What the First Advertising Campaigns on ChatGPT Reveal

On May 5, 2026, OpenAI opened direct ad buying inside ChatGPT. The announcement got lost in all the current noise, but it carries a real consequence: a channel with close to 900 million weekly users, one that for months had been reserved for a closed group of large brands, became available to any company with a modest budget and a willingness to experiment.

The platform shifted from charging per impression to charging per click, removed the spending minimum that kept it an exclusive arena, and added pixel-based measurement and a conversions API whose reliability, for now, is uneven. Targeting does not work by keywords, as in paid search, but by context signals within the conversation. Taken together, OpenAI did not invent a completely new model: it replicated the playbook Google and Meta have spent years perfecting and moved it into a different environment.

That familiarity is deceptive. The fact that the buying mechanics resemble paid search does not mean the channel behaves like paid search. The only way to find out what kind of medium ChatGPT is, is by running it, and the first campaigns we have run at JULIUS are already showing patterns worth examining carefully, especially because they partly contradict what the public conversation is taking for granted.

The first thing that stands out is cost. The click on ChatGPT comes out notably cheaper than the click on paid search for the same campaign, with the same market and objective. Not a little cheaper, but enough for the difference to draw attention and for the immediate reading to be almost automatic: new channel, still-thin competition, an auction that is not yet saturated. The conclusion many are drawing in public is that this is cheap, high-intent inventory, an arbitrage window worth taking advantage of before it closes. Similar to what we lived through fifteen years ago on Facebook, and still, to a decreasing degree, on TikTok.

The click-to-lead conversion reinforces that reading: the cost per lead turns out to be very competitive. And the figure that completes the optimistic narrative is the click. The click-through rate gets close to that of paid search and lands above that of paid social, which invites the thought that the ChatGPT user arrives at a moment of decision, comparing options and ready to act, and not simply scrolling through a feed.

Up to here, everything points to the same comfortable conclusion: a channel cheaper than paid search, with clicks and conversions that hold up in comparison. If one stops at these three numbers, the decision seems obvious. The problem is that stopping here leaves out the question that really matters for generating demand, which is not how much it costs to bring someone in, but who that someone is when they arrive.

The picture changes when you stop looking at how much the lead costs and start looking at which lead arrived. In the first campaigns, quality came in below the average paid search tends to deliver for the same account. This is worth saying with the caution a still-small sample imposes, a handful of conversions and a few weeks of operation, but the direction is clear and worth taking as an early signal, not a verdict.

Part of the difference shows up in the type of contact the user leaves. In B2B settings, where an email on a corporate domain tends to signal a qualified prospect, the proportion of personal emails, free domains like Gmail or Outlook, came in higher than in paid search. That detail, which might seem minor, points to something deeper. With a search engine, the simple act of typing a commercial query already works as a filter: someone looking for a vendor is, to some extent, declaring an intent. The conversation does not filter the same way. Someone asking ChatGPT is exploring, learning, or comparing, and does not necessarily show up with the identity they would buy under.

This does not mean the channel does not work. On the contrary, in our experience it produced leads where paid social, run in parallel, barely generated any. The point is finer: the channel does convert, and at a good cost, but the contact it delivers is softer than paid search. We see it not only in the type of email, but in later behavior: among the prospects who arrive through ChatGPT, the no-show rate to scheduled meetings is notably higher than among those who arrive through paid search. The intent is there, but it is a different intent.

The click exists and the conversion exists. What changes is the intent behind them. Someone arriving from a search engine is usually weighing whom to hire; someone arriving from ChatGPT is often still just understanding their problem. Confusing that conversational intent with purchase intent is the most likely misreading at this stage.

If the contact has a lower profile (on average) and the most common intent is exploratory, the consequence for strategy is direct: this is not as much a bottom-of-funnel channel as paid search. Paid search captures demand that already exists, someone who went out looking for a vendor and is close to deciding. ChatGPT could be stepping in, on average, earlier, when the user is still defining their need, comparing approaches, or forming criteria. It sits closer to discovery and consideration than to the bottom of the funnel. It is possible that in some cases the user contacts a vendor moved more by the impulse of the moment than by a decision already made.

Hence the most expensive mistake is not one of budget, but of framing. Porting the paid search playbook as is, bidding for the last click, sending traffic to a page built for someone who already wants to buy, and measuring the channel with the yardstick of immediate conversion, is asking ChatGPT to do a job that is not its own. Measured that way, the channel will appear to deliver low-quality leads. The problem would not be the channel, but the expectation it is measured against.

The picture is potentially more interesting. ChatGPT could offer an earlier entry into the funnel, in the middle, with an attractive cost per click, measurable conversion, and prospects that over time can be converted at a profitable unit cost and at a relevant scale. Today, for many lead generation campaigns, that is not available through any medium. The question is no longer whether the channel is good or bad, but what its role is within a robust demand generation strategy.

It is worth placing the cost advantage in time, because it is probably the most fleeting thing about the channel. The cheap click is, in large part, a function of the stage: few advertisers, an unsaturated auction, and a platform that is still learning how to charge. It is the same pattern we already lived through with Facebook and, in a shorter version, with TikTok: the window of low prices closed as advertiser demand grew. ChatGPT could follow that path, but it is not the only possible trajectory.

The price of an ad depends on how much advertiser demand there is against how much inventory exists, and that inventory could grow along two paths: a user base that keeps expanding and a greater number of ads per conversation. If supply grows quickly, the price increase could be contained even as more competitors enter. Facebook's history points the other way, because there demand grew faster than supply and the platform learned to charge more per impression, but that is not a law. The most honest way to put it is in terms of probability: the window probably gets more expensive over time, without our being able to anticipate how fast.

There is a second risk, less obvious and more fundamental. Of the two paths to expand inventory, raising the number of ads per conversation is precisely the one that most quickly erodes what makes the channel valuable. And what is valuable is the trust with which the user comes to ChatGPT: they ask without suspicion, explore honestly, and treat the answers as advice and not as a sale. That trust is the asset, and it is in tension with the business model that sustains it. OpenAI has laid out advertising revenue targets that go from 2.5 billion dollars this year to something close to 100 billion by 2030, and an ambition of that scale is hard to sustain with discreet advertising. For now the company says it sees no deterioration in its trust metrics, though independent surveys suggest a majority of users perceive that ads reduce the credibility of AI answers. The lever that would most relieve the pressure on price is, then, the same one that most threatens the asset.

None of this disqualifies the channel, but it does change the nature of the bet. What is being tested today is not a settled position one joins to stay, but a hypothesis still in progress.

If what the channel delivers is a softer prospect, under conditions still being defined, the practical conclusion is that the value is not in being present, but in knowing how to operate what comes through it. Capturing a cheap lead is easy; turning it into business when it arrives less qualified is what separates a campaign that works from one that only pads the report. That requires work the cost per click does not reflect: qualifying before passing the contact to sales, nurturing the one still exploring, and reading the data often enough to adjust while the platform keeps changing week to week.

None of this is new in nature, but it is new in its demands. Paid search forgives a certain operational laziness because it delivers already-mature demand; ChatGPT does not. The work this channel requires is sustained, not occasional, and most teams underestimate how much of it there is. The talent that does it can be structured in different ways: onshore and onsite, contracted through an agency, or integrated as a dedicated nearshore team that works as an extension of the in-house operation rather than as an outside vendor. What matters is not where the talent sits, but whether it can hold the standard week after week. Whoever turns on campaigns without that capability behind them will conclude that the channel does not work, when what is missing is the operation that makes it profitable.

At bottom, this reorders where competitive advantage sits. When access to the channel opens to everyone, as just happened, being on it stops being a differentiator almost immediately. What differentiates is the ability to read it well, adjust it quickly, and sustain the operation that turns an early contact into a customer. For a growing number of teams, a dedicated nearshore model has become a practical way to hold that standard with talent that operates as part of the team, without the cost and friction of building everything onsite. The channel gets commoditized; the ability to operate it does not.

The question many bring to ChatGPT is whether it is worth advertising there. It is the wrong question, or at least an incomplete one. The channel is cheap right now, it converts, and it opens a stage of the funnel other media do not reach well. But that cheap click hides a softer prospect, the price will probably rise as it did on the platforms that came before, and the business model that sustains the experience is in tension with what makes it valuable today.

The early advantage, then, is not in arriving first to the channel, but in reading it honestly: understanding what kind of intent it delivers, measuring it with the right yardstick, and building the operation that turns a conversation into a customer. Because what is being tested in these first campaigns is not a new channel to move to, but a bet on how people will search and decide when part of those decisions starts going through a conversation with a machine.

It is worth being there early, with one condition: doing it while understanding what is being tested, without confusing what the channel promises with what it still has to prove.