OpenAI just fused ChatGPT's ad business into its enterprise platform and quietly reserved the right to decide who's allowed to compete for attention on it. Here's what your team should actually do about it.

———

You saw the OpenAI news this week and felt two things at once. First: I need to be in this. Everyone with a media budget is watching ChatGPT turn into an ad platform in real time, and the fear of missing an early-mover advantage is real. The second, quieter, feeling underneath: I don't actually know what I'm walking into.

That second feeling is correct. Trust it.

This week alone, OpenAI announced Sponsored Agents — conversational ads that let people chat with a business's product directly inside ChatGPT. It signed Amazon as an ad tech partner. It shipped an ad manager built into the ChatGPT interface itself, so campaigns get planned, created and optimized without a human ever leaving the chat window. And, days before any of that made headlines, it quietly rewrote its advertising policy to give itself the right to reject any advertiser that conflicts with its own "business interests" or "competitive position."

Most of the coverage this week is about the first three. We think the fourth one is the real story — and almost nobody in growth marketing is talking about it the way they should be.

If you run paid media for a business — whether that's a founder watching CAC creep up every quarter or an operator deciding where next year's budget goes — this matters to you specifically. Not because you need to launch a ChatGPT ads campaign by Friday. Because the platform you're being invited onto just told you, in writing, how it thinks about power. And that tells you more about how to use it than any case study will.

What Actually Happened This Week

If we strip away the announcement noise, here's the shape of things.

OpenAI launched ads on ChatGPT in February 2026, and for months the read on the industry was that advertising was a side project — a monetization experiment running alongside the real business, which was selling enterprise subscriptions and API access. That read is now wrong, and OpenAI just proved it wrong on purpose.

This week, ChatGPT rolled out Sponsored Agents: ads that don't just show up as a card at the bottom of a response, they let a person open a live conversation with a business inside the chat itself. Wayfair was one of the first to test it, with a "chat with us" button that opens a branded exchange instead of sending someone to a landing page. At the same time, OpenAI shipped a plugin connecting its ad manager to ChatGPT Work — the enterprise product used by the exact people who approve ad budgets. Campaign creation and management now live inside ChatGPT itself, so a marketer can build, launch and adjust a campaign through the same interface they use for everything else.

Layer on the Amazon news. Amazon Ads now lets marketers reserve ChatGPT inventory, initially through managed service and eventually through Amazon's own demand-side platform. That's not a small integration. Amazon is, by scale, one of the two or three most consequential forces in digital advertising. Its willingness to build ChatGPT into its stack is a signal to every performance marketer watching from the sidelines: this channel is being taken seriously by the people whose job is to know where the money should go.

None of this happened by accident, and none of it happened slowly. OpenAI's ad product went from a basic impressions format in February to click-based buying, conversion optimization, first-party audience targeting and product feeds within months — the same feature set that took Meta and Google years to build. One ad-tech analyst described it as the most aggressive ad platform rollout they'd ever watched. That's not hyperbole; it's the timeline.

Here's the part worth sitting with. OpenAI's own finance chief has framed ChatGPT ads as combining the intent signal of search with the contextual richness of social — shorthand for something structurally new: a channel where discovery, research and purchase decision can all happen inside a single conversation, with no click-through required to move someone from curiosity to commitment.

That's the opportunity everyone's excited about. It's also, as we'll get into, the thing that should worry you.

We've Watched This Movie Before

Here's where our contrarian take starts, and it's not about whether ChatGPT ads work. Early data says they might: Capital One, Expedia and Cloudflare are already spending real money there, and Booking Holdings — one of OpenAI's own enterprise partners — is buying inventory too. The format works. That's not in dispute.

What we don't trust is the direction of the "agentic" push — OpenAI encouraging marketers to build, launch and optimize campaigns entirely inside ChatGPT, with AI handling bidding, targeting and creative generation on their behalf.

We've run this exact playbook before, just wearing a different logo.

Meta built Advantage+. Google built Performance Max. Both promised the same thing OpenAI is promising now: hand us your targeting and bidding decisions, let the model optimize, and you'll get better results with less work. And for a lot of businesses, in a lot of quarters, that promise delivered — right up until it didn't, and nobody could tell you why, because the black box doesn't explain itself. It just reports back a number and asks for more budget.

We had a client — a founder-led e-commerce business in a competitive category — who handed almost all of their Meta spend to Advantage+ campaigns because the case studies were good and their internal team was stretched thin. For two quarters, blended ROAS looked fine. Then it didn't, and there was no way to isolate why: which audience had gone stale, which creative had fatigued, which placement had stopped converting. The system had made all of those decisions invisibly, and it kept making them the same way even as performance degraded, because "the way that used to work" was still what it was optimizing toward. We had to rebuild manual audience segmentation and creative testing structure from scratch, underneath the automation, just to get visibility back.

That's the pattern. Automation that removes your ability to see and adjust the individual levers isn't actually saving you time — it's deferring a harder rebuild to later, at the exact moment your patience for uncertainty is lowest, usually when performance has already slipped.

OpenAI is running the identical move, just earlier and faster than Meta or Google did, because it's watched them do it and knows the model works: get marketers dependent on the platform's own interface for decision-making, and you own not just the auction, but the workflow. Every campaign built inside ChatGPT's ad manager is a campaign whose logic lives entirely on OpenAI's side of the relationship. You get outputs. You don't get the reasoning.

None of this means don't test Sponsored Agents or the new ad manager. It means don't confuse "the platform makes this easy" with "the platform is optimizing for you." Those are different claims, and only one of them is true. The platform optimizes for the platform. It always has. The businesses that do well on any new ad channel are the ones who remember that on day one, not the ones who relearn it during a bad quarter.

The Part Nobody's Talking About: OpenAI Is Also the Referee

Buried under the product announcements this week was a policy change that matters more than any of them.

OpenAI updated its advertising rules to explicitly reserve the right to reject advertisers whose products conflict with its "business interests" or "competitive position." In practice, that's already happening: Adobe — currently one of ChatGPT's top twenty advertisers by spend — has reportedly been blocked from promoting products that compete with OpenAI's own.

Sit with what that actually means. OpenAI is not a neutral pipe that happens to sell ad space, the way a highway billboard company doesn't care what you're selling as long as you pay. It is a company racing into coding tools, image generation, voice products and enterprise software — the same categories a growing number of its advertisers operate in. It is simultaneously your ad platform, your infrastructure vendor in some cases, and, increasingly, your competitor. It has now written into policy that when those roles conflict, it gets to decide who loses access to its audience.

This isn't a hypothetical governance question. It's a live business risk with a name and a date attached to it.

Every ad platform has rules about what you can advertise — nobody's arguing OpenAI should run unlimited unsafe ads. That's not the concern. The concern is a rule written broadly enough to cover "competitive position," a category that can expand to include your business the moment OpenAI decides to enter your market. Today that's mainly true for AI-adjacent software companies like Adobe. There's no reason to assume it stays limited to that category as OpenAI's own product surface keeps growing — it's already moving into coding, images, voice and enterprise software, and none of those categories have fixed edges.

Compare this to Meta and Google. Both are deeply flawed as advertising partners — the auction dynamics get worse every year, the automation black-boxes get thicker, the support gets worse the smaller you are. But neither one, as a matter of stated policy, reserves the right to cut off an advertiser because that advertiser might someday compete with a product Meta or Google hasn't built yet. The risk with legacy platforms is that they're expensive and opaque. The risk with OpenAI's ad platform is different in kind: the platform can decide you're not welcome anymore, for reasons that have nothing to do with your ad quality or your spend, and everything to do with what OpenAI ships next.

For most businesses reading this, that risk feels distant. You're not building a competing AI product; you're a DTC brand, a healthcare practice, a B2B software company trying to hit a CAC target. Fair enough — today. But betting a growth channel on a platform that is actively expanding its own product surface, while holding a policy that lets it deprioritize anyone standing in the way, is a bet on OpenAI's ambitions staying narrower than they've already proven to be.

We're not saying don't advertise there. We're saying: know exactly what you're exposed to, and don't build anything you can't afford to lose access to overnight.

What Happens to Your Measurement When the Funnel Collapses

Here's the operational problem underneath the strategic one, and it's the one that will actually cost you money first.

For over a decade, growth marketing has run on a funnel you could instrument: awareness in one channel, consideration in another, conversion tracked back through UTMs, pixels and multi-touch models that, however imperfect, gave you something to optimize against. Search told you intent. Social told you interest. Your CRO stack sat at the bottom, converting whoever survived the trip.

Conversational ads collapse that trip into a single interaction. If someone discovers a product, asks it questions, gets convinced and converts, all inside one ChatGPT conversation, there is no funnel to instrument. There's one event. Your existing attribution stack — the one you've spent real time and budget building — wasn't designed to see inside a conversation happening on someone else's platform, in someone else's interface, reported back to you through whatever data OpenAI decides to expose.

We work with a couple healthcare clients whose entire growth model depends on precise attribution — knowing exactly which touchpoint moved a patient from awareness to booked appointment, because their sales cycle is long and their CAC tolerance is tight. When a new discovery channel emerges that doesn't pass clean referrer data or consistent conversion signals, our first instinct isn't excitement about reach. It's a hard question: can we actually verify what this channel is doing, or are we about to optimize budget toward a number we can't audit?

That's not rhetorical, and it's not one to answer after spend has already shifted. Before ChatGPT ads get real budget from any business we work with, we ask three things. What data does OpenAI actually expose about the conversation, not just the outcome? Can we connect that outcome to our own tracking, independent of OpenAI's own reporting? And if the channel's reported performance and our own numbers disagree, whose number do we trust — and can we prove it?

Most businesses jumping into ChatGPT ads right now can't answer those three questions, because the tooling to answer them barely exists yet. That's fine for a small test budget. It's not fine for a growth channel you're betting a quarter on.

The businesses that get burned here won't be the ones who tested too late. They'll be the ones who scaled spend on a channel before they could independently verify what it was actually delivering — the same mistake performance marketers made with early Facebook attribution, and again with in-platform view-through conversions, and are still cleaning up from both.

How We're Actually Testing This With Clients

We're not telling our clients to ignore ChatGPT ads. We're telling them to treat it the way we treat every new channel: test it like you might be wrong, not like you've already decided you're right.

Concretely, that looks like a few things.

Small, bounded budget first. Not a strategic reallocation — a test with a defined ceiling, run long enough to get a real read, small enough that being wrong doesn't wreck the quarter. We've watched too many businesses treat a new channel's early hype as permission to skip the boring, disciplined part of testing.

Independent measurement before scale. If we can't verify a conversion outside of OpenAI's own reporting, we don't scale budget against it. That's not a special rule for ChatGPT ads — it's the same rule we apply to every platform's self-reported numbers, because every platform grades its own homework.

No dependency by design. This is where our no-long-term-contract model actually matters, not as a sales pitch but as an operating discipline. We don't build a client's growth engine around a single channel that a platform could restrict, deprioritize or price out of reach with one policy update. Diversification isn't a hedge we talk about — it's a structural requirement of how we build media plans, because we've watched what happens to businesses that don't.

Watching the policy, not just the product. Somebody on our team reads OpenAI's advertiser policy updates the same way we read Meta's and Google's, because the policy is where the real strategy shows up, months before it shows up in a product announcement. The rule change blocking competing advertisers happened quietly, days before the flashy Sponsored Agents launch got all the coverage. That's not an accident. That's how platforms move the pieces that matter while attention is on the pieces that don't.

This is the difference between a business that reacts to marketing news and one that operates on it. Anyone can read the headline and decide to "get into ChatGPT ads." Fewer people will read the policy update underneath it and ask what it structurally means for how much control they're handing over, and to whom.

We built ELLERA around the idea that a growth partner should do the actual work — not sit in strategy decks talking about channels, but be in the ad accounts, in the attribution data, running the tests and owning the results. This is exactly the kind of moment that tests whether an agency does that or just talks about it. New channel, real hype, genuine opportunity, real risk buried underneath — the businesses that come out ahead won't be the ones who moved fastest. They'll be the ones who moved carefully, with someone accountable for the downside, not just excited about the upside.

In practice, this means every client testing ChatGPT ads with us gets a plain-language weekly readout: what we spent, what we can verify converted, what we can't verify yet, and what we're doing about the gap. No dashboard buried behind a login only we can access. No monthly call where the story quietly changes shape. If a channel isn't working, we say so in week three, not next quarter — because the whole point of not locking clients into a long-term contract is that we have to keep earning the relationship with results they can actually see, not results we tell them to trust.

Where This Leaves You

Test ChatGPT ads. The early advertiser data is real, the format is genuinely new, and businesses that learn a channel early usually get cheaper attention than the ones who wait for it to mature and get expensive. We'd tell any client to run a bounded test right now.

But separate two questions that are getting collapsed into one everywhere this week: "Is this a channel worth testing?" and "Is this a channel worth depending on?" The answer to the first is yes. The answer to the second is: not yet, and maybe not for a while, because the company running it has told you plainly that it will use its position to protect its own competitive interests when the two conflict.

So here's the actual question worth asking about your own growth engine, not just OpenAI's: how many of your channels could restrict your access tomorrow, by policy, for reasons that have nothing to do with your ad quality or your spend — and would you even notice how exposed you were until it happened?

Most growth teams can't answer that. That's the real story this week, not the new ad format.