An AI support answer can become wrong even when the model has not changed. A price moves, an access rule is replaced or a policy gains an exception. The old answer may still sound confident because its source remains available.
Keeping answers current starts with source ownership, not prompt polishing. Each material change needs a controlled update, regression questions and a safe fallback.
Separate each fact from the system that proves it
List the facts an automated answer may use, then name the authority for each one. A service page may prove standard scope, a fee schedule may prove price and a booking system may prove one customer’s availability.
Do not combine those authorities into one undated document. A static article cannot prove a live appointment, and a live record does not explain the general cancellation rule. The knowledge-base gap guide helps classify missing, stale and authority-dependent answers; this workflow begins after a specific source change is known.
Microsoft Learn’s current knowledge-source guidance warns that sources may evolve and distinguishes verified official sources from material that needs caution.
Keep a small source freshness ledger
Create one row for every source that can materially change a customer answer. Keep the ledger short enough for an owner to maintain.
- 1ConfirmVerify the authoritative change.
- 2MapFind dependent answers and questions.
- 3ReplaceRetire the superseded source.
- 4RetestRun focused regression questions.
- 5ReleaseKeep evidence or use the fallback.
| Ledger field | What to record | Why it matters |
|---|---|---|
| Authority | The page, system or person that proves the fact | Prevents convenient copies becoming truth |
| Owner | The person responsible for approving a change | Gives the update a destination |
| Last verified | Date and evidence used for the check | Shows what was actually reviewed |
| Change trigger | Price, policy, service, process or access event | Starts review when the business changes |
| Affected questions | Real customer questions that depend on the fact | Defines a focused regression set |
| Safe fallback | Handoff or limited wording while evidence is uncertain | Stops an old answer sounding final |
NIST’s Generative AI Profile discusses provenance records with source, date, time and modification information. The ledger applies that idea to daily support work.
Trigger review from business changes
Scheduled reviews can catch quiet drift, but important updates should start when the source changes.

A professional-services team might trigger review when it changes a fee, service boundary, eligibility rule, document requirement or access process. The owner confirms the authoritative change; the support owner identifies every answer and test question that depends on it.
Retire superseded material instead of leaving two versions available. Record the effective date and transition conditions. A rule that applies only to newer contracts must ask for that context.
Retest the questions that could change
Update the source, then test the decision around it. A price change should cover the standard price, an old quote, a currency question and a request for an exception. An access change should cover the new method, an old credential, a failed attempt and a human-help request.
For each question, record the source retrieved, answer, preserved conditions and route when evidence is missing. Use authorized test records, not customer data. The AI agent release gate provides a broader framework; the website chatbot buying test covers platform controls and evidence.
A passing answer uses the current authority, preserves limits and hands over when a required fact is unavailable.
Use a safe answer while the update is uncertain
When two sources conflict, pause the affected automated answer. Do not let the newest-looking file win automatically. Ask the responsible owner which source governs and preserve the conflict as evidence.
The temporary reply should say what is verified, what remains uncertain and who will check it. Price, eligibility, refunds, access and promised service deserve human review until source and test results agree. The AI reply review guide helps choose send, edit, verify or handoff.
Close the change with evidence
An update is complete when the old source is retired, the new authority is identified, affected questions pass and an owner accepts the result. Keep a receipt with the source, effective date, tests, fixes, limits and reviewer.
The shared-inbox ownership model keeps the next action visible during takeover. A configured DripTell AI workspace can support replies from approved knowledge, but the business owns source truth and approval.
Frequently Asked Questions
How often should AI support knowledge be reviewed?
Use both scheduled and event-triggered review. Set a cadence based on consequence and change frequency, then review immediately after a material price, policy, process, service or access change. There is no universal interval for every source.
Should every source change trigger a full AI retest?
Retest the questions and workflows that depend on the changed fact, plus a stable critical set. Run a broader release test when the source type, retrieval method, permissions, model behavior or handoff path changes.
What if two approved sources disagree?
Pause the affected answer, identify the authority owner and resolve the conflict. Until then, state only verified facts and hand the decision to a person. Do not merge incompatible wording into a confident compromise.
Want help applying this guide?Ask the DripTell team
From reading to action
Want help applying this to your business?
Share your details. We will read the article you are on and start with a useful answer for your team.
DripTell Editorial
Practical guidance reviewed by the DripTell product and customer workflow team.
See how DripTell checks product claims, uses primary sources and handles corrections.
Editorial and source policy



