ChatGPT Ads are live. The auction is still nearly empty.

What are ChatGPT Ads, and how do they look?

ChatGPT Ads is OpenAI's advertising platform. Below a ChatGPT answer, a paid card shows up with a title, a short description, an image and a link, the chat_card. It looks much like a sponsored result in Google, except it sits inside the conversation instead of above a list of links.

The ad is always labelled as sponsored and kept visually separate from the answer. It never shapes the answer itself. OpenAI keeps the reply and the advertising strictly apart, so the model still answers with whatever helps the reader most. You buy the space under the answer; the answer itself stays out of your reach. That separation is what the whole channel rests on, because the moment ChatGPT started answering based on who paid, people would stop trusting it. OpenAI holds three rules here: ads are clearly labelled, they stay separate from the answer, and people control how their data feeds advertising.

A cheap auction, until everyone else notices

Start with the numbers. ChatGPT has over a billion users a month, 900 million a week, and handles around 2.8 billion queries a day. Among AI assistants it holds close to 42% of the market. This is not a side channel. It is a second Google, twenty years younger.

The platform ran in the US first. Ads went live there in February 2026, the self-serve Ads Manager in May. Then came Europe. On 18 August 2026 OpenAI announced the expansion to 31 countries, and on 31 August it opened the self-serve Ads Manager to advertisers across those markets. Within a few months of running, ChatGPT Ads passed a billion-dollar annual revenue run rate. This is not an experiment. It is a new media market, and it just opened up.

When the platform launched, a thousand impressions cost 60 dollars and you had to spend at least 200,000. By May OpenAI dropped the minimum and the price fell to around 25 dollars. Few advertisers, plenty of open space, so the price drops. We watched the same thing with Google Ads and Facebook. Whoever showed up early bought clicks for pennies. Two years later they paid ten times as much for the same click, once the auction filled up. We are at that first point now, on a channel that already reaches a billion people.

And it is not only startups moving in. Among the first advertisers in ChatGPT are Newegg, Best Buy, Lowe's and VistaPrint. Newegg put it plainly: they want to be part of the discussion the moment someone is working out what to buy. That is exactly the point.

A billion dollars in six months. And still behind plan.

In early September 2026 OpenAI announced the ads business had reached a one billion dollar annualised run rate, counting only real advertiser spend and excluding promotional credits. Impressive until you remember OpenAI had projected two and a half billion for 2026. So the platform is growing fast and missing its own forecast at the same time. eMarketer puts the market at roughly 5.4 billion by 2030.

That gap is good news for you. If the auction were filling on plan, prices would already be climbing. Instead there is room, because advertisers are holding back over proof rather than performance. Rajeev Nair of Lifesight told AdExchanger that conversion tracking and audience options are still rudimentary, and until that improves, budgets stay in the test column instead of the core allocation.

Put plainly: the auction is cheap because the big spenders are waiting on reporting. You do not have to wait, provided you build the measurement yourself.

There is no rate card. Here are other people's invoices.

OpenAI has published no benchmarks, and the interface offers neither auction insights nor impression share. All that exists are the numbers advertisers have chosen to share. Search Engine Journal collected them, and the spread is worth studying.

A range from four to twenty-three dollars per click is not a measurement error. It is evidence that the auction is still shallow and your price depends on who happens to be bidding in your topic and your market.

One number circulates as a benchmark and is not one. OpenAI suggests starting with a maximum bid of three to five dollars per click. That is a default for new accounts. The platform average may sit elsewhere. Anyone who confuses the two is budgeting against somebody else's starting line.

How does targeting work in ChatGPT?

Forget keywords. ChatGPT does not match ads to what someone types into a search box, but to what they are talking about. It targets by context: the topic of the conversation, past chats, and which ads the person already engaged with.

Say someone asks ChatGPT which CRM to pick for a growing sales team. Below the answer, a card from a CRM vendor appears. This is not a person who typed a keyword into Google. This is a person mid-decision, asking for advice. That is a different kind of attention than classic search advertising, and it is why it is worth so much. OpenAI makes this its main argument for the whole platform: people in ChatGPT do not just look up information, they compare options and decide. The ad puts you in the running at that exact moment.

Since August 2026 you can also target defined audiences. The core stays the same: you buy the live context of the conversation.

Who sees your ad, and who doesn't?

Most guides leave this out, and it changes everything. Only users on the free and Go tiers see ads. That is roughly 95% of everyone in ChatGPT. Anyone paying for Plus, Pro, Team, Business or Enterprise sees no ads at all.

The implication depends on who you sell to. For a broad consumer audience, this is excellent, you reach the mass for little money. For a narrow group of senior decision-makers who almost all pay for Plus or Pro, ads will miss some of them entirely. Nobody who just wants to sell you the channel will tell you this. We say it up front, because a campaign your audience never sees is money down the drain.

How do you pay, and how do you measure it?

There are three buying models. CPM pays per thousand impressions, good for building awareness. CPC pays per click. oCPC is the conversion model, pushing delivery toward clicks more likely to lead to action while still billing per valid click. oCPC is a standard Google and Meta have had for years. ChatGPT only added it in mid-2026, which tells you how young the platform still is.

Measurement rests on two things: the OpenAI Pixel and the Conversions API. Both track what a person did after clicking the ad, a purchase, a lead, a sign-up, and the results add up in Ads Manager. In practice that means installing Google Tag Manager, the pixel, and the Conversions API connection.

Measurement on this channel is young and attribution is imperfect. Install the pixel wrong and at the end of the month you see costs. This is where it breaks, and where most people who run it themselves over a weekend lose money.

Month one costs you three times more. That is normal.

This is the most important part of the article and the part nobody wants to hear.

Intelegencia documented six months of ChatGPT Ads for a US financial advisory platform that only earns when an advisor accepts a lead. Month one, cost per lead ran three times higher than paid search. By month four it dropped below it. At steady state it settled 35% cheaper, an 80% reduction over the six months. Click-through held around one percent throughout and the on-site conversion rate barely moved.

More useful than the numbers is one line from the case: at fourteen or thirty days the numbers said stop. The ninety-day review was set before a dollar was spent.

What they changed along the way: targeting rebuilt from keyword lists into conversation-led briefs, CRM approval data wired into campaign reporting, optimisation moved from form fills to approved leads, and context hints rewritten on a weekly cadence.

Which gives you a rule to put in the budget before you put anything in the campaign. Without the money and the patience for ninety days, do not start. A channel you cut off in week three owes you no answer, only an invoice.

The pixel takes an hour. A broken pixel takes a month to notice.

Measurement runs on the OpenAI pixel and the Conversions API, supporting twelve standard events plus one custom event, each with its own data type. This is where the expensive mistakes live, the ones that surface at month end when you have costs and no conversions.

Take that last line seriously. The difference between 57 and twenty is not rounding, it is threefold. The cause could be script blocking or a different click definition, but until you know which, you have nothing to base a budget decision on.

What the platform still cannot do

For balance, here is what you are used to elsewhere and will not find in ChatGPT yet.

None of this argues against using the channel. It argues for entering with your own measurement and a budget that can absorb six months of not knowing how you rank against anyone else.

You can see the competition, just not from inside the platform

When the platform gives you no comparison, you go around it. Marketing Miner has tracked ChatGPT ad placements since September 2026: what share of answers carry an ad at all, which advertisers appear, with which headlines and destination URLs, which specific prompts trigger an ad, and at what average position.

None of that exists in the Ads Manager, and it is the single most valuable thing you can know about a young channel. The share of answers carrying an ad tells you how fast the auction is filling. While it stays low you have a head start. Once it climbs, the cheap period is over and you are in a normal market.

When to wait on ChatGPT Ads

We are not here to sell you the channel at any cost. A few situations where you should wait or spend the money elsewhere:

• You are in a regulated field. OpenAI does not yet run ads in finance, healthcare, dating or politics. If you sit here, the door is shut.
• Your audience pays for ChatGPT. Ads reach only free and Go users. You will not reach a narrow premium B2B audience in full.
• You have no way to measure. Without a properly installed pixel and conversions, you spend blind. On a new channel, that is the most expensive mistake you can make.

We'll launch and run it for you

Launching ChatGPT Ads is not about pressing one button. It is about setting up measurement before you spend a single cent, then running the campaign toward results. OpenAI opens access two ways: self-serve through Ads Manager, or through agency and technology partners. We are the second route, except we also run the campaigns for you. The steps:

• First we tell you, for free, whether ChatGPT is your channel. If it is not, we save you the budget and the time.
• We set up the technical measurement. Google Tag Manager, OpenAI Pixel, Conversions API. So it counts from the first click.
• We launch the campaigns right away, while the auction is cheap and half empty.
• We optimise for conversions. Every week we look at what pays and move the budget there.

One more thing. Ads put you under ChatGPT's answers, for a fee. We can also get you into the answers themselves, for free. That is called GEO, and we wrote a separate guide on it. The ideal is to be in both places at once: in the answer and under it.

FAQ

Sources: OpenAI, Advertise in ChatGPT (ads.openai.com) · OpenAI, ChatGPT Ads expands across Europe (August 2026) · MediaPost and Digiday (CPC launch, self-serve Ads Manager, billion-dollar run rate) · industry coverage of ChatGPT Ads 2026 (oCPC, Pixel, Conversions API).

Want to be among the first brands inside ChatGPT while it is still cheap? Get in touch. The first assessment of whether it is your channel is free.