What Questions Should a Website Chatbot Be Able to Answer?

A website chatbot should answer frequent, factual and low-risk questions that the business can support from approved information. It should recognise when the answer depends on judgement, current account data or sensitive circumstances and pass those conversations to a person.

Limit its scope to questions it can answer dependably, and give it a way to pass on the rest.

Start with real customer questions, not a generic list supplied by the software vendor.

Begin with questions customers already ask

Collect questions from:

• Enquiry emails and contact forms.
• Telephone calls.
• Live-chat transcripts.
• Sales and support staff.
• Search Console queries.
• Website searches.
• Appointment preparation messages.
• Complaints and failed journeys.

Group repeated questions by purpose. Customers may be checking suitability, price factors, timing, location, preparation or the next step.

If the answer belongs clearly on a public page, improve that page as well as the chatbot knowledge. The bot should not become the only place where essential information exists.

Thinking about your own rebuild?

Tell us about your business, your current website and what you would like to improve. We will help you work out the next step.

Tell us about your project↗

Service scope and suitability

A chatbot should often be able to explain:

• Which services the business provides.
• Who a service is designed for.
• The locations or areas covered.
• Important exclusions.
• The difference between similar services.
• What information is needed to assess suitability.

Answers should be precise without making a final professional judgement where one is required.

For example, it can explain that a survey service covers commercial buildings within a region and list the details needed for a quotation. It should not confirm that a particular building is safe without an inspection.

Opening, availability and response arrangements

The bot can answer current operational questions when it has a reliable source:

• Opening or contact hours.
• Typical response arrangements.
• Whether appointments are required.
• How out-of-hours messages are handled.
• Which channels are available.
• How to cancel or rearrange according to published policy.

Avoid promises that depend on live capacity unless the chatbot reads a current booking or staffing system. “A team member usually replies within one working day” must match the real service and should not be presented as an immediate transfer.

Pricing and quotation questions

A chatbot can explain published prices and the factors affecting a bespoke quotation.

It may answer:

• Whether an initial consultation is free or paid.
• The published starting price and what it includes.
• Which details affect the quote.
• Accepted payment methods.
• Deposit and cancellation terms that are publicly documented.
• How to request a quotation.

It should not invent a price from incomplete information. When costs depend on scope, gather the minimum useful details and arrange human review.

Keep pricing sources current. If the website and chatbot disagree, customers will reasonably question both.

Process and preparation

Customers often need practical reassurance about what happens next.

Useful answers include:

• What the first conversation covers.
• Which documents, measurements or photographs to prepare.
• Who attends an appointment.
• Broad stages of the service.
• How long a published standard process normally takes, with appropriate caveats.
• How customers receive results or follow-up.

These answers can reduce avoidable administration and improve the quality of enquiries.

The guide to information a business website should include helps identify facts that belong on the site itself.

Policies and common administrative questions

The chatbot can point to clear policies and summarise straightforward parts carefully.

Possible subjects include:

• Returns or cancellation procedures.
• Delivery areas.
• Accessibility arrangements.
• Data-use information.
• Warranty or support routes.
• How to raise a complaint.

Link to the complete policy. Do not let a brief chat summary override formal terms or omit important conditions.

Questions that should go to a person

Define boundaries before launch.

Human review is usually appropriate for:

• Complaints and distressed customers.
• Complex suitability decisions.
• Negotiation and unusual commercial terms.
• Legal, medical, financial or regulated advice.
• Safety-critical questions.
• Account access and identity problems.
• Bespoke quotations.
• Vulnerable customers.
• Any answer the bot cannot support confidently.

The handoff wording should be honest and calm. It can explain that the question needs judgement and offer the appropriate contact route.

Questions the chatbot should refuse

Some requests should not be answered or collected.

Examples include requests for:

• Passwords or authentication codes.
• Full payment-card details in ordinary chat.
• Unnecessary health or identity information.
• Confidential information about another customer.
• Instructions to bypass business policies or security.
• Unsupported guarantees.

Configure the bot to stop, explain the safer route and avoid repeating sensitive text unnecessarily.

Read whether customer information is safe in an AI chatbot when setting these boundaries.

Build an approved source set

Answers should come from controlled information rather than the whole internet.

Create a source list containing:

• Current service pages.
• Approved FAQs.
• Public pricing and policy pages.
• Opening and contact information.
• Internal support material suitable for customer use.
• Named owners and review dates.

Remove duplicates and old versions. If two documents disagree, resolve the business rule before expecting the chatbot to choose correctly.

Give changing information a review schedule. Opening hours may need frequent checks; a stable explanation of the service process may change less often.

Write answers for conversation

Chat answers should be direct, but not abrupt.

A useful pattern is:

1. Give the answer.
2. State an important condition or limitation.
3. Link to the full page where useful.
4. Offer the relevant next step.

Avoid large copied paragraphs. Let the visitor ask a follow-up, but do not fragment a safety-critical condition across several messages.

Use the business’s natural tone. Do not pretend the bot is a human colleague.

Test more than the ideal wording

Customers misspell words, combine questions and omit context.

Test:

• Short and vague questions.
• Local terminology and abbreviations.
• Several questions in one message.
• Contradictory details.
• Requests outside the service area.
• Price questions without enough scope.
• Attempts to obtain private information.
• Requests to ignore previous rules.
• Frustrated wording.
• A direct request for a person.

Record whether the bot answered, clarified, refused or handed over correctly.

The theatre is not finished after opening night. Review real conversations and add missing approved answers carefully.

Design a useful handoff

The bot should know how to exit.

For each subject, define:

• The destination team or contact method.
• Information that can be passed with consent.
• The expected response time.
• Out-of-hours behaviour.
• Confirmation shown to the customer.
• What happens if the integration fails.

Allow customers to request a person without fighting through repeated questions. The comparison of AI chatbots and live chat explains where each route works best.

Measure answer quality

Useful measures include:

• Correct answers from approved sources.
• Questions the bot could not answer.
• Inappropriate confident answers.
• Successful human handoffs.
• Repeat contact about unresolved questions.
• Customer feedback in context.
• Time spent maintaining sources.

Measure whether the customer gets an answer or reaches a person who can help, alongside the proportion of conversations handled automatically.

Review a sample manually. Automated satisfaction labels can miss frustration, politeness and silent abandonment.

Frequently asked questions

How many questions should a chatbot know before launch?

There is no required number. Cover a small, well-tested group of frequent questions before expanding. Quality and maintenance matter more than a large count.

Should a chatbot answer questions already covered on the website?

Yes, when chat helps a visitor find the answer quickly. Keep the page itself clear because many visitors will not use chat.

Can it give quotations?

It can repeat a published fixed price or collect information for a quotation. Bespoke pricing should be calculated only when the rules and data are reliable, with human review where needed.

Should it answer competitor comparisons?

It can explain the business’s own service and factual differences supported by evidence. Avoid invented or unfair claims about competitors.

Rehearse the difficult questions first

Do not begin with the easy demonstration prepared by the vendor. Test ambiguity, exceptions, sensitive requests and handoffs.

Build the chatbot from real customer questions and approved sources. Give it a reliable way to pass questions outside its scope to a person.

Next, learn whether a chatbot can generate more website enquiries or review the practical AI website guide.

• AI chatbots
• Customer service
• website content
• Website planning