OpenAI is testing a new ChatGPT Ads campaign type called Agent that replaces the destination URL with a live Business Agent conversation running on the advertiser’s own site or infrastructure. The find was spotted in the ads manager interface by Juozas Kaziukėnas, who posted screenshots to LinkedIn, and reported by Barry Schwartz for Search Engine Roundtable on July 31. The change touches the most basic unit of a paid search ad: what happens after the click.
The mechanics described by Kaziukėnas follow a three-step build. OpenAI first scrapes the advertiser’s website to construct a business profile covering common customer questions, support information, and general context. The advertiser then configures a Business Agent with defined instructions that draws on that profile, connects to product feeds and MCP tools for live data, and can capture leads through custom forms. The resulting ad campaign points to that agent rather than to a landing page. Kaziukėnas wrote that Agent is already “in place in the ads manager for select advertisers,” running on the same underlying technology as OpenAI’s custom GPTs.
The significance sits below the product description. For three decades, the unit of paid search has been a click that lands on a page the advertiser built and controls. Agent removes that page and substitutes a conversation. Everything that depends on a landing page loses its footing: there is no scroll depth to measure, no form placement to test, no version of the page to run against another in a conventional landing-page experiment. Conversion tracking has to attach to a dialogue rather than a URL, which means the object being optimized stops being a layout and becomes the agent’s behavior, turn by turn.
That shift raises questions the source material does not answer. Who writes the agent’s instructions, the advertiser or OpenAI’s scraped business profile? What is the agent permitted to say on the advertiser’s behalf once a conversation drifts off-script, and who is accountable if it misstates a price or a policy? Most practically, how would a performance marketer A/B test a conversation the way they test a headline or a call-to-action button today? None of these have been addressed in what Kaziukėnas or Schwartz have published.
Business agents are not a new concept in search advertising. Kaziukėnas noted that Google has announced comparable agent-driven ad formats before, without describing how that comparison plays out for advertisers weighing either platform. What is new here is that OpenAI’s version is surfacing inside the standard ChatGPT Ads Manager alongside the existing standard and product-feed campaign types, rather than as a separate product.
The status of Agent should be read narrowly. This is an early-stage test visible to select advertisers, single-sourced from OpenAI’s own product surface via Search Engine Roundtable’s reporting. Kaziukėnas said he has not seen a live Agent ad or its resulting user experience, and Schwartz reported he does not see the option in his own account. OpenAI has not published a rollout timeline, eligibility criteria, or pricing for the format.
Search teams that manage ChatGPT ad budgets should not build measurement plans around Agent yet. The more useful move for the next ninety days is auditing what a Business Agent’s scraped profile would say about your business today, since that profile, not a landing page, may become the surface a prospect actually interacts with first.
Barry Schwartz reported the Agent campaign type for Search Engine Roundtable on July 31, 2026, citing a LinkedIn post by Juozas Kaziukėnas.
