How to keep customer-support answers up to date

Every reference answer should have a source, an owner, a version and validity conditions. Changes to prices, opening hours or processes need to reach the channels using that information. The FAQ knowledge base is a maintenance process as well as a collection of texts uploaded to a tool.

By Lyntev. Updated on .
About 5 minutes to read.

The answer starts with the sourceApproved information, defined access and a verifiable answer.

Translation published on . Original Portuguese guide; source revision .

By the end, you can:

  • Distinguish drafts, approved answers and expired content
  • Assign authority and review to each topic
  • Publish and roll back a change with verifiable tests

Find the versions already in circulation

Look for FAQs on the website, quick replies, internal documents, automatic messages and information used by assistants. Identify the same topic with different wording and check whether the conditions match. An answer can be correct in a recent document and remain wrong in the shortcut reception uses daily. The inventory should cover points of use, not just the central folder.

Start with topics having greater consequences: prices and inclusions, opening hours, appointments, cancellations, deliveries or entry requirements. Do not copy personal conversation data to build the knowledge base. Turn recurring questions into general wording and keep examples fictitious. Reference information should help answer questions without becoming an indiscriminate archive of what customers said.

Assign an owner to each subject

Define who can confirm facts for each topic and who approves publication. The person who writes best may not have authority to change a commercial term. If there is doubt, the answer remains pending until clarified, rather than being completed by AI with an assumption. Also identify who covers for the owner during absences.

Write the answer’s boundaries: which service, audience, period or situation it applies to. An old policy for one product must not be presented as a rule for all the others. When a condition depends on assessment, the knowledge base should explain that dependency and referral, preserving the distinction between public information and a decision about a specific request.

Separate drafting, approval and publication

Use statuses showing whether a text may already be used in support. A draft may contain options and questions; the published version needs validation. Record the change and retain the earlier reference for the period operations require. If there is an end date, prepare expiry handling before the answer stops being true.

Knowledge tools, such as the one documented by Microsoft, allow work with versions and publication. The principle can be applied in a simpler structure, but do not assume your platform controls this automatically. Confirm where the approved version lives and how the team distinguishes a current answer from one left in an old file.

Separate drafting, approval and publication
StatusMay it be used to answer customers?Condition for progressing
DraftNot yetFacts and scope reviewed
ApprovedOnly through the defined processPublication destination and timing confirmed
PublishedYes, within scopeUse and validity monitored
Expired or replacedNot as a current conditionAlternative or referral identified

Write answers with conditions and a next step

A reference answer should resolve the question directly and state what depends on confirmation. Add the fact’s internal source and referral instructions when no conclusion can be reached. Avoid turning the knowledge base into a catalogue of promotional phrases repeating advantages without helping the team decide what to say.

Include equivalent questions in customers’ language without duplicating the answer for every variation. “Can I change the time?” and “I need to reschedule” may point to the same procedure. If context changes the result, state which information is missing. AI must not treat a general answer as authorisation to confirm an exception the company has not yet approved.

Work through a change using a fictitious scenario

Estúdio Largo, a fictitious example, changes its support hours for a holiday period. The responsible person confirms the dates and prepares a new answer version. The team tests questions about opening, appointments and out-of-hours contact, including dates before and after the period. Only then does it publish to the identified points of use.

The website, reply shortcut and AI knowledge base need to show the same currently applicable condition. At the end of the period, the seasonal answer expires or is replaced according to the plan. The scenario demonstrates maintenance, not a sales result. Evidence of completion is the information observed in every channel, not merely a “saved” status in the main document.

Test the change and prepare a way back

Create questions covering the normal condition, time boundary, an exception and a request without enough information. Compare the answer against approved facts. If you use AI, observe numbers, negatives and handover because fluent wording can change a condition. Keep the test set to repeat when the model or source changes.

Define how to roll back an incorrect version and who may do so. Removing text from a file does not guarantee it has disappeared from caches, shortcuts or integrations. Confirm the tool’s actual mechanisms and propagation to points of use. If an update fails in one channel, limit the answer at that destination until the correct condition is available.

Use gaps to improve the knowledge base without blind learning

Keep a queue of unresolved questions with topic and minimum context. The team decides whether there is a general answer to add, whether customer information was missing or whether the matter always requires individual assessment. Do not automatically publish everything AI generated or everything a professional answered in a private conversation.

Review error reasons and how often an answer is used, without confusing frequent use with quality. A repeated question may indicate that the answer is unclear or difficult to find. The knowledge base improves when changes have authority and tests, not when it accumulates ever more contradictory text.

Keep a worksheet for each important answer

The worksheet should let you find the fact, owner, validity and every publication destination. Avoid identical review schedules for all topics: a temporary price may need different attention from a stable explanation. Record events requiring review, such as a change of product, contract, hours or process.

Bring support and content owners together periodically to resolve priority gaps. A small, current list is easier to control than a large unversioned knowledge base. This guide focuses on ongoing governance; initial source preparation and detailed assistant trials can follow the related guides while preserving one reference version.

Worksheet for your project

Question and language variations:
Reference answer and scope:
Source of facts and owner:
Status, version and approval date:
Start and end of validity, if applicable:
Conditions requiring referral:
Points of use: website, shortcuts, team and AI:
Test cases and observed result:
Publication method and confirmation at every destination:
Rollback plan:
Next review or event triggering it:

Sources and criteria

Content prepared with AI assistance and research using technical sources, with checks of the relevant examples and calculations. The references support the guide’s decisions; examples identified as fictitious let you try out the method.

Wellinton Oliveira

Written by

CEO of Lyntev

Technology with context, original analysis and sources you can consult.