A collection can sound precise in a ChatGPT ad while remaining loosely defined in the catalog. “Narrow desks” might mean a maximum measurement to the editor, a supplier label to the catalog team and any small furniture to the person adding new products. The same collection can gradually begin making different promises.
Write membership rules before sending paid visitors to it. Another colleague should be able to assess a new product and reach the same decision as you. This work concerns entry into the selection. How qualifying products are subsequently grouped around customer tasks or otherwise presented is a separate task.
Make the advertised promise testable
Start with one sentence describing what the products share. Translate its decisive words into factual conditions. If the ad states a maximum measurement, define which measurement, which unit and whether the boundary itself qualifies. A loose adjective such as “compact” is insufficient as an operating instruction.
In a hypothetical example, a shop advertises desks no wider than 100 cm. The rules could require that an item is intended as a desk and that the selected variant does not exceed the width limit. A console table measuring 90 cm passes the measurement condition but fails the product condition. A desk measuring 120 cm passes the product condition but fails the measurement.
These measurements are invented examples, not recommended product standards. The point is to make every necessary condition independently checkable. If the ad promise cannot be translated into reasonable rules, clarify it before adding more products. If the destination type remains undecided, start with choosing a category or product page.
Separate requirements from preferences
A requirement determines whether the item belongs. A preference can influence the order in which qualifying products appear. Do not confuse them. If drawers are convenient but not promised, a desk without drawers should not be excluded by a hidden rule. If the collection explicitly promises drawers, every included variant must meet that condition.
Use four fields in the working record: attribute, condition, source and treatment of missing information. The final field matters. An empty width is not zero centimeters and should not accidentally admit an item into a maximum-width selection. Place it in a separate review queue until the information has been checked.
Also decide which source takes precedence when information conflicts. If the product description and structured catalog disagree about dimensions, resolve the underlying error. Do not choose whichever value makes the product available for advertising. A boundary rule supports an accurate selection; it is not a device for filling the page.
Decide whether the rule applies to models or variants
A product family can include multiple widths. Having one narrow version does not make every version suitable for the collection. Define the level being assessed and what customers should encounter when they open the product card. The default variant must make the collection’s boundary understandable.
In the desk example, a model could come in 90 cm and 120 cm versions. If the smaller variant qualifies the model for inclusion, its link and card information should clearly refer to that variant. A card showing the smaller version’s starting price alongside a description of the larger version confuses the rule even if the internal catalog can explain it.
Apply the same review to accessories, bundles and products that only partly satisfy the promise. A separate desk leg is not a desk. A bundle may need its own assessment even when one component meets the dimension limit. Record these boundary cases so the next import does not recreate the same ambiguity.
Compare the saved selection with the rules
OpenAI’s product-feed guide describes product filters and clarifies that matching does not guarantee impressions. Treat filtering as a technical selection that must implement the editorial rule. Do not assume a store category name expresses every condition in the ad promise.
Ask the integration owner to show the products actually included in the saved selection. Inspect an item that should certainly qualify and one that should certainly fail. Add a case exactly on the boundary and one with missing information. This small review set reveals different problems from checking a few attractive products at the top of the list.
Review the advertising wording at the same time. OpenAI’s campaign documentation describes creative content as an ad object. A correct catalog selection therefore does not repair an overbroad promise. Keep wording and membership conditions aligned even when different people own them.
Boundary checks for the narrow-desk collection
- Desk, 100 cm
Qualifies if the other stated conditions are met. The boundary is inclusive.
Checkpoint - Desk, 120 cm
Excluded because its width exceeds the advertised promise.
Reconsider - Console table, 90 cm
Excluded despite its width because the product type does not qualify.
Reconsider - Desk, unknown width
Needs review before membership can be decided.
Do not allow invisible exceptions
A commercial wish to include a popular item is not a reason to quietly bypass a requirement. Either the product meets the rule or the collection promise needs to change. Another option is to present the item in a separate, clearly labeled recommendation outside the promised collection.
Describe the boundary on the page in customer language. Important restrictions should be understandable without reading internal filter expressions. The category introduction guide helps with that wording. Technical working notes may be longer than the customer explanation, but their meaning must agree.
Finally, assign someone to review membership when attributes, assortment or ad promises change. Retain the reasons for borderline entries and removals. Use the broader product-feed guide for source data and catalog responsibility. A clearly bounded collection can evolve without making customers guess why a product appears behind the ChatGPT ad.
Sources and scope
Define and review the rules deciding which products and variants can enter or must leave a collection promoted through paid ChatGPT ads.
Working methods and examples are editorial suggestions. Check current platform requirements and available features before implementation.