCT 10, 2026
OpenAI's Sponsored Agents Put a Salesperson Inside Every ChatGPT Conversation

On September 16, 2026, OpenAI quietly changed what an ad inside ChatGPT can do. It stopped being a static link you can ignore and started being something you can talk to. The feature is called Sponsored Agents, and it's the clearest sign yet of where OpenAI thinks ChatGPT's business model is headed — and a useful test case for how much trust people are willing to lend a chatbot that's also trying to sell them something.

What a Sponsored Agent actually does

The mechanics are simple, at least on the surface. When a relevant ad surfaces in a ChatGPT session, the user can choose to open a separate, clearly labeled conversation with a business-controlled AI agent — one built and tuned by the advertiser, not OpenAI. That agent can answer follow-up questions, walk through options, and hand the user off to the merchant's own site when they're ready to buy.

The two launch pilots make the strategy legible: Wayfair (furniture shopping) and Angi (home services). Both are categories where a customer typically needs to describe a messy, specific problem — "my couch doesn't fit through this doorway," "my faucet won't stop dripping" — before they're ready to transact. Angi's pitch is that a homeowner can go from describing a leaky faucet to booking a local contractor without ever leaving the chat window, in a session that's separate from, and more promotional than, the free Angi assistant some users already add to ChatGPT.

This has been building for months

Sponsored Agents didn't come out of nowhere. Fidji Simo, OpenAI's CEO of Applications — who previously spent nearly four years running Instacart, where she built its advertising business into nearly $1 billion in annual revenue — published OpenAI's advertising principles back in January 2026 after roundtable discussions with close to 100 employees about guardrails. Ads went live for U.S. Free and Go tier users on February 9, 2026, with an initial partner roster that included Target, Adobe, Williams-Sonoma, and Albertsons. A self-serve ad platform followed on May 6, removing budget minimums so smaller businesses could buy in. Sponsored Agents is the next rung on that ladder: not just a labeled link, but a full conversational sales rep licensed to operate inside the product people use to ask for honest advice.

OpenAI's own framing leans hard on Google's playbook rather than Meta's. The company has said ads are always separated and clearly labeled, that conversations with ChatGPT itself stay private from advertisers, and that user data isn't sold. Simo has also said ads will likely remain "a minority" of OpenAI's revenue for the foreseeable future — this is an intent-capture play, closer to search ads than a feed algorithm optimizing for engagement.

Why this is a harder problem than a banner ad

The uncomfortable part isn't the ad — it's the format. A banner ad looks like a banner ad. A Sponsored Agent looks and sounds exactly like the assistant you were just talking to, because it's built on the same conversational interface. That distinction matters more here than it did on the open web: academic research on conversational advertising has found that people consistently struggle to identify embedded ad content when it arrives in natural, back-and-forth dialogue rather than a visually distinct unit. Labeling helps, but a label competing against a fluent, helpful-sounding paragraph is a much weaker signal than a box with "Sponsored" stamped across the top.

"Protecting the trust people place in ChatGPT remains our North Star" — OpenAI, in its framing of the advertising rollout.

That trust question got harder to wave away the same week Sponsored Agents launched. Reporting from 404 Media, based on leaked internal documents, described a separate OpenAI program — internally called Project Lily — in which contractors review full ChatGPT conversations and user-memory summaries as part of product-improvement work. None of that is the same system as Sponsored Agents, and OpenAI has been explicit that advertisers don't see private conversations. But it lands at the same moment users are being asked to accept a chat window that sometimes represents them and sometimes represents a merchant, and to tell the difference in real time.

The advertising industry's early read is lukewarm rather than hostile. Coverage of agency reaction has described media buyers as "underwhelmed," citing skepticism about whether OpenAI's in-platform click metrics line up with what advertisers can independently verify — a familiar complaint from the early days of any new ad platform, but one that carries more weight when the unit being sold is a novel, hard-to-audit AI conversation rather than an impression count.

The part worth watching

Every major assistant vendor is watching this pilot, because every one of them will eventually face the same fork: subscriptions alone likely can't fund frontier-model inference at consumer scale forever, and the two obvious alternatives — ads or selling access to user data — both cut directly against the "trusted advisor" positioning that makes people rely on these tools in the first place. OpenAI is betting it can thread that needle by keeping the ad surface small, clearly bounded, and opt-in, and by choosing categories (home repair, furniture) where a sales conversation is genuinely close to what the user wanted anyway.

Whether that holds depends less on OpenAI's policy language than on whether ordinary users can reliably tell, mid-conversation, when they've been handed off to a salesperson wearing the assistant's voice. That's an interface and trust problem as much as an advertising one — and it's the one worth tracking as Sponsored Agents rolls out beyond its first two pilots.

Sources