There's an assumption baked into every ad campaign you've ever run: when someone clicks, they go somewhere. A product page, a landing page, a checkout flow. The entire job of CRO, page speed optimization, and above-the-fold design exists because of that assumption.
OpenAI is testing a campaign type that removes it.
The "Agent" campaign type
Reported by Search Engine Land this week, OpenAI has been testing a new campaign type inside ChatGPT Ads Manager named Agent. Unlike standard campaigns — which point to a destination URL — the Agent campaign type points to a Business Agent conversation.
When a user clicks the ad, they don't go to your website. They enter a ChatGPT conversation with an AI agent configured by the advertiser.
The format is currently available to a limited group of advertisers. No public launch date has been announced.
How OpenAI builds the agent
The setup runs in three stages, per Search Engine Land.
First, OpenAI scrapes the advertiser's website to build a business profile. The system is trying to understand what the business does, what questions customers commonly ask, and what the brand's support context looks like.
Second, the advertiser configures the agent with custom instructions. Those instructions sit on top of the profile OpenAI built from the website. Advertisers can connect additional data sources — product feeds, Model Context Protocol (MCP) tools for live inventory or pricing, and lead generation forms.
Third, the campaign points users into that agent instead of a URL. Targeting and budget are set the same as any other campaign. The difference is the click destination: not a website, a conversation.
The agent can answer product questions, surface catalog recommendations, and collect lead information — all inside the ChatGPT interface, before a user ever visits your domain.
What this breaks in your current setup
If you're running ChatGPT Ads campaigns today, you're sending clicks to landing pages and measuring sessions, time on site, and conversion events from the pixel. The Agent format doesn't generate any of that. There's no page visit, no session, no bounce rate. The conversion happens in the conversation.
That's not a small change. It's a different measurement model, a different creative brief, and a different funnel structure.
The creative input isn't a headline and description pointing at a landing page. It's the quality of your website content (because OpenAI scrapes it to build the business profile), the completeness of your product feed (because that feeds the agent's catalog), the MCP tools you connect (because those supply live inventory and pricing data), and the custom instructions you write (because those shape how the agent handles objections and surfaces the right products).
Every dollar you've spent optimizing conversion rates on landing pages does not transfer to this format.
Who this actually works for
The Agent format has an obvious fit for categories where pre-purchase questions are high and the decision is considered — complex product specs, custom configurations, service businesses where lead capture involves back-and-forth.
For DTC brands selling sub-$40 items where the purchase decision takes two seconds, there's less upside. A conversation adds friction before a decision that's already made. That friction needs to earn its place.
The brands best positioned to test it are the ones with clean product data, structured website content that makes sense when scraped, and clear answers to the five questions their customers ask before buying. If your website is thin pages and fragmented copy, the agent OpenAI builds from it will reflect that.
The test window question
OpenAI has moved fast with ChatGPT Ads. The platform went from CPM-only to CPC, then geo targeting, then oCPC and mobile attribution integrations — in under six months. The Agent campaign type follows that pattern: a new capability in limited testing that will likely open more broadly as the Ads Manager matures.
Whether the Agent format becomes a mainstream placement or stays niche depends on advertiser results, which OpenAI hasn't published. The performance of conversation-destination ads against URL-destination ads hasn't been reported publicly. Early access exists; data does not.
What makes sense to do now regardless: ensure your website copy is structured and current, since OpenAI pulls from it to build the agent profile. Ensure your product feed is complete — titles, descriptions, prices, categories. Both inputs determine the quality of an agent you'd run in this format, and both have value well beyond ChatGPT.
If you want to see how your current measurement setup handles AI channel traffic and where the tracking gaps are, the free audit at Gromerce shows you in three minutes.
The click-to-website assumption has held since the first banner ad. It was never a law.
Sources: Search Engine Land, OpenAI, August 2026
