Simple Agent
Blog
website agent scopeAI agent question boundarieswebsite chatbot questionsvisitor question handlingwebsite agent demo

How to Define What Your Website Agent Should Answer

October 16, 20267 minSimple Agent Team

How to Define What Your Website Agent Should Answer

A website agent is most useful when visitors can tell what it can help with—and when it knows what to do if a question falls outside that scope. Before embedding one, decide which questions it should answer, when it should ask for more detail, and when it should hand the conversation to a person.

This review gives you a practical way to set those boundaries using questions visitors already ask. The result is a clear starting point for your website agent scope, grounded in your business’s own information.

1. Collect real visitor questions

Start with questions people have actually asked. Look through support messages, sales conversations, and website forms. Include questions that come up repeatedly, but keep unusual questions too if they reveal a useful edge case.

Remove names, email addresses, phone numbers, order references, and other personal details before using messages as examples. Keep the wording close to what the visitor asked, while replacing identifying information with neutral descriptions.

For example, change “Can I move my appointment on May 14 for order 48291?” to “Can I reschedule an appointment?” Preserve the intent, not the person’s details.

Group related questions without assuming they all need the same response. “What does the service include?” and “Will this work for my specific situation?” may sound related, but the second may require more context or human judgment.

A useful first list might include:

  • Questions about services, products, and published policies
  • Questions about pricing or how to get started
  • Questions that depend on a visitor’s circumstances
  • Requests about an existing account, order, booking, or case
  • Complaints, exceptions, or urgent requests

This list is your source material for deciding which website chatbot questions the agent should handle. Avoid writing hypothetical questions just to make the list look comprehensive. You can add scenarios later when they help expose an unclear boundary.

2. Choose answer, clarify, or handoff

For each question, choose one of three paths: answer, clarify, or handoff. Then write down what information the agent needs to take that path.

Answer

Choose answer when the response is available in material you have approved for the agent, such as a service description, published policy, or help article.

Record the source the answer should rely on and what a complete response needs to cover. If the question is “What is included in the consultation?”, your review might specify the published list of inclusions and any relevant conditions. If that information is missing or unclear, resolve the source content before expecting the agent to answer reliably.

Clarify

Choose clarify when the agent could help after the visitor provides a specific piece of information. State what detail is needed and why.

For “Which plan is right for me?”, the missing information might be the visitor’s intended use. A good clarifying question asks for that detail directly instead of guessing based on a broad prompt.

Decide how many details are necessary to continue. Asking for information that will not change the next step creates friction without improving visitor question handling.

Handoff

Choose handoff when a person needs to make a decision, review a case, or take an action the agent cannot perform. State what the agent should tell the visitor and what the person receiving the handoff needs to know.

A handoff does not need to be vague. You can define whether the agent should explain how to contact your team, invite the visitor to use a form, or summarize the kind of issue that needs attention—provided that process is actually available to your business.

A simple review table can make these choices concrete:

Visitor question Path Information needed
“What does this service include?” Answer Approved service description
“Which option fits my situation?” Clarify The visitor’s intended use
“Can you change my existing booking?” Handoff The booking issue and the available contact route

This table turns broad AI agent question boundaries into decisions you can review before embedding.

3. Mark questions that need caution

Some questions require information an agent may not have, or judgment that should stay with a person. Mark these clearly rather than hoping the agent will infer the right boundary.

Pay special attention to questions that depend on:

  • Current information: availability, schedules, changing prices, or recent policy changes
  • Account-specific details: an order, booking, membership, payment, or personal record
  • Sensitive information: health, finances, identity, or private circumstances
  • Human judgment: exceptions, disputes, eligibility decisions, or advice that depends on context

For each one, decide whether the agent should answer from an approved current source, ask a clarifying question, or hand the visitor to your team. If you have not provided an appropriate source, the agent should not fill the gap with a plausible-sounding answer.

Be specific about what “current” means for your business. A published service page may be suitable for explaining what a service includes, while today’s availability may need a person to check. Similarly, a general policy answer does not establish what happened in a visitor’s account.

These boundaries help keep the agent useful without implying it can see private records, make decisions, or perform actions you have not configured it to handle.

4. Try representative examples in your configured agent

Include a straightforward question, one that needs clarification, and one that should trigger a handoff. You can also try a question where the answer depends on information your sources do not provide. See the Simple Agent plans.

To review answers against your own material, use the agent configured with those sources.

For each example, check:

  • Did the response address the question the visitor actually asked?
  • If a detail was missing, did the agent ask a useful clarifying question?
  • If the question needed a person, did the response avoid guessing?
  • When retrieved source material was available, did the answer show source citations at the relevant chunk level?
  • Did the answer stay within the approved information?

A citation is useful when it helps you check what supports an answer. Its presence does not make an unsupported claim reliable, and not every response will necessarily have retrieved source material to cite. Review the response itself alongside any citation.

Write down what should change: the boundary, the source content, or the example response. Then try the revised scenario again. For more setup and product details, see the Simple Agent documentation.

5. Embed after the review, then revisit it

When the question list and paths make sense, embed the widget using its script tag. Review your site’s own publishing process to decide where the script belongs. For platform-specific steps, follow verified instructions for that platform; general guidance about adding a script does not establish that a particular integration or plugin exists.

After embedding, treat your categories as a working scope. Revisit them when your services, policies, or source content change. If you add a new offering, check which visitor questions it may create and whether the answers are present in the approved sources. If a policy changes, review examples that rely on the old wording.

You do not need to revisit every category after every small edit. Focus on changes that affect what visitors can ask, what the correct answer is, or when a person needs to take over. Keep the review table with the source material it depends on so the next update is easy to trace.

Defining what the agent should answer is a practical content review: collect real questions, choose a path for each, and check representative examples against approved information. Pick a Simple Agent plan, save a few questions from your list as agent tests, then use the results to finalize your scope before adding the script to your site.

Ready to build your AI agent?

Guided setup in 90 seconds. One flat price, published monthly volume.

Create my agent