ChatGPT Ads are paid placements inside ChatGPT conversations. OpenAI opened self-serve buying on May 5, 2026, and any verified business in one of seven countries can now run them — with cost-per-click or cost-per-thousand-impression bidding, conversion tracking, and no meaningful minimum spend. Several large categories are still gated.
May 5, 2026
7
CPC, CPM, conversion-optimized
August 8, 2026
Here's the thing about every other page on this topic. They all quote the same number — three to five dollars a click — and they all present it as what ChatGPT ads cost.
It isn't. It's the maximum bid OpenAI recommends you start at, published in its own help documentation. A maximum bid is a ceiling you set. A cost is what you actually pay after the auction runs. On a young platform those two numbers can be a long way apart, and nobody outside the platform can tell you yours.
So we publish ours. Every month, by vertical, from real accounts: what ChatGPT Ads actually cost. That is the only part of this page you can't get anywhere else, and it's the reason the rest of it is worth reading.
What a ChatGPT ad actually looks like
Ads appear below a relevant conversation, not inside the answer. Each unit carries six things: the advertiser name, a favicon, a title, body copy, an image, and a landing page.
The character limits are tighter than anything you'll be used to. A title is between 3 and 50 characters. The body caps at 100. For comparison, a Google responsive search ad gives you up to fifteen 30-character headlines and four 90-character descriptions to play with. Here you get one of each, and they have to carry the whole offer.
Translation: you cannot pad your way to relevance. Write the offer, not the brand.
Where they run, and who sees them
Two separate questions, and they get muddled constantly.
Where you can advertise from. OpenAI's Ads Manager is currently available to advertisers in seven countries: Australia, Canada, Japan, Korea, New Zealand, the United Kingdom and the United States. Outside those, you can register interest and wait.
Who sees the ads. OpenAI does not show ads to users on Plus, Pro, or any Business plan, and does not show them to anyone it knows or predicts to be under eighteen. That leaves the free tiers, which is where the overwhelming majority of ChatGPT's users are.
What they cost
OpenAI recommends a starting maximum bid of three to five US dollars per click, and sets the default maximum for impression buying at sixty dollars per thousand impressions.
The auction is relevance-weighted and second-price. Two consequences follow, and both favour small advertisers.
You pay what it takes to beat the advertiser below you, not your maximum. And relevance is weighted, so a tight, genuinely useful ad on a modest bid can outrank a vague one bidding higher. Budget size matters less here than it does almost anywhere else — for now.
On budgets: the platform will accept a campaign lifetime spend limit of one dollar. There is no floor worth planning around. What there is worth planning around is whether your test can produce a readable result, which is a different question and a better one.
For what a click has actually cost across the accounts we run, by vertical, see what ChatGPT Ads actually cost. We publish the methodology and the sample size with it, because a benchmark without those is a rumour.
What you can actually target
Less than you think, and this is the single biggest adjustment for anyone coming from Google Ads.
At campaign level there is exactly one targeting control: location. Country, region, or designated market area. That's the list.
At ad-group level you get context_hints — a free-text list of topics and phrases describing when your product is useful. You can supply up to two thousand of them. They look like keywords and they are not. OpenAI's own wording:
These hints help guide ad matching, but they are not exact-match keywords and do not guarantee delivery in specific conversations.
There's a matching exclusion_hints field, which is the closest thing to a negative keyword list the platform has. It is a suggestion, not an instruction.
No age targeting. No gender or income. No device targeting. No dayparting beyond a campaign start and end date. No search terms report, so no query-level view of what you paid for.
That last one deserves a moment. On Google, the search terms report is where wasted money gets found. I once opened a business-to-business fintech account that had been running on autopilot — right keywords, wrong match modifiers, and consumer traffic nobody had excluded eating roughly eighty per cent of the budget. The evidence was sitting in the search terms. On ChatGPT Ads there is no equivalent report, so the discipline has to move upstream: watch conversions, not clicks, and don't leave it running unattended.
The full field-by-field comparison is on ChatGPT Ads vs Google Ads.
Who can't advertise yet
During what OpenAI calls the initial test period, allowed categories are deliberately narrow — household and consumer goods, local services, travel and entertainment, digital products and education.
Healthcare and medicine, financial services, and legal services are gated, not banned: approved-advertiser categories reviewed manually, case by case. A handful of things inside them are disallowed outright, including cryptocurrency, credit repair and debt settlement. So are adult content, alcohol and tobacco, gambling, political content and recreational drugs.
If you are gated, submit before you plan. We have one in review as I write this, and I cannot tell you what the wait is, because it went in recently enough that I do not know yet and OpenAI does not publish a target. That uncertainty is the whole argument for starting it early: an approval you filed weeks ago costs you nothing while you work on something else, and an approval you file the week the campaign was meant to launch decides your timeline for you.
If you are in a gated or blocked category, read the full breakdown and what to run instead before you plan a quarter around this. If you have already decided it is worth pursuing, the two routes into a restricted category covers what approval reviews and how to structure an offer that is not restricted in the first place.
How to set one up
The short version: create the account, create a campaign with a budget and a bidding model, create an ad group with your context hints and a maximum bid, upload a creative, and submit. Reviews usually take a few minutes.
Three things reliably catch people out. The bidding model is fixed at creation — pick the wrong one and you rebuild the campaign. An ad only serves when the ad and its ad group and its campaign are all enabled and review has passed, which is where "active but not spending" hides. And the 50-character title limit will break the headline you brought with you.
So do them in the right order. Settle the bidding model before you name the campaign, before you write a hint list, before you size a budget — it is the one decision on this platform you cannot revisit, and I have paid the rebuild for getting it backwards. Everything else on the setup screen can be edited later. That field cannot.
The full walkthrough, field by field, is in how to advertise on ChatGPT.
What you can measure
Reporting covers impressions, clicks, spend, click-through rate, average cost per click, average cost per thousand impressions, and conversions. That is the complete list today.
Conversions run through a measurement pixel or a server-to-server conversions application programming interface — the same shape as Meta's. Static tracking parameters survive the click, so tagged landing page URLs show up in whatever analytics you already use.
If your conversion tracking isn't already trustworthy, fix that first. On a channel with no query-level reporting, outcome measurement is the only measurement you have.
Which is why the account structure matters more here than the reporting tab does. Build it so the structure answers the question the reports can't. One theme per ad group, one landing page per ad group, and never two themes sharing a destination — then the page that converts tells you which theme paid, because nothing else will. It is more setup than one ad group and one page, and it is the difference between reading this channel and guessing at it. The reasoning, and what you give up, is in ChatGPT Ads vs Google Ads.
Should you test it yet?
I get asked which platform is best constantly, and the honest answer is that I don't much care — results are the only indicator that means anything. Research tells you where to start. After launch, the money should move to wherever customers are cheapest, and that is as true for a three-month-old channel as a twenty-five-year-old one.
So the question isn't "is ChatGPT Ads good." It's whether you can afford an honest experiment right now. Four gates, in order:
- Is your existing channel clean? If there's obvious waste in the Google account, that's where the next five hundred dollars earns most. Fix the leak before drilling a new hole.
- Can you measure an outcome? No working conversion tracking, no test. You'd be buying traffic you can't judge.
- Is your category allowed and your country covered? Both are checkable in ten minutes, and both are hard stops.
- Will somebody watch it? Autopilot is how ad budgets die. On a channel whose only negative control is a free-text hint list, that goes double.
Clear all four and it's worth a capped test. Fail any one and it isn't — yet.
Then give it long enough to say something. The one test I can put a timeframe on ran alongside Google for an automotive client, and the numbers needed one to two weeks before they were worth acting on. So budget two weeks of patience, not two days — and decide now what you will do at the end of them, because a test with no decision attached just becomes a line item. That is a single test in a single vertical, which makes it a planning figure, not a benchmark. The real ones, by vertical and with the sample sizes attached, are in what ChatGPT Ads actually cost.
Where this channel is genuinely interesting is the shape of the demand. People bring ChatGPT the messy version of a buying question, with far more context than they'd ever type into a search box. That's a real intent signal, and right now there are fewer advertisers competing for it. New channel, old rules: a clear offer, tracked conversions, and a budget sized to learn rather than to look serious.
How Bytown runs ChatGPT Ads for you
Bytown connects to OpenAI Ads with an application programming interface key — one you generate in the Settings tab of your own Ads Manager, scoped to your ad account. No agency owns the account. You do.
From there Bytown builds the campaign, the ad groups and the ads, sets budgets and bids, and pauses, resumes or archives anything that needs it. It reads performance back in alongside your Google, Meta and LinkedIn numbers, so ChatGPT sits in the same view as everything else rather than in its own tab.
The part that matters most: agents never decide alone on anything that touches money. Every spend-affecting change is staged and surfaced for a human to approve before it executes. You approve the campaign before a dollar moves.
That is also why testing this channel is cheap now in a way it wasn't two years ago. Setting up accounts and rebuilding campaigns per platform used to be the expensive part of trying anything new. Bytown creates the campaign once and pushes platform-specific versions out, which turns "should we test ChatGPT Ads" from a project into a decision.
Self-serve plans and the managed option are on the pricing page. One flat fee either way — no percentage of ad spend, because a fee that grows when your budget grows is a fee that quietly wants your budget to grow.
What's changed recently
Early platforms move under you. Dates below are from OpenAI's own developer changelog.
- June 3, 2026 — Location targeting shipped, with country, region and DMA support. Before this, campaigns ran everywhere available.
- June 11, 2026 — Segmented reporting arrived: breakdowns by product, country and device.
- June 16, 2026 — Conversion-optimized bidding, so you can optimize toward a conversion event while still paying per click.
- July 16, 2026 — The Conversions API gained browser-reference matching, tightening attribution.
Three material capabilities inside a fortnight in June. Plan for the rules to keep moving, and re-read the policy rather than a summary of it.
Common questions
Does ChatGPT have ads? Yes. They have been generally available to self-serve advertisers since May 5, 2026, and appear below relevant conversations for users on free tiers.
How much do ChatGPT ads cost? OpenAI recommends starting at a three-to-five-dollar maximum bid per click. What you actually pay is decided by a relevance-weighted second-price auction, so it is usually less, and it varies by category. We publish our real numbers by vertical.
Is there a minimum spend? No meaningful one. The platform accepts a one-dollar campaign lifetime limit. Size the test to produce a readable result instead.
Can I target keywords like on Google? No. Location is the only campaign targeting. Ad-group context hints guide matching but are explicitly not exact-match and do not guarantee delivery.
Will it work for a local business? It can, with a caveat. Location targeting goes down to designated market area, so geography is workable. But somebody who needs a plumber today is not workshopping it with an assistant — urgent local demand still belongs on search.
Should I move budget off Google to test this? No. Run it alongside, funded separately. Budget is a variable, not a number: start small, move quickly, and scale only what the evidence supports.
Run ChatGPT Ads without building them twice
Describe your business once. Bytown drafts the campaign for ChatGPT, Google, Meta and LinkedIn, and you approve every one before a dollar moves.
Generate a free campaign Want it run for you instead? Talk to a marketer.