Twelve answers from four months of running ChatGPT Ads, on our own money and our clients'. Each one cost us something to learn.
1. Your country and your currency are decided once
The country and currency of an advertiser account are fixed when the account is created, together with the legal entity behind it, and neither can be edited afterwards. A wrong choice means opening a new account and verifying the business again. Decide where the business is registered and which currency it will be billed in before you touch the signup form. The full picture of who can advertise and from where sits in its own guide.
2. There are no agency accounts
OpenAI has no manager account and no shared billing for agencies. The advertiser creates the account, verifies the business and adds the card, then invites the agency in as a user. An agency cannot lend its own eligibility to a business registered in a country that is not open yet. The account, the billing and the campaign history stay with the advertiser, so changing agency is a change of access and nothing gets rebuilt. We wrote up how agency access actually works separately.
3. The minimum daily budget follows the country you register in
Every currency has its own minimum daily budget, and the one that applies is the account's, whatever countries the ads are shown in. A US account starts at 25 dollars a day. A UK or euro account starts at the equivalent of roughly 17 to 20 dollars, and an Indian account at under 8. Aiming at a more expensive market from a cheaper account does not raise the floor. The per-currency detail is in what ChatGPT ads really cost.
4. A new account may only be allowed to advertise at home
Some new self-serve accounts can advertise only in their home country at first, and choosing other countries does not override it. The account becomes eligible for other countries after identity verification is approved and after it reaches a required amount of spend at home. That threshold is not published, and creating a campaign or setting a budget does not count towards it. If the plan is cross-border, budget for a home-market campaign first.
5. A daily budget is a weekly average
A daily budget is paced across the week. On a single day the platform can spend up to twice the daily figure, and across the week it will not go past seven times it. The first days of a campaign often run over, because the week still has unspent room in it, and a campaign that uses its weekly allowance early can stop delivering while it still shows as active. Quote a budget as an amount per day with the weekly and monthly totals beside it.
6. Only one of the five rejection reasons is about your words
Ads come back with one of five reason strings, and only the first is mainly about the copy: platform integrity and misrepresentation, landing page mismatch, a crawler challenge, a failed crawl, and a generic policy string. The landing page is the single strongest factor in whether an ad clears, and the theme of an ad group decides more than any one sentence inside it. About one ad in ten is classifier noise with no cause worth chasing.
7. Claims and regulated vocabulary hold ads back
Percentages, multiples of return, guarantees and clinical or regulated vocabulary are the words most likely to keep an ad from clearing review, even when the same claims sit on your own website. Copy that describes what the product does and what changes for the buyer clears far more reliably. Start without claims, then add them back one at a time once the campaign is approved. The format rules are in creative specs that actually serve.
8. A reason attached to an ad in review is not a rejection
An ad sitting in review often carries a reason string, and on its own it means nothing: every ad in review can carry the same string, including ads that go on to be approved. Only the verdict is diagnostic. Time is the other signal: an ad that has carried a reason for 30 minutes without changing is parked, and only a new creative moves it.
9. The card cuts your title at 224 pixels
The API accepts 50 characters of title and 100 of description. The rendered card gives the title one line in a 224 pixel box and the description two lines of the same width, which is roughly 30 and 65 characters depending on the letters you use. In our builds, over a third of the descriptions written to the character limit arrived cut. Measure the copy in the box it renders in, which is what the creative specs guide walks through.
10. The pixel can accept everything and record nothing
The pixel validates every event in the browser before sending it, and on any problem it stops there: the conversion is never transmitted, a diagnostic goes in its place, nothing is thrown and nothing is logged. A site can look fully instrumented while the conversions table reads zero on every row. The event name has to be one of eleven documented names, so anything carried over from another platform is dropped. The only place this shows is the warnings panel on the data source. Setting it up correctly is covered in conversion tracking on ChatGPT ads.
11. Editing an approved ad takes it off the air
Any change to a live ad's creative, including the destination URL and the image, sends it back to review and stops it serving until it clears again. Renaming it does not. A campaign-wide URL change does this to every approved ad at once, so it belongs in a slot where losing delivery for an hour is acceptable.
12. What a budget actually buys
Across the campaigns we manage, the average cost per thousand impressions is about 15 dollars and the average cost per click about 2 dollars. The cost per thousand has moved between about 10 and 18 dollars from one fortnight to the next. At 30 dollars a day that is roughly 2,000 impressions a day. Both averages sit below the ranges OpenAI publishes, and they move with the topic and with who else is bidding, so no one can promise them. What the reporting does and does not show is in measuring ChatGPT ads.
Sources
OpenAI's Help Center and advertising policies, written answers from OpenAI Ads Support to our own questions, and the campaigns we have been running for ourselves and for clients since July.
