Inconsistent AI metadata generation behavior in AWS DataZone for Column Names

Question

Hi team,

I’m observing a strange behavior with AI-generated metadata in AWS DataZone.

We have two tables with similar structures (Both Glue Tables with no specific difference). For one of them, the AI metadata generator produces good business-friendly column names, but for the other, it keeps only the technical column names (even though the same column names were interpreted perfectly in the first table).

What’s more confusing is that within the same table, some columns are assigned proper business names while others retain their technical names — even though similar columns in another table were correctly enriched.

We’ve tried:

Running the metadata generator multiple times

Using different machines and users

But the behavior remains inconsistent.

Could you please help us understand why this inconsistency occurs and how we can fix or stabilize the AI metadata generation in DataZone?

Thanks in advance!

— Reena
![Table1](/media/postImages/original/IMVO0aK7mYRcC9_30usAJ-4A)

![Table2](/media/postImages/original/IM3lYCRDOxT2OSh_L7-Hg3NQ)

Taz · Answer

Hey,

Hope you're keeping well.

AI-generated metadata in AWS DataZone can vary because the model relies on available schema context, sample data, and any existing business glossary terms in your DataZone domain. If a Glue table has sparse or ambiguous column data types, limited profiling results, or no glossary matches, the enrichment may default to the technical name. To improve consistency, make sure both tables are profiled in Glue with representative data, and consider adding or mapping business terms in your DataZone glossary so the AI generator has stronger hints. You can also review the DataZone project settings to confirm the same glossary and profiling configurations are applied to both datasets.

Thanks and regards,  
Taz

Inconsistent AI metadata generation behavior in AWS DataZone for Column Names

Add your answer

Relevant content