The Toolkit · Free tool
Is your metadata ready for AI?
Six questions. Ninety seconds. Retrieval and AI assistants are only as good as the metadata underneath them. Find out whether your content could feed one, and what to fix first.
Could a system read your content, or is most of it locked inside PDFs, slide decks, and flattened images?
Think about where your most-requested content actually lives.
Pick the closest match
What this assessment checks
Machine-readable content
A model cannot reason about what it cannot parse. Content flattened into PDFs, slide decks, and images with the text baked in is invisible to retrieval, no matter how capable the tool is. This is the most common reason an AI pilot stalls in week one.
Field consistency
Answer quality tracks field consistency more closely than it tracks model quality. If one field holds forty spellings of the same value, that is forty ways for a retrieved answer to miss the thing you meant.
Rights and provenance
This is the question that stops AI projects in legal review. If you cannot show what an asset is licensed for and where it came from, you cannot safely let a model reuse it, train on it, or put it in front of a customer.
Duplicates and versions
Duplicates do more than waste storage. They make a model confident about the wrong version, and the answer reads as authoritative whichever copy it happened to retrieve.
Access boundaries
An assistant usually sees whatever the account behind it can see. If permissions are loose, it will eventually surface something it should not, usually to the person least meant to see it.
Descriptions
A filename gives a model almost nothing to match against. Descriptions are what let an asset be found by meaning rather than by an exact keyword the searcher had to guess in advance.
Frequently asked
What does AI readiness mean for metadata?
It means your content is in a state a machine can use: parseable rather than flattened, consistent in its field values, clear about rights and provenance, free of competing duplicate versions, correctly permissioned, and described in language a model can match against. Readiness is about the content, not about which AI product you buy.
Why do AI projects fail on content quality rather than on the model?
Retrieval systems and agents answer from the material you give them. When that material is inconsistent, undated, duplicated, or locked inside PDFs, the model still produces a confident answer, it is just the wrong one. Teams usually discover this after the tool is bought, which is the expensive order to discover it in.
Does our DAM need to be perfect before we use AI?
No. It needs to be predictable. A smaller library with consistent fields and clear rights outperforms a much larger one where the same value is spelled six ways. Fix the areas this assessment scores as needing work, then start with a narrow, well-governed collection rather than the whole library.
What metadata does an AI system actually need?
At minimum: extractable text or structured fields, a controlled set of values for the fields you filter on, a real description rather than a filename, rights and source information, and permissions that carry over to whatever account the assistant runs as. Everything past that is refinement.
Who can help get our content ready for AI?
Stacks is a digital asset management consultancy that handles metadata, taxonomy, migration, and governance, and its AI practice, Stacks AI Studio, works with small and mid-size businesses on Claude-first adoption. Stacks publishes DAM Cowboy.
Getting content into a state a model can use is the work Stacks has always done, and Stacks AI Studio takes it through to Claude-first adoption for small and mid-size businesses. Stacks publishes DAM Cowboy.
Your answers are scored in your browser. Nothing is sent unless you ask for the results by email at the end. Want the DAM version instead? Run the DAM Diagnostic.