ChatGPT Ads Gets Sponsored Agents for Brand Chatbots

Every ChatGPT ad used to end at a landing page. Now it can end in a live sales pitch. On September 16, 2026, OpenAI announced ChatGPT Sponsored Agents, a new ad format that lets a business run a conversational bot inside ChatGPT itself once someone clicks its ad. That changes what "running an ad" means for anyone who deploys AI on a company's behalf, because now the brand's chatbot is competing for the same chat window as the model doing the advising.

What ChatGPT Sponsored Agents actually do

Sponsored Agents are being tested with a small group of US advertisers, including Wayfair, according to OpenAI's own announcement and reporting from Quartz. Click a sponsored ad and instead of leaving ChatGPT, you drop into a separate, clearly labeled thread with the advertiser's agent. That conversation is walled off from both ChatGPT's own answers and the chat you were already having, so the model that was helping you plan dinner doesn't blur into the one trying to sell you a couch.

OpenAI shipped three companion pieces alongside it, aimed at the people building these campaigns rather than the people clicking them:

• A natural-language Ads Manager: type a prompt describing the campaign and the system drafts it, but a human has to confirm before anything goes live.
• HubSpot as the first CRM partner, so a business can build and track ChatGPT ad campaigns without leaving HubSpot.
• A free Shopify app, live since September 3 and expanding internationally since September 23, that syncs a merchant's product catalog and installs conversion tracking straight from the Shopify dashboard.

You: Build a campaign for our fall running shoe line targeting mobile
users 25-40 who searched "best trail running shoes."
Budget: $5,000/week. Optimize for purchases. Attribution: 14-day click.

Ads Manager: Draft ready — 3 ad groups, 6 headline variants,
Shopify catalog synced (214 SKUs). Confirm to launch?

That's the new workflow: describe the campaign, review a draft, confirm. No campaign-builder UI to learn, but also no visibility into the prompt or guardrails behind the agent's replies until you've already shipped it, per the rollout details covered by ppc.land.

What ChatGPT ads pricing actually looks like

OpenAI hasn't published a rate card for Sponsored Agents specifically, and access is still invite-only in the US test. What is public: ChatGPT Ads' conversion-optimization billing, priced on impressions, is now generally available, and attribution windows are configurable (7, 14, or 30-day click, 0 or 1-day view). That's the lever that will actually move your reported cost per acquisition once Sponsored Agents open up more broadly. Budget for this the way you'd budget any new paid channel: assume the efficiency pitch is unproven until your own attribution data says otherwise.

Why this matters if you build agents for a living

This is the same trust problem you already have with any customer-facing LLM agent, just wearing an ad badge. A Sponsored Agent is a GPT marketing chatbot speaking for the brand, running on a surface the brand doesn't control end to end. If you're the FDE who built that brand's support bot or product agent, OpenAI is now also operating a version of it, fed by your Shopify catalog and HubSpot data, inside somebody else's chat window and moderation stack.

Practically, that means the guardrails you'd put on any production agent now need to extend to a channel you can't instrument yourself: what happens when the Sponsored Agent misquotes a price, invents a return policy, or gets prompt-injected by a user probing its limits. You control the data going in. You don't control the logs, evals, or kill switch on the conversation coming out.

What to do Monday morning

If your company sells anything OpenAI would call a product, check whether you qualify for the US test or can request early access. If you're not in it yet, decide now whether a future Sponsored Agent gets the same fact-check pass as your own support bot before launch, not after some customer screenshots a wrong answer. And if you already run a standalone GPT-based brand bot, start comparing its answers against what OpenAI's agent says about you, because both may soon be live in the same app at once.

Getting the ad copy right was never the hard part of deploying a brand agent. Making sure it doesn't say something wrong to a paying customer, in a channel you don't fully own, is exactly the kind of production problem worth planning for before it's your incident to fix.