What Are ChatGPT Ads?

Last verified 8 August 2026. Version 2.0. Next scheduled review 8 September 2026.

---

Search for what it costs to advertise on ChatGPT and you will find the same sentence nearly everywhere: there is no minimum spend.

OpenAI's help centre says the current minimum daily spend is 25 USD per campaign. That page was updated three days before this version was published. In the United Kingdom the floor is 15 GBP, in Japan 2,500 JPY, in South Korea 25,000 KRW, in Brazil 40 BRL.

Version 1.0 of this page said the minimum was removed entirely. It was, on 5 May 2026, and the sentence was correct when we wrote it. Then a per-campaign daily floor appeared and we did not catch it until this review. That is the failure mode of this whole category. Almost nothing published about ChatGPT ads is dishonest. It is just old, and the platform has changed underneath it roughly once a month since February.

So this version is built to be checked. Every claim carries a tier and a source, the review date is at the top, and where we got it wrong last time we say so.

What ChatGPT ads are

ChatGPT ads are sponsored placements that appear in a labelled chat_card below a ChatGPT response, matched to the topic of the live conversation rather than to keywords. They launched for United States users on 9 February 2026 and are now bought through a self-serve platform at ads.openai.com.

OpenAI states that ads run on a separate system from the response, that the model generates its answer without advertiser input, and that conversation data is not sold to advertisers. There are no keyword lists, no match types and no negative keywords. The advertiser-side targeting lever is called context hints.

The auction is relevance weighted and second price. Your bid sets a ceiling, and ad quality, landing page fit and how closely your hints match the conversation all move what you actually pay. On this platform creative work is a cost lever, not only a conversion lever.

This is the format that reaches what we call the reasoning window, the interval between asking a question and forming a conclusion. That framing still holds. What has changed since we first wrote it is everything operational underneath.

Who sees them

Ads may appear for users on the Free and Go plans. Plus, Pro, Business, Enterprise and Edu accounts do not have ads, and OpenAI does not show ads to accounts identified as belonging to people under 18, using account-level age information and age prediction.

Two details in the current documentation change the size of that audience and almost nothing published in this category accounts for either.

Logged-out users are served. OpenAI states that ads shown in logged-out experiences are appropriate for all audiences, and that mature-content ads are limited to eligible logged-in users. Guidance that describes the audience as logged-in only, including earlier versions of this page, is out of date.

Free users can switch ads off without paying. There is an Ads-Free option on the Free plan. It removes ads in exchange for lower message limits and reduced access to tools like image generation. Anyone who takes it leaves the addressable audience while remaining a Free user. Nobody has published what share of Free users has taken it, OpenAI included, so the honest position is that the ad-eligible population is smaller than the Free and Go population by an unknown amount.

Ads also do not appear in Temporary Chats, and during this test they do not appear in the ChatGPT Atlas browser at all.

Then the part that damages B2B media plans specifically. If your buyer works somewhere that bought ChatGPT Business or Enterprise, that person is not in the addressable audience while they are at work. Anyone selling ChatGPT ads as a route to enterprise decision makers should be asked directly how they intend to reach people the platform does not serve ads to.

The four gates

Most published guidance treats availability as one question. It is four, and they resolve independently. A campaign clears all four or it serves nothing.

Gate one: where your business is registered

Advertiser eligibility follows the country your business is registered in, not the country your customers are in. OpenAI's Ads Manager Availability page lists nine countries. Eight are marked Available: Australia, Brazil, Canada, Japan, Korea, New Zealand, the United Kingdom and the United States. Mexico is marked Coming Soon.

That page was updated twelve hours before this version was published. Treat every list you find, this one included, as a snapshot with a date on it.

Gate two: where your buyer is

Opening an account and reaching users are separate permissions, and merging them is the single most common error in the published guidance.

Brazil sits on the advertiser list while Brazilian users are not yet served ads. A Brazilian company can therefore build a campaign it cannot run at home, and target the United States instead. A reader in one of the mismatched markets who trusts a merged list gets told something false about their own country.

Gate three: what the location picker will actually accept

This gate is documented nowhere, so we counted it ourselves.

OpenAI publishes its full targeting catalogue as a CSV at developers.openai.com/ads/openai-geotargets.csv. Retrieved 8 August 2026, it contains 42,110 rows. Of those, 41,488 are United States postal codes. It lists 249 countries as selectable targets, including India, the United Arab Emirates, Saudi Arabia, Singapore, Germany and France, none of which serve ads to users.

Below country level the catalogue thins out sharply:

So the catalogue is a roadmap, not an availability list, and it will show a buyer their own excluded country in a dropdown. OpenAI's own guidance concedes this indirectly by telling you to search for a location during campaign setup to confirm whether it is currently supported. If you are planning a media buy from that CSV, you are planning against a file that describes an intention rather than a permission.

The practical read for anyone outside the United States: this platform is a US geo-targeting product with country-level access for six other markets. Sub-national targeting anywhere else does not exist yet.

Gate four: what you sell

Ad policies v1.3, updated 15 July 2026, limits advertising to a set of consumer categories: household and consumer goods, local services, travel and experiences, and digital products and education.

Financial services, healthcare and legal services deserve close reading, because the shorthand circulating about them is wrong in a way that costs money. The policy does say OpenAI may approve financial and health advertisers case by case, with proof of licensure. It then adds a line most summaries drop entirely: ads for financial services outside the United States are generally prohibited, and ads for health services outside the United States are generally prohibited. The carve-out is US only. Legal services are not permitted in any market, although legal education and LSAT preparation are.

Gambling, alcohol, tobacco, dating, recreational drugs, counterfeit goods and political content are disallowed outright.

Underneath all of that sits a placement layer that has nothing to do with what you sell. Ads do not run beside conversations about mental or personal health, emotionally reliant exchanges, or what the policy calls sensitive user journeys. In April 2026, policy v1.1 stopped categorically blocking ads from medical, legal and financial advice contexts, which changed where ads may appear rather than who may buy them. Those two changes are frequently reported as one.

What it costs, and what the floor really is

OpenAI's guidance recommends a starting maximum CPC bid of 3 to 5 USD, with 60 USD as the default maximum CPM bid. Trade press reporting puts observed CPM clearing between 25 and 60 USD, with most landing between 35 and 50. Those ranges disagree with each other and every one of them is reported rather than confirmed.

The minimum is 25 USD per day per campaign in the United States. Two things follow from that number, and neither is published anywhere else.

What the floor actually buys. At the 60 USD default CPM, 25 USD a day is roughly 417 impressions. At a 3 USD CPC it is about 8 clicks a day, and at 5 USD about 5. Collecting 100 clicks at the floor takes somewhere between twelve and twenty days. Anyone telling you to test this channel on the minimum is describing a three week experiment, not a week.

When Reach beats Clicks. Buying on CPM only wins if your click-through rate clears the ratio between the two prices. At a 60 USD CPM against a 3 USD CPC, the crossover is a 2.00% CTR. Against a 5 USD CPC it is 1.20%. If CPMs clear nearer 35 USD, the crossover falls to 1.17% and 0.70%. Below your crossover, CPM costs you more per click than CPC would have. Since the platform publishes no CTR benchmarks at all, you cannot know your crossover before you run, which is the honest argument for starting on Clicks in B2B and moving to Reach only once you have measured your own rate.

The local floors are also not equivalent to each other. At August 2026 rates, 15 GBP and 2,500 JPY and 25,000 KRW all sit meaningfully below 25 USD, and 40 BRL sits far below all of them. Whether that is deliberate pricing for market maturity or an artefact of rounding, OpenAI has not said.

One more thing before you set a budget. OpenAI describes the campaign-total budget as a total spending limit and not a pacing control, and states in its own wording that the budget is not distributed evenly across campaign dates and that spend may accumulate quickly. A total budget can be consumed in a fraction of the window you set it for.

How you buy

Here OpenAI's own documentation contradicts itself, and we would rather show you that than pick a side.

Create Campaigns for ChatGPT, updated three days before publication, lists three objectives: CPM, CPC and oCPC. Ads in ChatGPT: The Basics, updated two days before publication, lists two, describing only the Reach objective on CPM and the Clicks objective on CPC, with no mention of oCPC at all. Both are current primary sources. Neither is a stale cache.

The likeliest reading is that oCPC is rolling out and the introductory page has not caught up, but that is inference and we are labelling it as inference. If you are planning around oCPC, confirm it is present in your own account rather than in anybody's article, this one included.

On what oCPC is: it is conversion-optimised cost per click, and it is not CPA, though a good deal of coverage calls it CPA. OpenAI's documentation is explicit that you pay per valid click and not per conversion. The system optimises delivery toward clicks more likely to convert, and you continue to carry the risk on whether they do.

Conversion measurement runs through a pixel and a Conversions API, both shipped 5 May 2026. Static UTM parameters survive ad clicks into whatever analytics stack you already run.

Nobody has benchmarks, including the people selling benchmarks

This is the section where a page like this normally shows a performance table. There will not be one.

OpenAI's own FAQ, updated eight days before this version was published, states that ChatGPT Ads does not yet have performance benchmarks across advertisers, industries or campaign types, and that results depend on objective, creative, landing page experience, relevance and setup.

Agencies are nevertheless publishing ChatGPT ads benchmarks. Some of that is real data from a single account over a couple of weeks, which is a data point and not a benchmark. Some of it is Google and Meta performance in fresh labelling. The defensible reading is that a platform in open beta, three months into self-serve, whose operator says delivery systems, inventory and ad formats will keep changing, cannot yet produce a stable benchmark for anybody.

When you see a ROAS figure for this channel, ask how many accounts, over how many days, in which vertical, at what spend. The answer is usually one, fifteen, one, and small.

The rule that complicates our own pitch

There is a clause in the ad policies that is inconvenient for the category of agency we belong to, this one included.

Ads must be clearly distinguishable from the ChatGPT product experience. OpenAI reserves the right to remove, or require changes to, ads that imitate the appearance, functionality or voice of ChatGPT in a way that could reasonably make a user believe the ad is part of the product.

A significant amount of what is currently sold as AI-native ad copy is exactly that: writing the ad so it reads like the assistant. We build methodology in this area and this clause draws a real boundary around it. Matching the register of a conversation is permitted. Impersonating the assistant is not, and the line between them is drawn by OpenAI's reviewers rather than by any agency. Anyone promising you conversational ad copy should be able to tell you where they think that line sits and what happens to your creative if they are wrong.

Review is mostly automated, using classifiers and language models with human oversight, escalating on severity, confidence and risk. Restricted categories draw manual review. Ads and landing pages the systems cannot evaluate are not eligible to run at all.

What we cannot tell you

There is no published approval rate or timeline for manual review in finance, health or legal, which is precisely the information a restricted advertiser needs before committing.

Nobody outside OpenAI knows what happens to clearing prices as international advertisers enter the auction.

There is no public evidence on whether context hints behave like a controllable targeting lever over time, or whether the relevance model overrides them at scale.

And the question that matters most across both of our companies has no published answer at all: whether paid placement inside ChatGPT competes with organic citation in the same answer, or complements it. We sell one side of that surface and our parent company sells the other. We do not know, and we will keep saying so until somebody measures it properly.

If you are in India or the UAE

You cannot open an advertiser account. Neither country appears on the advertiser list or the serving list, and eligibility follows your registration rather than your market. The targeting catalogue listing India and the UAE as selectable does not change that, and gate three explains why.

The two workable routes are a registered entity in an eligible market, or operating the account of a client who already has one. Both carry tax and legal consequences that sit outside this page. The route that does not work is waiting for the catalogue to mean what it appears to mean.

Quick answers

Is there a minimum spend for ChatGPT ads? Yes. The current minimum daily spend is 25 USD per campaign in the United States, 15 GBP in the United Kingdom, 2,500 JPY in Japan, 25,000 KRW in South Korea, 25 AUD, 25 CAD and 25 NZD respectively, 40 BRL in Brazil and 150 MXN in Mexico. The 200,000 USD entry commitment was removed on 5 May 2026, which is a different thing and is the source of the widespread claim that no minimum exists.

Which countries can open a ChatGPT advertiser account? Nine are listed: Australia, Brazil, Canada, Japan, Korea, Mexico, New Zealand, the United Kingdom and the United States. Eight are marked Available and Mexico is marked Coming Soon. Eligibility follows the country your business is registered in.

Can I advertise on ChatGPT from India or the UAE? No. Neither appears on the advertiser list or the serving list. A registered entity in an eligible market, or running a client's existing account, are the only routes.

Can I target India with ChatGPT ads? The location catalogue lists India as selectable, and 248 other countries besides. Listing is not serving. Confirm any location in the campaign picker during setup rather than from the CSV.

Can a financial services company advertise on ChatGPT? In the United States, possibly, through case-by-case approval with proof of licensure across a defined list including insurance, mortgages, credit cards and investment services. Outside the United States, ads for financial services are generally prohibited. Credit repair, debt settlement and alternative investments are not allowed anywhere.

What is the difference between CPC and oCPC? Both charge per valid click. oCPC optimises delivery toward clicks more likely to produce your tracked conversion event. Neither charges per conversion, which is why calling oCPC a CPA objective is wrong.

What is a good ROAS on ChatGPT ads? Unknown. OpenAI states the platform does not yet have performance benchmarks across advertisers, industries or campaign types. Any figure you are shown is one account over a short window.

Should I buy on CPM or CPC? At a 60 USD CPM against a 3 USD CPC, CPM only wins above a 2.00% click-through rate. Against a 5 USD CPC the crossover is 1.20%. You cannot know your rate before you run, so start on Clicks and move to Reach once you have measured it.

Who does not see ChatGPT ads? Plus, Pro, Business, Enterprise and Edu accounts, accounts identified as belonging to people under 18, anyone in a Temporary Chat, anyone using the ChatGPT Atlas browser during this test, and any Free user who has switched to the Ads-Free option in exchange for lower message limits. Logged-out users are served, contrary to a lot of published guidance.

Can my ad be written to sound like ChatGPT? Matching the register of a conversation is fine. Imitating the appearance, functionality or voice of ChatGPT so a user might believe the ad is part of the product is grounds for removal or forced modification under the ad policies.

---

Sources

---

Related reading

• ChatGPT Ads Troubleshooting: Every Documented Failure Mode. Nine documented failure modes and what to send support.
• Blocking AI Crawlers Can Get Your ChatGPT Ads Rejected. Why OAI-AdsBot, not GPTBot, is the one that matters.
• ChatGPT Ads vs Google Ads vs Meta Ads. Where this channel sits against the two you already run.
• Why People Respond to Ads Differently Inside a Conversation. The reasoning window, and why creative behaves differently inside it.
• The 10 Best ChatGPT Ads Agencies for B2B Brands. If you would rather not run this yourself.
• ChatGPT Ads and Google AI Mode Ads. The two surfaces we build placements on.
• Our methodology, starting with the prompt opportunity set.
• Glossary. AEO, GEO, chat_card, prompt-native advertising.

InPromptAds buys and manages paid placement inside AI assistants. It is a Newtation company, and Newtation works the organic side of the same surface. We publish what each of them can and cannot do, including when the answer is that we do not know.

---

Changelog. Version 2.0, 8 August 2026. Corrected the claim that no minimum spend exists, which was accurate on 5 May 2026 and is not accurate now. Added the four gates structure, the location catalogue count, the finance and health US-only qualification, oCPC in place of the incorrect CPA description, the interface imitation clause, and breakeven CTR arithmetic. Updated advertiser and serving country lists. Version 1.0, 14 May 2026, first publication.

Field kit

Tools

• Eligibility Checker
• Budget Calculator
• OAI-AdsBot Checker
• Context Hints Generator
• Buy Scorecard

Site

• ChatGPT Ads
• Google AI Mode Ads
• Methodology
• Glossary
• Contact

OpenAI (primary)

• ChatGPT Ads Help Center
• Ads in ChatGPT: The Basics
• Measure Results
• Ad Policies
• Ads Manager
• OAI-AdsBot verification