Test answers with realistic customer questions
Use a repeatable set of questions to check relevance, context, links, and handling of uncertainty.
Create a small, repeatable quality check that catches weak answers before customers do.
On this page
Build a small test setCover different situationsEvaluate the whole responseCorrect and retestBuild a small test set
Choose real buying questions rather than only greetings or broad prompts. Keep the expected facts and source URLs beside each case. Test in the same language and widget configuration as the storefront.
Cover different situations
Use these patterns and adapt them to products you actually sell.
| Case | What a useful answer should do |
|---|---|
| Exact product specification | Identify the model and explain the requested attribute using the relevant product information. |
| Follow-up question | Keep the product context when the customer asks ‘Does it also work by cable?’ |
| Current delivery or price | Use current shop information, distinguish variants, and avoid making up an unavailable date. |
| Ambiguous reference | Use available conversation or page context and ask a short clarification only when needed. |
| Unknown information | Explain the useful part, identify the unknown fact, and offer a specific next step. |
| Unsuitable product | Explain the mismatch and suggest a relevant alternative only when supported. |
Evaluate the whole response
Check the answer, product links, variant, language, and tone. A technically correct sentence can still be unhelpful if it ignores the customer's main question. A polite apology does not replace an answer or a useful next step.
Correct and retest
If a fact is missing, improve the source or connect the appropriate additional page. If the source is correct but the answer is poor, use the available answer-improvement workflow and include the expected wording and evidence. Repeat both the failing case and a related case after the change.
Still need help?
Tell us what you expected and what happened.