AI Approach
Evidence-grounded AI in regulated pharmaceutical work
Outputs without citations cannot be reviewed. A practical standard for using AI in pharmacovigilance, regulatory, labelling and quality operations.
COMPLIVISE Editorial · 19 June 2026 · 5 min read
In a regulated environment, an AI output that cannot be traced to a source is not an answer. It is a claim.
Three non-negotiables
1. Every output carries its source. A proposed classification, extraction or comparison must show the document, the page and the passage it came from. Reviewers should be able to disagree in one click.
2. A person holds the decision. The system proposes; a qualified reviewer accepts, edits or rejects. That action is recorded with the reviewer's identity and timestamp.
3. The record survives the model. If the model is replaced next year, the decision history, citations and approvals remain readable exactly as they were.
Questions worth asking a vendor
- Can I see the passage behind this output, without leaving the screen?
- What is recorded when a reviewer overrides the system?
- How is the output affected when the source document is superseded?
- Can the audit trail be exported independently of the vendor?
If a demo cannot answer these in the product itself, the answer is process, not software.
What good looks like in practice
A literature record with an inline citation panel. A label comparison that shows the exact CCDS clause next to the country text. A quality deviation whose root-cause summary links to the investigation attachments it drew from.
None of these are exotic. They are the difference between an assistant you can defend in an inspection and one you cannot.