> ## Documentation Index
> Fetch the complete documentation index at: https://help.agentsy.pro/llms.txt
> Use this file to discover all available pages before exploring further.

# Design the conversation

> Plan an AI Experience that understands visitor intent, offers useful guidance, and introduces a human conversation at the right moment.

A successful AI Experience does more than answer questions. It helps visitors explore a topic, understand their options, and decide what to do next—without making them feel like they are filling out a form.

In this guide, you'll design the conversational journey your visitors should experience before you configure the assistant in Agentsy.

## 1. Start with the visitor's intent

Visitors may arrive with different goals. Some want to learn, others want to compare options, and some are ready to speak with you.

| Visitor intent | What the assistant should do |
| - | - |
| Explore | Explain the topic and help the visitor identify useful starting points. |
| Understand | Clarify unfamiliar concepts and answer specific questions. |
| Compare | Explain relevant differences and trade-offs without inventing facts. |
| Apply | Help the visitor think through their situation using your methodology. |
| Take action | Explain the appropriate next step and offer a human handoff when relevant. |

**Design principle:** Respond to what the visitor is trying to accomplish now. Don't force everyone through the same sequence of questions.

## 2. Write a welcoming opening

The first message should make three things clear:

* Who the assistant represents.
* What it can help the visitor explore.
* How to begin with minimal effort.

For example, a leadership consultant might use:

> Hi! I'm the AI guide for this leadership practice. I can help you explore common leadership challenges and approaches. What's on your mind today?

Avoid asking for a name, company, job title, email address, and project details all at once. Collect only information that is necessary for the interaction, and use the platform's configured registration flow where applicable.

## 3. Make the conversation adaptive

A useful conversation generally follows this pattern:

1. **Recognize the question.** Acknowledge the visitor's topic or objective.
2. **Provide immediate value.** Offer a useful explanation or first insight before asking for more details.
3. **Clarify only when necessary.** Ask one focused question if the answer would materially improve the guidance.
4. **Explore the relevant options.** Use your framework, examples, or knowledge resources.
5. **Suggest a next step.** Offer further exploration or a human conversation when it matches the visitor's intent.

This is a flexible pattern, not a script. A visitor asking a straightforward factual question should receive a straightforward answer.

> [!NOTE]
>
> The assistant should not claim to have assessed a visitor's individual circumstances when it lacks the information, expertise, or authorization to do so.

## 4. Use questions with purpose

Questions should help the visitor, not simply gather lead data.

**Helpful:**

> Are you mainly trying to improve communication within your team, or are you working through a specific conflict?

**Less helpful:**

> What is your full name, job title, company size, annual budget, and phone number?

When planning a question, ask yourself: *Will the answer change the guidance I provide?* If not, the question may be unnecessary.

## 5. Let your methodology shape the experience

Your AI Experience becomes distinctive when it reflects how **you** approach a problem.

For example, a consultant's method might be:

1. Understand the challenge.
2. Identify the underlying cause.
3. Explore practical alternatives.
4. Recommend an appropriate next step.

The assistant can use that framework to organize its guidance without revealing confidential methods or presenting a generic checklist as personalized professional advice.

If you have a named framework, define its steps and meaning clearly in your knowledge materials. The assistant should not invent additional steps or claim outcomes your methodology does not support.

## 6. Design the human handoff

The goal is not to replace every human interaction. For many businesses, the most valuable outcome is a better-informed conversation between the visitor and the expert.

A handoff may be appropriate when the visitor:

* Explicitly asks to speak with someone.
* Wants a quote, proposal, appointment, or current availability that the assistant cannot verify.
* Needs a personalized decision or service beyond the assistant's permitted scope.
* Has explored the topic enough to express clear interest in working together.

A natural transition could be:

> We've explored the main considerations. If you'd like to discuss how these apply to your situation, you can book a conversation with the team.

Only refer to a booking button, form, or contact option if it is actually configured for the AI Experience. Don't invent appointment availability or imply a booking has been made.

> [!TIP]
>
> A visitor who wants to keep learning should still be able to ask questions after a handoff is suggested. Avoid repeating the same call to action in every reply.

## 7. Set conversation boundaries

Define what the assistant **can** and **cannot** do before launch.

Examples of useful boundaries include:

* Answer using the approved knowledge and acknowledge uncertainty.
* Never invent prices, inventory, legal requirements, or service availability.
* Avoid collecting unnecessary personal or sensitive information.
* Distinguish educational information from individualized professional advice.
* Refer high-stakes or regulated decisions to a qualified human professional.
* Match the visitor's language when supported and appropriate.

For regulated subjects such as health, insurance, legal, or financial services, establish additional safeguards that reflect the business's actual obligations. A disclaimer alone is not a substitute for those safeguards.

## 8. Draft your conversation blueprint

Use this template to plan the experience before translating it into system instructions.

| Element | Your answer |
| - | - |
| Assistant identity | Who does the assistant represent? |
| Primary visitor goal | What should a visitor be able to understand or accomplish? |
| Opening message | How will you invite the visitor to begin? |
| Main topics | Which questions and situations should it handle? |
| Your approach | What methodology or principles should shape its answers? |
| Clarifying questions | When is more context genuinely necessary? |
| Boundaries | What must it avoid or refer to a human? |
| Handoff triggers | What visitor signals indicate a human conversation is appropriate? |
| Next-step options | What real actions are available to the visitor? |

### Example: An AI Experience for a business consultant

**Visitor:** "I think my business needs automation, but I don't know where to start."

**Assistant:** "A good starting point is to identify work that's repetitive, time-consuming, and guided by clear rules. For example, customer follow-ups or moving information between systems can be candidates. What's one process your team repeats frequently?"

**Visitor:** "We manually prepare weekly client reports."

**Assistant:** "That sounds worth examining. I'd first look at where the data comes from, how much manual formatting is involved, and whether someone needs to review the results. We can walk through those factors together, or you can discuss a more detailed assessment with the consultant."

This conversation provides immediate value, asks only relevant questions, and makes the human next step optional.

## 9. Test the design with realistic conversations

Before launching, test at least these scenarios:

* A visitor asks a simple question and expects a direct answer.
* A visitor provides little context or changes the subject.
* A visitor asks about something outside the approved knowledge.
* A visitor wants a specific recommendation the assistant is not authorized to make.
* A visitor is ready to book, buy, or speak with a human.
* A visitor declines the handoff and continues asking questions.

Check whether the assistant stays helpful, accurate, and aligned with your intended experience. Refine your instructions and knowledge based on the results.

## What comes next?

Once you've designed the conversation, prepare the information that will support its answers.

> **[Prepare your knowledge](/ai-experience-builder/prepare-your-knowledge)**
>
> Organize the expertise, documents, and reference materials your assistant will use.

> **[System instructions](/create-and-configure/system-instructions)**
>
> Learn how to translate your conversation blueprint into assistant behavior.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.