What the platform charges for
OpenAI's current billing guide describes Views bought on CPM, Clicks bought on CPC, and eligible conversion campaigns with click or impression billing. Conversion optimization describes the delivery goal; it does not promise a fixed acquisition price. Confirm the choices shown in the actual account before committing a test.
Card-billed self-service accounts accrue spend after delivery and charge at an assigned payment threshold or month end. A threshold is a charge trigger, not a campaign budget. Two equal card charges can relate to separate accrued spending; review the corresponding invoices rather than infer duplication from the amounts alone.
Sources: OpenAI: Billing and payment · OpenAI: Campaign objectives and minimum budgets
Minimum spend is an access floor
At this review date, OpenAI's campaign guide lists minimum daily budgets of EUR 15 and USD 25, among currency-specific amounts. The applicable minimum follows the ad account's billing currency. Recheck it during setup because platform terms change.
Meeting that minimum says little about whether a campaign can answer your business question. A long buying cycle or infrequent qualified lead may require more observation than a small budget allows. Decide what evidence would make the test informative before treating the minimum as a recommended pilot size.
Understand the spending envelope
A daily budget is an average across the applicable week. OpenAI's daily-budget guidance allows up to twice the selected amount in a day and seven times across the week for an unchanged budget. Changes introduce proration rules. A campaign-total budget controls the overall campaign allowance.
For illustration, a EUR 20 unchanged daily budget has a EUR 40 maximum daily amount and EUR 140 weekly allowance under those rules. This is arithmetic about spending controls, not a forecast of delivery, an approved budget or an estimate of results. Confirm dates and account timezone as part of the plan.
Sources: OpenAI: Daily budgets
Budget, bid and outcome cost answer different questions
A budget answers how much you allow the test to spend. A bid controls participation in the relevant auction. Cost per accepted lead is an observed result: divide the spend assigned to the test by the number of accepted leads. When that count is zero, the cost per lead is undefined rather than zero.
Write down what accepted means. A successfully submitted form can still be unsuitable for the business. Keep that event count separate from qualified enquiries and booked meetings. For a service business, downstream suitability can matter more than obtaining the lowest-priced contact form.
Plan the whole experiment
Our recommended planning worksheet separates media allowance, setup work, creative and destination work, and measurement maintenance. Those categories help you understand the commitment. They are not Crispy Scale prices or a claim about what every agency includes.
Set a primary outcome, target market, observation period and decision rule. Decide who reviews lead quality and what happens if tracking fails. A spend allowance without those responsibilities can create activity without an interpretable result.
OpenAI says the beta does not yet have universal performance benchmarks. Avoid building a forecast from someone else's unsupported CPC or CPA claim. Use verified prior business data where appropriate, label assumptions and revise the plan when actual evidence arrives.
Read reports before making a spending decision
Spend reporting can lag delivery. OpenAI's quickstart warns that a displayed zero does not prove nothing has accrued. Keep invoices, settled reporting and the test's approved allowance aligned when reviewing the experiment.
For a discussion about a specific offer and acquisition goal, see ChatGPT Ads management. The next useful input is a bounded business question and real measurement, rather than an expectation that a new channel must beat an established one.
Sources: OpenAI: First campaign quickstart · OpenAI: Advertiser FAQ and benchmark limitations
How this guide was prepared
Platform mechanics are based on the official sources below. Test-design recommendations are Crispy Scale’s evaluation method. Illustrations are examples; they are not client campaign results.
Sources checked
Official documentation checked . Platform details can change.