Choosing an approach
FAQ vs Chatbot vs Live Chat: What Does Your Site Need?
Choose an FAQ, live chat or AI website assistant using customer questions, available staff and source requirements, with a practical decision worksheet.
By HostDesk · Updated · 6 min read
An FAQ can be enough when customers need a few stable answers that are easy to find. Live chat fits questions that need a person and a team able to respond. An AI assistant can help with varied questions about documented information, provided it has reliable sources, clear limits and a route for unresolved requests.
Choose the smallest approach that serves the customer well. Start with the question, the available evidence and the person or system responsible for the next step.
Use the customer question to choose the approach
A useful comparison goes beyond whether a tool can display a chat bubble. Ask how the answer is established, who maintains it and what the visitor can do when the question falls outside its scope.
These are decision criteria rather than a ranking of named vendors. Specific products combine these approaches in different ways, so verify the workflow you intend to use.
| Customer need | A suitable starting approach | What you must provide | Limit to check |
|---|---|---|---|
| A few repeated policy or process questions | Clear FAQ or relevant page sections. | Approved answers and descriptive links. | Can customers find the answer where they need it? |
| Individual judgment, complaints or exceptions | A person through live chat, phone or a request form. | A real response process and responsible team. | Do you have staff available at the advertised time? |
| Varied wording across documented services and policies | Source-grounded AI answers with a handoff. | Current knowledge, testing and maintenance. | Does it stay within what the sources establish? |
| Current stock, order or appointment state | An authorized current-data workflow. | Appropriate access and identity controls. | Does it retrieve the relevant live fact? |
| A booking, payment or other completed task | The actual authorized execution path. | Clear authorization and completion evidence. | Does the receipt prove completion or only a click? |
When a clear FAQ is enough
If the same small set of questions dominates, improve the answer and its placement first. Explain the conditions near the price or offer, then link to the complete policy where appropriate.
An FAQ has to remain current too. Assign someone to review it when the underlying service or policy changes. Several versions of the same answer can create confusion regardless of the interface.
For a store, make shipping and returns reachable through the routes customers use. Baymard's own usability research describes people looking for this information on product pages and through the footer. That supports checking findability; it does not prove your site needs a chatbot.
- Use the customer's words in the question.
- Put the useful answer first and keep important exceptions visible.
- Link to the relevant next step rather than a generic homepage.
- Use real enquiry history to decide which questions deserve space.
When access to a person matters
Some questions require negotiation, diagnosis, reassurance or a decision someone is authorized to make. A human response route is useful only when the business can operate it as described.
Set expectations around actual availability. If no one can respond overnight, use a request flow with an honest acknowledgement rather than imply that a person is waiting.
Decide who receives a request, what context they need and how you will notice a failed delivery. Keep collection proportionate to the task. A general question may need no personal details; a callback request needs a way to contact the visitor.
| Operating question | Your answer |
|---|---|
| Coverage | Who responds, during which hours and time zone? |
| Authority | Which decisions can that person make? |
| Handoff | Where does an unanswered conversation go? |
| Expectation | What can you truthfully tell the visitor about follow-up? |
| Failure | How will you find and recover an unreceived request? |
When an AI website assistant is a reasonable fit
An AI assistant is worth evaluating when people ask varied questions across information the business can document. The trial should show whether it understands useful follow-ups, cites relevant evidence and handles an unsupported detail without inventing one.
Do not use conversational fluency as the only selection criterion. Test a change of service, an incomplete question and a commercial exception. Inspect what happens when the answer requires a current source or a person.
An assistant also creates maintenance work. Someone still owns the policies, approved answers and next-step destinations. Ask how changes reach the assistant and how you can inspect a wrong answer.
Four examples of the minimum useful setup
The examples below are fictional choices, not measured customer results. Use them to identify the work your site actually needs.
A site can also combine approaches: clear information for everyone, conversational help where it adds value and a person for exceptions.
| Situation | Start here | What would justify more |
|---|---|---|
| A consultant receives the same three process questions. | Clarify the service page and FAQ; keep a working enquiry form. | Repeated varied questions that customers still struggle to answer. |
| A repair team must assess each job before quoting. | Explain the process and collect a useful request for a person. | Evidence that documented questions can be answered without a bespoke assessment. |
| A business has many supported offerings and policies. | Test source-grounded answers against representative questions. | Successful evidence checks, useful follow-ups and a working handoff. |
| A store receives questions about individual orders. | Use its authorized order/help route. | A verified integration with the required identity boundary and current data. |
Run a small trial with your own evidence
Choose a limited set of real question types and write the expected outcomes. Record how each approach handles them, including questions that need a person. Use the same cases so the comparison is fair.
If you estimate staff time, use your own observed counts and handling times. For example, the arithmetic is monthly eligible questions multiplied by average minutes spent on those questions. It is an estimate of workload, not guaranteed time saved by a new tool.
Keep adoption costs in the decision: setup, review, staff coverage, ongoing knowledge maintenance and plan limits. Avoid comparing a flat subscription with a competitor price quoted under a different billing period or usage allowance.
HostDesk offers a bounded public-page preview you can use during this evaluation. Check what the sample supports and what remains unresolved before deciding whether it fits your business.
- List the questions and evidence.
- Choose the minimum approach likely to serve them.
- Test ordinary, ambiguous and unsupported cases.
- Inspect the actual next step and failure behavior.
- Compare operating work and observed outcomes before committing.