How do I use AI with a large Catalog without losing control?
We currently recommend starting in Build & Manage with AI before creating the items: give it the product or service list and business context, let it propose Folders, Semantic Tree / Classification, Tags, AI Search Attributes, and Search Names, review the structure, then let it create and place the items in reviewable batches.
- Let AI analyze the whole set before item creation.
- Review the proposed structure and business facts, then apply bounded Proposals.
- After Publish and indexing, test the result with realistic customer questions.
Recommended order for a large file or list
- 1
Give AI the source containing the products or services, useful business context, and the goal of the Catalog.
- 2
Ask it to analyze the full set and propose Folders, the Semantic Tree, and Tags before creating items.
- 3
Ask it to propose AI Search Attributes, values, Aliases, and Search Names that support precise questions without inventing unsupported business facts.
- 4
Review and adjust the proposed structure and facts where human input is needed.
- 5
Once approved, let AI create the items and place them into the right structure and values in reviewable batches. The current Catalog AI contract accepts at most 100 items in one logical proposal.
- 6
Review → Publish → allow required indexing to finish → test in Test Assistant.
Example: a 1,000-item Catalog
Having 1,000 items does not mean sending 1,000 direct edits in one operation. Split the work into clear batches so a mistake can be stopped, reviewed, and retried without putting the entire Catalog into one change.
After each batch
- Confirm the intended items were created or changed.
- Monitor any required reindexing.
- Test one natural discovery question.
- Test an exact lookup or count when you added AI Search Attributes.
- Start a fresh Test after an important change so old conversation context does not hide the result.
Frequently asked questions
Do I need to create 1,000 items manually?
No. AI can help organize and prepare changes, but large work should still be split into reviewable batches rather than one unreviewable one-shot operation.
Can AI invent a Search Name or a missing value?
AI may suggest a search-friendly form, but you should confirm it represents the same real fact. Missing business values should stay missing until you have a reliable source.