The OpenAI ad format that ends the click

OpenAI is quietly building an ad format inside ChatGPT that replaces the destination website with a business-specific AI agent. Click the ad, and instead of landing on a page, you land in a conversation. The conversation is trained on the business, wired into their product feed, hooked up to live data via MCP tools, and configured to answer questions, recommend products, and capture leads.

That's not a new ad unit. That's a new economic model for the open web.

For thirty years, the click has been the atomic unit of digital advertising. Every measurement system, every attribution model, every landing-page test, every conversion rate optimisation cottage industry — all of it exists because the click leads somewhere, and that somewhere is a page you own. OpenAI is proposing that the somewhere is a conversation they own. Your website becomes a data source. The agent becomes the storefront.

Most of the coverage I've seen frames this as an interesting product experiment. It isn't. It's the logical endpoint of a shift that's been happening in slow motion since AI Overviews launched, and if it works, it changes what "digital advertising" actually means.

What OpenAI is actually building

The workflow, based on what's been spotted in ChatGPT Ads Manager, has three parts.

First, ChatGPT crawls the advertiser's site and auto-generates a business profile — common customer questions, support information, product context. Second, the advertiser configures a business agent, layering custom instructions on top of that profile, plus product feeds, MCP tools for live data, and lead-capture forms. Third, campaigns point users into a conversation with the agent rather than to a landing page.

The technical foundation appears to be the same one that powers Custom GPTs, which means the plumbing exists and works. The novel bit is the ad slot pointing at it.

Nobody has seen the end-user experience widely yet. But the infrastructure tells you enough to understand the direction. OpenAI wants the click to end inside ChatGPT.

Why this is bigger than an ad format

The click going away isn't a UX change. It's a distribution change.

The click going away isn't a UX change. It's a distribution change.

If the destination is your website, you control the pixel, the copy, the follow-up email flow, the re-marketing cookie, the analytics, the checkout, and the customer relationship. You own the surface where conversion happens.

If the destination is a ChatGPT conversation with an agent OpenAI hosts, you own none of that. OpenAI owns the interface, the session data, the conversation history, and — crucially — the ability to change the terms whenever they want. Your product feed is an input. Your agent instructions are a configuration file on someone else's platform. The relationship with the customer is mediated by a system you don't run.

The click going away isn't a UX change. It's a distribution change.

We've seen this pattern before. Facebook Pages in 2012. Amazon in 2016. Apple's App Store since day one. Every time a platform inserts itself between a business and its customer, the platform's leverage grows and the business's shrinks. The business gets efficiency in exchange for dependency. Usually the efficiency is real. Usually the dependency is worse than it looked at the start.

What this does to SEO, GEO and the rest of the acronyms

If ad clicks stop landing on your site, organic clicks are next.

That's not speculation. It's already happening in miniature. AI Overviews absorb the answer, the citation is a tiny link most people don't take, and the click never happens. iPullRank's recent 13-billion-search analysis put the aggregate zero-click rate near half of all Google searches. The direction of travel is clear.

What OpenAI's chatbot-native ads formalise is this: the destination is optional. If the AI system can answer the question, transact the purchase, capture the lead, and schedule the meeting without ever sending the user to your site, then your site's job changes. It's no longer where the customer arrives. It's where the AI system extracts data.

The GEO discourse hasn't caught up with this. Most GEO advice is still framed around "how do I get cited so users click through." That framing assumes the click matters. In an agent-mediated model, the click either doesn't exist or points at a conversation on the platform's turf. Optimising for citations is optimising for a step in a funnel that increasingly ends elsewhere.

The honest reframing: you're not optimising for clicks anymore. You're optimising to be the data source AI systems draw from when they answer, recommend, and — now — transact.

The feed is the new landing page

Notice what's in OpenAI's business-agent configuration. Product feeds. MCP tools for live business data. Custom lead-generation forms. Auto-generated business profiles from crawling your site.

None of that is content in the traditional SEO sense. It's structured data, real-time inventory, and machine-readable business information. If you want to show up well in this system, you need clean product feeds, accurate structured data, working MCP endpoints, and a site that a crawler can extract clean facts from.

This is what I've been arguing for a while: the fundamentals that made you findable in Google search are the same fundamentals that make you usable by AI agents. Speed, semantic HTML, accessible markup, schema, clean information architecture. The businesses that invested in structural quality over the last five years are going to be reusable across a dozen new discovery surfaces. The businesses that treated their website as a marketing brochure are about to find out that a marketing brochure isn't a data source.

The businesses that treated their website as a marketing brochure are about to find out that a marketing brochure isn't a data source.

The uncomfortable implication for a lot of in-house teams: your content team may have less leverage than your data team from here on. Product feed accuracy, inventory-sync reliability, and schema hygiene are becoming higher-stakes than another blog post. The feed-quality lever has been quietly outweighing the content lever for a while. This just makes it obvious.

The measurement problem gets worse

If clicks stop landing on your site, how do you measure whether any of this works?

You can't. Not with existing tools. Google Analytics measures sessions that begin with a landing page. Attribution models trace back from a conversion event that happens on your domain. Ad platforms report clicks that leave their platform and go to yours.

An agent-mediated ad doesn't fit any of that. The "click" is a session start inside ChatGPT. The "conversion" might be a lead form submitted to OpenAI, forwarded to you as an email. There's no pixel. There's no cookie. There's no referrer. You know a lead came in. You don't know what conversation produced it, what the user asked, or what the agent said in response. OpenAI does.

That means the platform holds the analytics leverage as well as the distribution leverage. Whatever reporting they choose to expose is what you see. Whatever they choose to withhold, you don't. If this reminds you of every walled-garden ad platform ever built, that's because it is one, with an extra layer of opacity.

I've written before about the AI search measurement trap — teams budgeting off metrics they don't fully understand from surfaces they don't control. This makes it worse, because at least AI Overviews still deliver a click you can attribute if it happens. Agent-mediated advertising doesn't even give you that.

What to actually do about it

Not much yet. This is early. The end-user experience hasn't been widely observed. The format may take a year to become meaningful, or it may quietly fizzle when advertisers realise they don't want to hand OpenAI their customer relationship. Both are plausible.

But the strategic direction is clear enough to make three decisions now.

First, treat your structured data and product feeds as customer-facing surfaces, not backend infrastructure. Anywhere a machine reads your business — feed accuracy, schema markup, MCP endpoints if you have them, business profile data on every major platform — is now a distribution surface. Audit it like one.

Second, build the parts of your customer relationship that a platform can't intermediate. Email lists. Direct relationships. Branded search. Communities. The businesses that survive platform shifts are the ones that own something the platform can't easily replicate. If everything you have lives on someone else's surface, you don't have a business; you have a channel dependency.

Third, get honest about the strategic question of whether to participate at all. If OpenAI's ad format launches at scale, some businesses will win big by getting in early. Others will accelerate their own commoditisation by handing over the customer relationship for a short-term efficiency gain. Which one you are depends on your category, your margins, and your alternative distribution. Nobody can answer that for you, but it's a real question that most marketing teams aren't equipped to ask.

The click made the open web work. Businesses paid platforms for attention, platforms sent users to businesses, and everyone owned their piece of the transaction. The agent-mediated model breaks that. The platform sends users to itself, and the business becomes a data feed.

That's the loop. And we're about to build it.