ChatGPT Ads campaign setup: the configuration, screen by screen

What a campaign actually holds

In Ads Manager a campaign carries four things and only four: an objective, a budget, dates and targeting. Everything else lives one level down.

The documentation adds a useful organising rule: a good campaign groups together the ad groups and ads that support a shared business goal. In other words the campaign is the budget and strategy unit, not the thematic unit. If two offers share an objective and a budget envelope, they can live in one campaign; if they target different conversations, they split at the ad group level.

Structure reminder: campaign, then ad group, then ad. The full map of which setting sits at which level is on our ChatGPT Ads Manager page.

Step 1: the objective, which locks your billing

The campaign objective determines both how your ads are priced and how delivery is optimized. Three objectives are available:

Two warnings before you click:

• oCPC does not bill per conversion. It optimizes toward a conversion event, but payment stays per click. It also requires working conversion tracking, with exactly one active standard conversion event serving as the optimization goal.
• An existing campaign cannot be converted to oCPC, and neither the objective nor the event can be changed after creation. You have to create a new campaign.

The economics of each model are on our ChatGPT Ads pricing page.

Step 2: the budget and the flight dates

The campaign budget controls how much you are willing to spend across all the ads in the campaign. Two modes:

• The campaign-total budget: the amount to spend over the full duration. The documentation presents it as the stricter control for total spend.
• The daily budget: an amount per day, presented as a delivery target. Actual spend may fluctuate above or below the selected amount on an individual day.

That is not a wording nuance. If your constraint is accounting, take the campaign total. If your constraint is steady pacing, take the daily budget and accept that the day's figure is not a hard ceiling.

OpenAI documents a minimum daily spend of $25 per campaign, which varies by the account's billing currency, with a minimums table published in the Help Center. Since the currency is locked at ad account creation, so is your floor.

The campaign carries its own start and end dates. Two practical effects, usually discovered too late:

• Ads stop serving after the end date. It is one of the documented causes of a "Not serving" status, and probably the most awkward one to explain in a meeting.
• A campaign only delivers if today falls inside its window. Scheduling a campaign for next quarter and wondering why nothing shows the next morning is a classic.

The time zone that carves up those days is the ad account's, chosen at creation and not editable afterwards. For a European campaign run from an account configured on a US time zone, that offset shows up in every chart.

Step 3: geographic targeting

Targeting is set at the campaign level and controls where ads can deliver. Ads Manager Beta supports:

• country targeting;
• in the United States, targeting by state, designated market area (DMA) and ZIP code;
• since the August 24, 2026 update to the official documentation, finer geographic targeting where available (cities, regions, postal codes outside the United States), without OpenAI publishing the list of markets or the granularity covered; we flag it as confirmed in principle, not detailed in its scope.

For a European advertiser, caution is still warranted until the exact scope is published market by market: an "Île-de-France" or "Bavaria" campaign cannot yet be reliably built from geographic targeting alone. The full breakdown of options by country and our method for compensating for a still-uncertain granularity (context hints, language, landing page) are on our ChatGPT Ads location targeting page.

Context reminder: ChatGPT Ads has been available in 31 European markets since August 24, 2026, that is the 27 EU member states plus Norway, Iceland, Liechtenstein and Switzerland.

Platform targeting: iOS, Android and Web

Still part of the August 24, 2026 update, the campaign creation documentation confirms three-value platform targeting: iOS, Android and Web: you can restrict delivery to one, two, or all three. It is a setting separate from geographic targeting, at the same campaign level.

One thing to know before routing budget by platform: OpenAI itself has stated that reporting stays organized into two categories despite this three-value targeting. Since the exact labels of the two reporting categories are not published in detail as of today, we flag this as unconfirmed rather than guess at a mapping. In practice: you can target finely, but check in Ads Manager how your three targets are actually grouped once in the reports, before drawing conclusions on the compared performance of iOS, Android and Web. Details on our iOS vs Android vs Web targeting page.

Step 4: custom audiences

Custom audiences are built from lists of customers or prospects, using email addresses or phone numbers. Once an audience exists, it is used at campaign level in two ways:

• for inclusion, limiting campaign eligibility to people in the selected audiences;
• for exclusion, preventing campaign delivery to those people.

Exclusion is the more neglected of the two and often the more profitable: excluding active customers from an acquisition campaign stops you paying for clicks from people you already have.

Mind the legal frame: uploading a contact list is processing personal data. In Europe the lawful basis, the notice to individuals and the retention period need to be settled before the upload, not after. ChatGPT Ads also launched in Europe without ad personalization, and we find no public documentation spelling out custom audience behaviour market by market: we flag that as unconfirmed.

Step 5: the ad group, theme and bid

Below the campaign, the ad group is a collection of ads organized around a specific theme or intent area. Two decisive things are entered here.

Context hints. They describe the conversations, topics or terms where your product or service may be relevant. The documentation is clear: these hints guide matching but are not exact-match targeting rules. Think of them as broad thematic signals. Our writing method is on the ChatGPT Ads context hints page.

The maximum bid. It is entered here, not at campaign level: a maximum CPM bid for a Reach campaign, a maximum CPC bid for a Clicks campaign. For CPC, OpenAI recommends starting between $3 and $5 per click, and states there is no reliable, generalized per-industry performance benchmark yet.

Maximize Results, the automatic bidding strategy. Since OpenAI added it, this strategy is selected by default on new eligible ad groups: instead of entering a fixed maximum bid yourself, you let Ads Manager adjust the bid automatically to get as many results as possible within your budget. OpenAI is explicit about its limit: Maximize Results guarantees no target CPA, CPC or ROAS, it prioritizes the volume of results. The setting is changed in the ad group's "Bid strategy" section, where manual bidding remains available. Detailed comparison on our ChatGPT Ads Maximize Results page.

The documented hygiene rule: keep each ad group focused. If your context hints cover meaningfully different products, audiences or use cases, create separate ad groups rather than piling everything into one.

Step 6: the ads, then review

The bottom level is the ad itself: headline, description, image and landing URL. Writing them and their format constraints are covered on our ChatGPT Ads creative page.

One thing to know before you even start writing: the landing page must be valid and reachable, and it must not block OpenAI's user agents, in particular OAI-AdsBot and OAI-SearchBot. That is a common failure cause and invisible from inside Ads Manager, which our ChatGPT Ads landing page checker tests directly.

All ads go through review before they can serve. An ad may be rejected if it is misleading, inconsistent or includes disallowed content. If it is rejected, you can edit and resubmit, or request another review if you believe the rejection was in error.

The product feed campaign case

Retailers get a separate path: campaigns built from a product feed. Instead of writing one ad per SKU, you connect a catalogue, mark which products are eligible for advertising, and the platform builds ads from catalogue attributes.

The campaign logic is unchanged (objective, budget, dates, targeting), but feed quality becomes the dominant variable: titles, descriptions, prices, availability, images and destination URLs have to be accurate and current. The full walkthrough is on our ChatGPT Ads for ecommerce page.

What cannot be changed after creation

The short list of irreversibles, worth rereading before you confirm:

• The campaign objective and, on oCPC, the chosen conversion event. An existing campaign does not switch to oCPC: you create a new one.
• At account level, the country, currency and time zone, which among other things set the applicable daily minimum.

Everything else, budget, dates, targeting, context hints, bids and creative, can be adjusted along the way. But an adjustment is not free: changing bids, budgets or creative substantially resets your learning in practice, because you are no longer comparing the same conditions.

After launch: the first 24 hours

OpenAI advises waiting at least 24 hours after launch before flagging a delivery issue. During that window, work the documented checklist rather than touching settings:

1. Is account verification complete? A banner in Ads Manager flags it if not.
2. Is billing fully set up, with a valid payment method?
3. Are the campaign and ads active?
4. Does today fall inside the campaign window?
5. Have the ads completed review?
6. Has the campaign budget already been reached?

If everything is green and nothing delivers, check the bid-strength guidance in Ads Manager before rewriting your context hints: a bid that is too low simply wins no impressions.

The most common structuring mistakes

What we see over and over on new accounts:

• One campaign for the whole catalogue. The budget becomes unsteerable and the results unreadable.
• A catch-all ad group, with context hints spanning three different businesses. The documentation explicitly asks you to split.
• Looking for the bid at campaign level, not finding it, and concluding the platform is incomplete.
• Treating the daily budget as a hard cap, then being surprised by a single day's spend.
• Launching on oCPC with no working measurement, which the documented prerequisites rule out anyway.
• Killing it after six hours. No conclusion is available at that point.

What is not publicly documented

Points we flag as unconfirmed for lack of an official source:

• The maximum number of campaigns, ad groups or ads per account is not published.
• Review turnaround times carry no public commitment.
• The precise auction mechanism, whether manual or via Maximize Results, is not described in how it works internally.
• The exact scope of fine geographic targeting (which markets, which granularity) is not published market by market; only its existence in principle is, since August 24, 2026.
• The detail of the two-category reporting for three-platform targeting (iOS, Android, Web) is not spelled out.
• The exact behaviour of custom audiences market by market in Europe is not publicly detailed.
• No guarantee of delivery, volume or outcome is offered by OpenAI, and nobody can offer one on its behalf, nor does Maximize Results.