Data quality for generative AI: what changes when you feed a chatbot your documents
Many companies connect an AI assistant to their documents so it answers with their own information (the technique is called RAG). AI does not improve what you give it: if documents are outdated, duplicated or poorly protected, so are the answers. Here are the quality controls worth applying.
The idea in one sentence
It is like giving a very fast intern a shelf of filing cabinets: they answer in seconds, but from whatever is in the cabinets, including old papers.
Six checks before connecting documents
- Owner: each collection has a person who answers for it.
- Currency: current version identified; obsolete ones removed or flagged.
- Duplicates: no conflicting versions of the same document.
- Permissions: the AI only retrieves what the asking user may see.
- Personal and confidential data: identified and handled per your policies.
- Traceability: each answer states which document it came from.
After go-live
- Review a sample of answers each month.
- Collect user reports of errors.
- Update the collection when a policy or price changes.
Common mistakes
- Loading “the whole SharePoint” unfiltered.
- Ignoring each document’s permissions.
- Not telling users they are talking to an AI or that it can be wrong.
Frequently asked questions
What is RAG?
The AI searches your documents and answers based on them.
Does it eliminate hallucinations?
It reduces them but does not eliminate them: cite sources and allow verification.
Informational content, not legal advice.
Which quality rules must your data meet?
Data quality rules: an editable template. €39. Support template, not legal advice.
View product
What's your Data Governance maturity?
Free assessment with your priority gaps, plus the self-assessment quiz and savings calculator on the Data Governance path.