ChatGPT ads run at a $60 CPM, or a recommended starting bid of $3 to $5 per click. There’s no minimum spend.
That last part is new. In January the entry price was $200,000, and the number that matters most now isn’t the rate card. It’s that a relevance-weighted auction decides what you end up paying.
Here’s the pricing, the history that explains the confusing figures elsewhere, and what to budget for a first test.
What It Used to Cost
The entry barrier collapsed in three months. Knowing the shape of that fall explains most of the contradictory numbers you’ll find in older articles.
- January: a $200,000 minimum commitment, confirmed by OpenAI to Adweek.
- April: cut to $50,000, still gating a managed pilot.
- May: removed when self-serve opened to US businesses.
- Today: no account minimum, though a campaign still needs a $25 daily budget.
Some agencies were quoted $250,000 rather than $200,000 in the earliest phase, which is why both figures circulate. Both were real. Neither applies now. Our dated record of every change sets the full sequence out.
So if an article tells you ChatGPT ads need six figures, check its date. That was true in winter and it stopped being true on 5 May 2026.
Why the collapse matters more than the rate
It matters because a $200,000 floor was a filter. It meant the only advertisers in the auction were large ones with agency support, and it kept competition thin for anybody who could clear it.
Removing it changes who you’re bidding against. The rate card stayed put while the number of bidders went up, which is the usual way an auction gets more expensive without any published price changing.
The Three Ways You Can Pay
Campaigns run on one of three objectives, and all of them settle through the same auction.
Cost per thousand impressions
The published CPM is $60, and it’s held there since the February pilot. OpenAI’s stated justification is the intent quality of the audience, on the argument that somebody mid-conversation about a problem is worth more than somebody scrolling past.
Cost per click
Click bidding arrived with the self-serve launch. OpenAI’s recommended starting bid is $3 to $5, which tells you where to begin rather than what you’re promised.
Cost per conversion
Conversion-optimised campaigns are live. You set a bid cap for a conversion and the system predicts click-by-click how likely one is, then bids accordingly.
There’s a prerequisite: conversions have to be flowing through the pixel or the conversions API before you can use it. OpenAI publishes no starting bid here, so you’re working from your own cost per acquisition rather than from guidance.
Which one to choose first
Impression bidding buys you presence and tells you whether your hints deliver at all. Click bidding buys you traffic and tells you whether the card earns attention.
Most first tests learn more from clicks, because delivery without engagement is the failure mode you want to catch early. Presence you can’t measure is the expensive kind.
Conversion bidding isn’t a starting option, since it needs conversion data already flowing before it has anything to optimise toward. It’s where a working campaign goes next, not where a new one begins.
What Sets Your Price
The auction is relevance-weighted and second-price. Those are two separate mechanics, and both reward a well-built account.
Relevance-weighted means the best-matched ad wins rather than the highest bid. Second-price means the winner pays based on the next bid down, not on its own.
Together they punish the reflex that works elsewhere. Raising your bid to outrun a better-matched competitor is expensive and unreliable, because you’re bidding against their relevance rather than their budget.
The cheapest lever is your landing page
Matching reads your landing page alongside your context hints, your ad title and your ad copy. A page that’s vague about what you sell weakens delivery before any bid gets considered.
Tightening that page costs nothing per click and improves the input the auction weighs most. It’s the first thing to fix and the last thing most advertisers look at.
Your category sets the floor
You’re competing with whoever else wants the same conversations. A crowded category with high customer value bids the clearing price up, and a quiet one leaves it near the starting guidance.
That’s why a single published rate can’t tell you what you’ll pay. The rate is the entry point to an auction whose price is set by your competition.
What Other Advertisers Report Paying
Here the ground gets softer, and it’s worth being clear about why.
The figures below circulate widely in marketing coverage. None of them comes from OpenAI, and we can’t verify them against an account we control, so treat them as market chatter rather than as rates.
Clearing rates are reported to have softened below the published CPM as inventory grew. Click costs in software and finance are reported well above the starting guidance, while retail and ecommerce are said to sit close to it.
The pattern is the ordinary one. Categories with high customer value bid more, and categories with thin margins bid less. If your Google Ads costs sit above average for your sector, expect the same here.
We’d rather give you the published numbers and name the rest as unverified than repeat a precise figure we can’t stand behind. A cost article that presents blog estimates as rates is how a budget gets set on fiction.
What a First Test Should Cost
With no minimum, the question changes. It’s no longer what you’re allowed to spend, it’s what buys a readable result.
A test needs enough volume for the auction to find your matches, and enough time for delivery to settle. Too small and you learn nothing except that delivery was thin.
- Budget: size it so a fortnight produces real click volume, not a handful.
- Duration: run long enough that early delivery noise settles.
- Structure: keep hints narrow at first, then widen once something works.
Instrument tracking before the first impression rather than after. Pixel and conversion tracking both exist, and retrofitting attribution to a campaign already running wastes the spend you’ve already made. The full setup walkthrough covers wiring it up.
One wrinkle to know before you read a first invoice. Since late July the daily budget has been a seven-day average rather than a hard cap, so individual days run hot or cold while the week reconciles.
Reporting won’t tell you which questions triggered you
Data comes back aggregated, and individual conversation data isn’t shared with advertisers. You’ll see that a hint delivered and converted. You won’t see the prompts behind it.
Budget for that blind spot. If your usual optimisation loop runs on mining query reports, it has nowhere to run here, and you’ll be iterating on hints instead of on search terms.
Decide what a click is worth before you bid
The starting guidance is a number OpenAI picked, not a number your business can afford. Work back from your own conversion rate and order value, and you’ll know whether $5 a click is cheap or ruinous long before the auction tells you.
Advertisers who skip that step end up reading their cost per click as good or bad against the guidance rather than against their margin.
Why the CPM Looks High Next to Search
It looks steep because the instinctive comparison is display advertising, where a fraction of that rate is normal.
The comparison that fits better is paid search. An impression here happens inside a conversation where somebody has described a problem in their own words, which is closer to a query than to a banner.
Whether that premium is justified is your call, not ours, and it turns entirely on how closely the conversations you match resemble buying intent. A hint that delivers against idle curiosity is expensive at any rate.
What Isn’t Included in the Price
The media cost is the part with a published rate. Three costs sit outside it.
Creative is one, and the format is unforgiving. The fields accept more, but the display truncates, so OpenAI’s guidance lands near 16 characters of headline and 32 of description.
Landing pages are the second, and they’re an input the auction reads, so a thin page costs you delivery as well as conversions.
Measurement is the third, since pixel and conversion tracking need wiring up before launch rather than after it.
If somebody runs the account for you, their fee sits outside the media cost too. Our OpenAI ads agency page sets out what that covers.
The Cost Nobody Quotes You
Every one of these numbers buys a card below the answer. None of them buys a mention inside it.
A brand cited in the response is there because the model drew on a source that mentions it, and that placement doesn’t stop when a budget does. The sponsored result sits below the answer and is a separate purchase entirely.
Worth pricing both before committing to either. The ad has a rate card and an off switch. The citation has neither, which cuts in both directions.
Frequently Asked Questions
How much do ChatGPT ads cost
The published rate is $60 per thousand impressions, and OpenAI’s recommended starting bid for click campaigns is $3 to $5. Your cost depends on relevance and competition, because the auction is relevance-weighted.
Is there a minimum spend for ChatGPT ads
No. The minimum was $200,000 in January 2026 and $50,000 in April, and it was removed on 5 May 2026 when the self-serve Ads Manager opened to US businesses.
Can I bid on a cost per conversion basis
Yes. Conversion-optimised campaigns let you set a bid cap for a conversion, but they require conversions to be flowing through the pixel or the conversions API first.
Why is my competitor winning with a lower bid
It happens because the auction weighs relevance, so a better-matched ad can beat a higher bid. Improving your hints and landing page usually costs less than raising the bid.
Does a bigger budget get me into the ChatGPT answer
No. Advertising buys a labelled card below the answer, and OpenAI’s stated position is that ads don’t influence the response itself.
Can I see which conversations my ads appeared in
No. Reporting is aggregated, and individual conversation data isn’t shared with advertisers.
Start With the Page, Not the Bid
The barrier to entry is gone, so the only question left is whether the auction thinks you’re a good match. That’s decided by your landing page and your hints, and neither costs anything to improve.
Fix those first. Then find out what a click is worth to you.