OpenAI's newest experiment in chatbot-native advertising ends somewhere no ad has ended in thirty years: not on the advertiser's website, but inside a ChatGPT conversation built from it. The click survives. The destination the digital economy was built around does not.

  • OpenAI is testing an Agent campaign type in ChatGPT Ads Manager where clicking an ad opens a business-specific AI agent instead of a landing page.
  • The agent is assembled from a crawl of the advertiser's own website, plus product feeds, MCP tools for live data, and lead-capture forms.
  • The performance case is real: Criteo data from 2026 shows LLM referrals converting at roughly 1.5x other digital channels. The strategic cost is that the post-click relationship, data, and destination move onto OpenAI's surface.
  • Publishers should read this as confirmation of the trajectory: the platform intends to keep the entire session, and the durable revenue model is being paid as the source that feeds it.

What chatbot-native advertising is

Chatbot-native advertising is an ad format in which the destination is a conversation with an AI agent rather than a webpage. The ad still gets clicked, but the click opens a platform-hosted dialogue that can answer questions, recommend products, book appointments, and qualify leads before the user ever reaches the advertiser's site, if they ever do.

It is the logical endpoint of a shift already underway: AI Search removed the click from discovery, and this format removes the website from conversion.

The landing page economy vs the agent destination economy

In the open-web economy, the funnel had a fixed shape: ad, click, your website. The landing page was the one asset the advertiser fully owned, along with its analytics, its CRM capture, and its optimization learnings. Publishers were funded as the surfaces those journeys crossed on the way.

In the agentic-web economy, the funnel folds into the platform: ad, click, a hosted agent. The advertiser configures the destination but no longer owns it. Both sides of the open web become raw material: publisher content is ingested to build the answer, and now the brand's own website is crawled to build the storefront. Nobody needs to visit either.

What MarTech and Search Engine Land reported, and what we verified

MarTech's Anu Adegbola reported on August 4, 2026 that OpenAI is laying groundwork for ads whose destination is a business agent, following Search Engine Land's July 31 report and Barry Schwartz's Search Engine Roundtable findings. The workflow has three steps: ChatGPT crawls the business's site to generate a profile, the advertiser configures an agent with custom instructions, product feeds, MCP tools, and lead forms, and the campaign then points to that agent instead of a URL.

One caution we verified across all three sources: the capability sits in Ads Manager for a select group of advertisers, and the end-user experience has not been publicly observed. OpenAI has announced nothing. This is infrastructure spotted in the wild, directional but not yet a shipped product, and we treat it as such.

Why higher conversion is the wrong scoreboard

The performance case for agent destinations

The strongest version of the optimist case deserves stating plainly. Conversational destinations cut friction: instant answers, personalized recommendations, lead capture in one surface. Criteo's 2026 retailer data shows LLM referrals converting at about 1.5x other channels, and OpenAI shipped a full measurement stack in months, with CPC bidding and a Conversions API on May 5 and conversion-optimized campaigns from June 5. Precedent supports it too: advertisers happily traded owned destinations for Facebook Lead Ads a decade ago when the numbers worked.

What the advertiser hands over with the click

This trade is categorically larger than lead ads. OpenAI's own ads documentation states that chats, names, emails, and user data stay inside ChatGPT; advertisers receive outcomes, not conversations. The landing page was the last owned asset in the paid funnel. Replace it with a hosted agent and the customer relationship, the intent exhaust, and the optimization learnings all accrue to the platform. Google rented you the shelf but never owned your store. In this model, the platform hosts the store, and it is also the landlord, the till, and the market researcher.

What this means for brands and for publishers

For CMOs, media buyers, and agencies: test the agent, keep the asset

Test the format when it opens; refusing a 1.5x-converting surface is not a strategy. But ringfence it: keep first-party capture inside the agent via your own forms and MCP endpoints, negotiate for conversation-level insight before scaling spend, and budget for a new discipline, agent configuration, that will matter more than landing page CRO within two years.

For publishers: the session is closing, price the source

If even the advertiser's destination is being absorbed into the assistant, referral recovery is a dead thesis. The platform's design goal is a session that never leaves. The revenue that survives is being paid as the source AI engines cite and ingest: citation-based advertising, with publishers inside the value chain rather than beneath it.

Three signals that show where the funnel is heading

First, velocity: OpenAI went from a $200,000-minimum managed pilot on February 9, 2026 to open self-serve with no minimum on May 5, then CPA bidding by June 5. A company moving that fast does not build agent infrastructure for fun. Second, ambition: OpenAI projects $2.5 billion in 2026 ad revenue per figures reported by Digiday and The Information, a target that requires owning more of the funnel than a text unit under an answer. Third, competitive gravity: watch whether Google mirrors the format inside AI Mode, because when both duopolists host the destination, the open web's last funded reason to be visited disappears.

Conclusion

Hold on to this: when the conversation becomes the destination, the advertiser's website and the publisher's page stop being visited for the same reason, and only the platform monetizes the session. Chatbot-native advertising makes that enclosure explicit. Smalk AI exists to build the counterweight: Generative Engine Advertising, native ads for AI agents that give brands visibility inside AI answers while opening a revenue stream for the publishers whose content those answers and agents are built from, a two-sided rail rather than a one-sided enclosure. What to watch next: whether the Agent format ships publicly and whether Google answers with a hosted destination of its own; that is the moment the post-click open web either gets a payment rail or gets archived.