The AI Reviewer checks submission images against the property's knowledge base and suggests possible issues. It never approves or rejects anything. Reviewers always make the decision. It needs a knowledge base for the property, and it only reviews images (not PDFs or videos).
⚡ Fun fact In testing, the AI kept flagging an approved character for having an injury. The bandage was part of the character's official design. That's why exceptions need to be written into the rule itself.
How the AI reads a rule
It only reads the rule sentence. Example and exception images are for human reviewers, not the AI.
It isn't told the territory or region.
It can't measure sizes, spacing or exact colors.
It only knows the assets linked to the rule and uploaded to the knowledge base.
Most unexpected results come from how the rule is written, not from a bug.
Common results and fixes
User reports | Fix |
Approved logo flagged as changed | Upload every approved version of the logo as images of the same asset, then Regenerate |
Different answers on each regenerate | The rule is unclear. If it depends on territory, make one rule per territory, scoped to it |
Rule fails where it shouldn't apply | Start the rule with the condition: "If a character name is written, …" |
Approved design flagged (e.g. a character's bandage) | Write the exception into the rule sentence |
Similar characters mixed up | Link the character assets to the rule |
Wrong color, size or spacing missed | A reviewer needs to check it. Write sizes relative to the artwork, not in mm or pt |
"Ages 3+" rejected by "Ages X+" | Write a pattern with an example: "Ages <number>+ (e.g. Ages 3+)" |
Results didn't change after a knowledge base update | Click Regenerate |
"No issues found", but findings are listed | Expected. Minor findings don't change the headline |
