ChatGPT ads: what they cost, who sees them, and how to run one

Most of what's written about ChatGPT ads was written by people who have never opened the Ads Manager. You can tell, because it's all timeline and no fields — nobody mentions that the budget box defaults to a hundred dollars a day, or that an ad won't publish without an image and the interface won't tell you why.

I run two live campaigns in there. This page is what I know from the inside: what it costs, who it actually reaches, how the targeting works, and whether it's worth your money right now. Where I have real numbers from my own account, I've used those instead of published averages.

What a ChatGPT ad actually is

It's a card that appears underneath a ChatGPT answer. Title, one line of description, a square image, a link, and the word Sponsored. That's the entire format — there's one ad type, called a chat card, and no video, no carousel, no expandable anything.

It does not appear inside the answer. OpenAI has been firm on this point and it's the design decision the whole product rests on: the model's response is generated without reference to who's paying, and the ad sits below it in a visually separated block. If that stopped being true the product would collapse, so expect it to keep being true.

Which means the mental model to carry in is not "sponsored search result." It's closer to a recommendation card at the end of a conversation the person was already having — they weren't searching for you, they were working through a problem, and you appeared next to the answer.

How this opened up, with dates

The speed here is the whole story, and it's worth having the dates straight because most write-ups are working from a stale snapshot.

• February 9, 2026 — ads launch in the US for Free and Go tier users. Access limited to large brands and holding companies. $200,000 minimum spend. CPM-only bidding at a $60 default.
• Late March — pilot extends past its planned end date and expands to Canada, Australia and New Zealand.
• April 10 — a beta self-serve Ads Manager appears for pilot advertisers. Minimum drops to $50,000. CPC bidding enters limited testing at $3–$5 recommended bids.
• May 5 — the floor is removed entirely. Self-serve opens to every US business. CPC bidding goes fully live alongside CPM, and a measurement pixel plus Conversions API ship the same day.
• Since — product feed campaigns for retailers, a Maximize Results automated bid strategy, platform targeting, one-day view-through conversions, and expansion into Japan, Korea, Brazil and Mexico. Europe is rolling out now.

Eighty-six days from a $200,000 door to no door at all. For comparison, Google Ads took years to reach the same posture. Reported figures put OpenAI past $100 million in ad revenue within six weeks of launch, against a $2.5 billion target for the year.

Who actually sees ChatGPT ads — read this before anything else

This is the single most important qualifying fact and it's usually buried in paragraph nine somewhere.

Ads only serve to Free and Go tier users. Anyone paying for Plus, Pro, Business, Enterprise or Education never sees an ad, at all, ever. Users predicted to be under 18 don't see them either, and ads are kept away from health, mental health and political contexts.

Roughly 95% of ChatGPT's users are on the free tier, and ChatGPT reaches somewhere near 920 million people a week — so the reachable pool is enormous. But the exclusion is not random. The people paying twenty or two hundred dollars a month for ChatGPT are disproportionately the professionals, executives and heavy users. If your buyer is a VP at a large company, a developer, or anyone who lives in this tool all day, you cannot reach them here and no amount of budget changes that.

If your buyer is a normal consumer, a small business owner, a homeowner, a student, someone comparing services — that's the free tier, and it's most of the internet.

What ChatGPT ads cost

There is no minimum spend. That's the headline. You can open an account and run twenty dollars a day.

Published benchmarks put CPC at $3–$5 as a recommended starting bid, with CPM having fallen from a $60 launch default to roughly $25 observed. Early platform click-through rate has been reported around 0.91%, against a Google Search benchmark near 6.4% — a gap that looks structural rather than creative, because a conversational interface simply engages people differently from a results page.

Here's what my own account actually shows, which is the part I can vouch for:

• Cost per click: $2.31–$2.69 against a $3.50 bid cap. Well under the published range, because my category has almost no competition in it yet.
• Click-through rate: 1.37% on my best card, across 1,315 impressions. That's roughly 50% above the reported platform average.
• Daily budget: $25. The field defaults to $100 — change it before you click through, or you've committed to $3,000 a month by accident.

Two honest caveats. My numbers come from an uncrowded category; a competitive vertical will cost more. And low CPC on a new platform partly reflects thin demand, which is exactly the thing that goes away as more advertisers arrive.

How targeting works — and why Google habits will hurt you

This is where most people get it wrong, and it's the difference between a campaign that works and one that quietly burns money.

There are no keywords. You do not bid on phrases. What you supply instead are context hints: short descriptions of the conversations you want your ad to appear beside. OpenAI describes them explicitly as broad thematic guidance, not exact-match targeting, and recommends writing them as the questions, needs and situations people actually bring to ChatGPT.

In practice the difference looks like this:

• Wrong: roofing contractor marketing, home services advertising, contractor growth — a keyword list wearing a disguise.
• Right: A roofing contractor is asking how to get more customers without buying shared contacts from Angi or Thumbtack.

The second one has a person in it, in a moment, with a problem. That's what the matching system is built to read. Write five or six of them as full sentences, and read each one out loud — if it doesn't sound like something a human would type into ChatGPT at eleven at night, rewrite it.

You also control geography and platform. What you don't get is demographic targeting, interest categories, or anything resembling Meta's audience builder. Matching happens inside OpenAI's infrastructure and no personal user data reaches the advertiser.

Where ChatGPT ads are available

As of now, Ads Manager is live in nine markets: the United States, United Kingdom, Canada, Australia, New Zealand, Japan, Korea, Brazil and Mexico. Thirty-one European countries are rolling out.

If you're outside those, you can't buy yet. The useful thing to do in the meantime is decide whether your buyer sits on the free tier, and write your context hints, so you're not starting from zero on the day it opens in your market.

What's still missing

An honest account of a young platform has to include what it can't do yet:

• No retargeting. You cannot build an audience of people who visited your site and chase them here.
• No CPA bidding. Announced, no date. Today it's CPC or CPM, with an automated Maximize Results option.
• No third-party measurement. OpenAI has said partners are coming without naming any, which means for now the platform reports on its own performance.
• Limited categories. Eligibility currently centres on lower-risk verticals — household goods, local services, travel, entertainment, digital products, education.
• Attribution is thin. A pixel and a Conversions API exist; the full picture doesn't yet.

That list is the reason to test with small money rather than move a budget. It is not a reason to wait, because the pricing advantage disappears at roughly the same speed the tooling improves.

How it compares to Google and Meta

Google Search catches people who have already put a query into words. That's the highest-intent traffic on the internet and it's priced accordingly — and its click-through benchmark, near 6.4%, is seven times what ChatGPT is currently doing.

Meta buys cheap reach against a scrolling thumb, with the best demographic targeting in the business.

ChatGPT is a third thing. The person isn't scrolling and isn't searching — they're mid-problem, working something out, often several exchanges deep into describing their situation. The ad arrives at the end of an answer, which is a strange and interesting place to be. Lower click-through, richer context, and for now, less competition.

Nobody sensible is moving their budget here. The play is a small parallel test while it's cheap, on the theory that being early to a channel is worth more than being efficient in a crowded one. That's a judgment call, not a fact, and you should make it with money you can afford to learn with.

Setting up your first campaign

The structure is three levels: campaign holds the budget and geography, ad group holds the bid and the context hints, ads are the cards themselves.

The short version, in order: register at ads.openai.com, install the measurement pixel on your site before you spend anything, create the campaign on a Clicks objective, attach a conversion event, set the budget down from its $100 default, choose Manual Max CPC rather than the automated option for your first run, write your context hints as sentences, and build four cards rather than one.

Four traps that cost me money or time before I caught them:

• The budget defaults to $100/day and the bid to $4.00. Both need changing.
• Titles truncate above about 24 characters even though the field allows 50. It warns you, quietly, after the fact.
• `oppref` and `olref` are reserved. Put either in your tracking parameters and the form rejects the whole field.
• Every ad requires a square image. Without one the review button simply stays greyed out and never explains itself.

And the one that isn't a platform quirk: do not send the click to your homepage. Build one page, one subject, one action. Somebody clicked a card about a specific thing in the middle of thinking about that thing; a homepage with eight menu items loses them, and you paid two or three dollars for the visit.

Is it worth it for your business?

Run the three questions in order and be honest on each one.

Is your buyer on the free tier? If they're a senior professional at a large company, probably not, and nothing else on this page matters. Stop here.

Can you afford to learn rather than earn? Twenty-five dollars a day for sixty days is fifteen hundred dollars to find out whether a brand-new channel works for your category. If that money needs to come back this quarter, spend it somewhere proven instead.

Do you have one page worth sending traffic to? If the honest answer is "my homepage," fix that first. It's free and it's the single biggest determinant of whether any of this pays.

Three yeses and this is one of the more interesting small bets available in advertising right now. One no and it isn't — and I'd rather tell you that than sell you something.

Questions people ask

If you'd rather watch someone build one than read about it, that's exactly what the 30-minute session is. And if video advertising is what you're actually after, here's what my agency does on YouTube for local and regional service businesses.