↑
← All articles

AI and workflow

Show when an AI answer's source was last reviewed.

Make source ownership and review dates visible so people know when an AI-assisted answer needs a fresh check.

A hypothetical operations team uses an AI assistant to answer internal questions about expenses, leave and supplier approvals. The answer sounds clear, but the policy behind it changed three months ago. The colleague reading it cannot see which source was used or when anyone last confirmed that source was current. An AI workflow should not make old guidance look newly reliable. Show the source, its owner and its review position close to the answer, then define what happens when that evidence is missing or stale.

Attach evidence to the answer

For each answer, identify the document, record or approved knowledge entry that supports it. A generic link to a shared folder is not enough. People need to reach the relevant source and understand whether it is authoritative for the question they asked.

Display a meaningful review date rather than only the file's last modified date. A formatting change does not prove that the policy was checked. Record who is responsible for the source and when they last confirmed its substance.

Define what counts as too old

Freshness depends on the subject. A process that changes weekly needs a different threshold from an annual policy. Agree a review interval for each source type and identify events that should trigger an earlier check, such as a regulatory change, supplier change or revised approval route.

Do not quietly continue when a source passes its threshold. The workflow might flag the answer, limit it to general guidance or return the question to a named person. The right response depends on the consequence of being wrong.

  • Which approved source supports this answer?
  • Who owns its accuracy?
  • When was its substance last reviewed?
  • What should happen when the evidence is missing or out of date?

Test the control with safe examples

Create synthetic questions that use a current source, an expired source, two conflicting sources and no source at all. Write the expected behaviour before running the test. Check whether the person can see the evidence and whether the workflow stops or escalates as designed.

Repeat the test when sources, instructions or the AI service change. A control that worked for one version can weaken when content is reorganised or the workflow starts retrieving information differently.

Make review part of normal ownership

Give overdue sources a visible queue and an accountable owner. Remove or quarantine material that is no longer approved rather than leaving the AI workflow to choose between old and new versions. Keep a short record of the review decision so future changes have context.

This does not guarantee that every answer is correct. It gives colleagues evidence to judge, a clear warning when confidence should fall and a route back to an accountable person.

Rethink's AI review and enablement work can help teams define practical controls around one real workflow before expanding its use.

Explore AI review and enablement →

Work with Rethink

Discuss your
project.

Talk to Rethink