QORPOHow to validate AI-assisted Gameplay Tags in Unreal Engine
Validate AI-assisted Unreal Engine content against the project's Gameplay Tag dictionary, then test whether the proposed tags match the game's actual rules. A tag can be registered and still be wrong for the asset. For a clean review, check registration, allowed tags for the asset type and what happens when the consuming system reads them.
Start with a short list of tags the task may use
Epic's Gameplay Tags documentation describes hierarchical tags registered in a dictionary. A project can add tags through project settings, data tables or native code. Before asking an assistant to populate a batch of items, give it the approved vocabulary for the job and an example of how each relevant branch is used.
Say the project has a family called Item.Consumable, with Item.Consumable.Healing and Item.Consumable.Buff below it. The task is to draft tags for healing items from approved design data. Ask the assistant to use the existing healing tag for unambiguous rows and flag unclear rows for review. An item described as a stamina tonic is a good reason to stop and ask which existing category applies. A tag that sounds plausible can still make a filter return the wrong items.
For a broader review of generated changes, see the QORPO guide to reviewing AI-generated game code. For tags specifically, start with one asset family and one consumer. That keeps the first pass small enough to check by hand.
Validate registration and asset rules separately
Run the first check against the dictionary. Catch spelling errors, unknown branches and values that came from another project. For proposals to introduce genuinely new tags, make the intended consumer and authoring owner visible in the review. The dictionary changes alongside the content only when that relationship is clear.
The second check asks what is allowed on this kind of asset. A consumable definition can hold a registered quest tag in a field intended for an item family. The field's type may accept it and the project may compile. A project-specific rule can still reject it. Epic's Data Validation documentation describes running validation on individual assets, folders or the project. Keep the error specific enough for a designer to fix, with the asset, field and expected tag family.
Model example: a build has three rows. The bandage has Item.Consumable.Healing and passes. The tonic contains an unregistered tag and fails the dictionary check. The quest key has a registered tag that belongs to the wrong branch for its field and fails the asset rule. Each failure tells the author which decision to revisit.
Test the query with real content
Open the inventory or ability screen that uses the tags. Add one asset that should match, one that should fail and, where relevant, one child tag. Observe the output and check the intended matching mode. In a hierarchical query, a parent category may legitimately include all its children. Another query may need an exact match. The consuming code or Blueprint is where that difference becomes visible.
Use the same three cases in later reviews so a policy change exposes its impact early. A fixture should record the expected player-visible result as well as the data. The guide to writing expected results for game tests explains how to describe that output before generating a batch of inputs.
Handle existing assets when the vocabulary changes
A rename is a migration. Find the assets and queries using the old tag, including content already present in a build. Review the project version's tag redirect and migration behavior, then test representative saved content before removing the old value. A clean import of new assets leaves the older path untested.
When reviewing a large proposed change, sort by tag and scan the distribution. A sudden concentration of rows in one broad category can reveal ambiguous instructions. Compare a handful against the source design and run the consumer check before accepting the whole batch.
What should block a merge?
An unknown tag, an asset in the wrong category, a changed query result with no approved design change, or an unverified rename that affects existing content should hold the merge. Give the reviewer a small asset diff and a short list of expected results. That makes it possible to spot an intent error without reading every generated row. Browse the QORPO development articles for adjacent game production checks.