A useful campaign structure lets you answer three questions: what are we offering, who might need it, and what decision will the results support? Start there before creating a separate campaign for every product, phrase or headline.
This reference explains the platform hierarchy and a practical way to plan it. For access and the broader launch process, start with getting started with ChatGPT advertising.
The four levels of a ChatGPT Ads account
The platform hierarchy is advertiser account → campaign → ad group → ad. A campaign belongs to an account; ad groups belong to a campaign; ads belong to an ad group. OpenAI’s campaign reference, ad-group reference, ad reference
Think of each level as a different planning question:
| Level | Planning question | What to write down before setup |
|---|---|---|
| Advertiser account | Which business is advertising? | Account owner, billing responsibility and who approves changes |
| Campaign | What are we testing within an agreed scope? | Offer, market, spending boundary and review date |
| Ad group | Which customer situation deserves its own approach? | One coherent need, relevant hints and a reason to separate it |
| Ad | What will the customer see and open? | A specific message, suitable image and matching destination |
OpenAI places campaign timing, geographic targeting and campaign spending limits at campaign level. Context hints and maximum bids are ad-group settings. A standard chat-card ad carries the title, body, image and destination. The references above describe the current fields; check the setup available in your own account before submitting.
Separate a campaign when the operating boundary changes
The following is our planning framework, rather than a platform rule requiring a particular account layout. Separate things when doing so creates useful control or a clearer decision. Keep them together when the difference is merely cosmetic.
| Situation | Sensible starting point | Reason |
|---|---|---|
| Another client owns the offer and spend | Use that client’s authorized advertiser account | Keep ownership and reporting clear |
| Two markets need separate spending decisions | Separate campaigns | You can review each market against its own cap and destination |
| A seasonal promotion has distinct dates and terms | Separate campaign | Its operational boundary differs from the ongoing offer |
| One product serves two distinct customer needs | Separate ad groups within a campaign | The messages and pages may need to answer different questions |
| You want to compare two headlines for the same need | Consider ads within the same group | A new campaign would add management without changing the offer |
| Several hints describe nearly the same situation | Review them in one group | Extra groups would make a small test harder to interpret |
Creating two ads does not establish a controlled experiment or promise equal delivery. Record what changed and how much exposure each received before drawing conclusions. The measurement guides explain how to judge what the numbers can support.
A worked structure for a furniture retailer
Imagine a fictional retailer selling compact desks. It can deliver in one selected market and wants to learn which space constraint brings useful product visits. The proposed structure is:
Advertiser account: Fictional furniture retailer
└── Campaign: Compact desks / selected market / initial test
├── Ad group: Small spare room
│ ├── Hints: choosing a desk for limited floor space
│ ├── Ad: Compact dimensions
│ └── Page: Desks with width and depth comparisons
└── Ad group: Shared living room
├── Hints: fitting a work area into a shared room
├── Ad: A workspace that fits
└── Page: Desks shown in shared-room layouts“Page” and “Hints” in this sketch are planning notes, not extra levels in the platform hierarchy. The actual destination belongs to the ad; the hints belong to its group.
Both groups concern the same product category, but they answer different questions. A person comparing desk depths needs measurements immediately. Someone planning a shared room may need layout photographs before choosing a size. See complete ad examples for how to connect that distinction to the copy.
If the available budget cannot support learning from both groups, start with the more important need. A tidy diagram is not a reason to fund an unnecessarily broad test.
Where the AthillyAds Brand book and offers fit
AthillyAds has preparation steps before those platform objects. The Brand book stores the business facts and voice you review. An offer identifies what you want to promote. The resulting campaign plan contains proposed intent groups, hints and ad cards for review.
An offer is a planning object inside AthillyAds, not another OpenAI account level. A completed plan also does not mean an ad is approved, connected or delivering. Direct launch requires an eligible advertising API connection; an exported draft still needs the appropriate submission process.
Use the Brand book to correct a business fact that affects many ads. Edit the offer when the product, service or proposition changes. Edit the individual card when only its wording or destination needs work. The AthillyAds features overview explains that workflow.
Name the parts so someone else can review them
A name such as compact_desks_se_small_room_v1 can be more useful than Campaign 7. Choose a convention that includes the offer, market and test idea without recording personal information or guessed facts about individual users.
Keep a small change log with the date, the affected campaign or group, what changed and why. Use the same stable names in your UTM tracking convention. This makes a later analytics review easier even when the people handling copy and measurement are different.
Before launch, read every group against its ads and destinations. Remove anything you cannot explain, confirm the operational settings, and complete the pre-flight checklist. A structure earns its place when it makes the next decision clearer.