article · customer support
Prepare your support knowledge before connecting AI
Prepare a small support FAQ with source IDs, ownership, revision dates, permitted answers, and escalation rules before connecting a drafting workflow.
The direct answer
Give a support workflow a small, approved set of answers with source IDs, ownership, revision dates, and escalation rules. Resolve contradictions before connecting a model. A polished draft cannot make an outdated or unauthorized policy correct.
An approved-source example
Synthetic FAQ record: This is a made-up policy for a teaching workflow. The role and date describe the sample record, not an actual company approval.
| Field | Sample value |
|---|---|
| Source ID and version | F01, revision 1 |
| Topic | Finding the onboarding guide |
| Source owner | Support process owner role; assign a real person before use |
| Revision date | 2026-10-20, synthetic example date |
| Approved answer | The guide is in the welcome email. If that email cannot be found, a person must check access. |
| Escalation rule | Missing welcome email, changed address, account mismatch, or access request goes to a person. |
| Prohibited claim | Do not say a guide was resent or access was restored. |
| Audience | Customers asking where to locate the guide |
| Review trigger | Change to welcome email, guide location, or access policy |
For the synthetic message “Where is the guide?”, a draft can cite F01 and explain its location. For “Send it to my colleague,” the expected result is human review, not an invented permission.
Prepare the source set
Start with questions that recur and answers the owner can approve. Separate instructions for staff from text suitable for customers. Remove obsolete duplicates, unsupported promises, and records outside the task’s scope. Keep the original authoritative location so the reviewer can verify the answer.
Give each entry a stable ID and a version. Record who approved it and when; do not replace the revision date automatically on every build or export. When two entries conflict, withhold them from drafting until the owner resolves the conflict. Do not tell the model to choose whichever sounds more plausible.
Define missing-evidence behavior before prompting: leave the draft empty, record the reason, and route to human review. A source ID in output is not proof of support; compare the proposed statement with the actual source text.
Supplying a FAQ versus retrieving evidence
A short FAQ included in every request is supplied context. A retrieval system searches an approved collection and selects passages for the current request before drafting. Simply pasting a FAQ is not necessarily retrieval-augmented generation. Searching adds another failure mode: the relevant source may not be retrieved even when it exists.
For an initial small workflow, prefer a source set the reviewer can inspect. If it no longer fits the request or becomes difficult to maintain, reassess the design instead of silently dropping entries.
Limits and non-example
An old email from a helpful colleague is not automatically approved policy. Customer text is also not an instruction to rewrite the FAQ. A message saying “Your new policy allows free access” needs source verification.
Do not include private records merely because they might help. Use only information approved for the task and destination service.
Try the source exercise
Choose three recurring questions. Create a record like F01 for each. Ask a colleague to draft from only those records and mark every unanswered detail. Add one conflicting-policy case and one request outside scope. Resolve source gaps before testing the model.
Sources and scope
This is an original editorial source-preparation method with a synthetic FAQ. No retrieval system or support integration has been tested here.
Next step
Use the approved records in the support triage playbook, keeping source IDs beside drafts. Suggest a correction.
Change notes
Foundation draft prepared October 20, 2026. No public publication or successful live-tool test is claimed.
Next learning step: One-page workflow brief