A practical guide from Allieta
Start with the questions customers already ask. Give each answer a current, customer-safe source, remove contradictions, and test the reply before launch. A well-maintained set of useful documents matters more than uploading everything.
Choose sources around customer questions
Collect questions from recent support conversations, sales enquiries, and your existing FAQ. Group them by the job the customer is trying to do: choose a product, get started, understand a policy, or ask for help. Pick one reliable source for each answer.
An online shop might begin with delivery regions, return conditions, product compatibility, and a contact route. A software business might begin with plan differences, setup instructions, and common account questions. These are starting examples, not evidence that Allieta reads your order or account system.
- Does the source answer a real customer question?
- Is this the current approved version?
- Can a customer safely receive this information?
- Is someone responsible for keeping it accurate?
Make each explanation complete enough to stand alone
Write the direct answer first, then its conditions and exceptions. Replace vague phrases such as ‘usually available’ with the actual policy your team is prepared to honour. Explain product names and abbreviations that a new visitor would not know.
Keep related conditions together. A return policy that describes a deadline in one document and exclusions in another can lead to incomplete answers. If the same rule appears in several places, agree which version governs and update the rest.
Separate general information from individual decisions
A public refund policy explains the rules; it does not establish whether a specific purchase qualifies. Identify which questions need account access, identity checks, or human judgment. Your knowledge base should explain the next step without pretending the AI has completed those checks.
Leave out passwords, API keys, confidential contracts, and private customer records. In Allieta, these sources supply a customer-facing chatbot. Do not assume uploading a document creates a separate permission boundary for each visitor.
Test answers and the gaps between them
Use the playground with the words your customers use, including misspellings, vague phrasing, and follow-up questions. Compare the response with the approved source. A response that sounds fluent can still omit a condition.
Include a question the content cannot answer and a question based on an old policy. Record what should happen: a correct explanation, a clarifying question, or a route to a person. In Allieta, inspect the retrieved passages to help diagnose the result.
Make content changes part of the support routine
When a product or policy changes, update the original information and the connected source. Refresh affected website sources or replace uploaded documents, then rerun the related questions. Do not assume an edit to the original document has already reached the chatbot.
Review unanswered questions alongside real conversations. Decide whether the gap needs better content, a clearer customer question, or a human process. Assign an owner and a next review date so the same gap does not return.
A source-review worksheet
Copy this structure into your team’s review document and fill it with your own evidence.
| Question | Approved source | Check before launch |
|---|---|---|
| Where do you deliver? | Current delivery policy | Regions, exclusions, and any timing conditions |
| Can I return an opened item? | Current returns policy | Eligibility, exceptions, and a human route |
| Which plan suits me? | Published plan details | Limits, required capabilities, and billing terms |
| Where is my order? | Account-specific workflow | Do not infer order status from a general policy |