> ## 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. # Conversation Insights > Learn how to interpret visitor questions and conversations to improve your AI Experience and identify opportunities for meaningful follow-up. # Conversation Insights Every conversation with your AI Experience can help you understand what visitors want to know, what they find confusing, and when they may benefit from speaking with someone on your team. Conversation insights turn these interactions into practical learning opportunities. Rather than looking only at how many people started a chat, focus on **what they asked, what they needed, and what happened next**. > [!NOTE] > > The information available to you depends on your Agentsy account, the experience's access settings, and the conversation and analytics features enabled for your workspace. This guide describes how to interpret available information; it does not imply that every metric or report is available in every account. ## What can you learn from conversations? Reviewing available conversation records and engagement information can help you recognize: * **Recurring questions:** Topics visitors ask about repeatedly. * **Unclear explanations:** Questions that suggest your website or assistant needs more clarity. * **Knowledge gaps:** Topics your assistant cannot answer confidently with its current materials. * **Visitor needs:** Problems, goals, and concerns visitors describe in their own words. * **Signals for follow-up:** Explicit requests for a consultation, a quote, a demonstration, or other human assistance. * **Opportunities to improve:** Moments when an answer was incomplete, overly generic, or unhelpful. These are signals to investigate, not automatic conclusions about individual visitors. ## 1. Start with the questions visitors ask Look for themes across conversations instead of drawing conclusions from a single message. For example, visitors to a business consultant's AI Experience might repeatedly ask: * "How do I know if my business is ready to expand?" * "What should I prioritize before hiring?" * "How would your methodology apply to a small team?" Together, these questions may suggest that visitors need clearer guidance on business readiness and prioritization. **Possible improvement:** Add a focused knowledge document explaining the consultant's readiness framework, then test whether the assistant can apply it to different scenarios. ## 2. Distinguish curiosity from expressed intent Not every conversation indicates a sales opportunity. | Visitor message | What it may indicate | Appropriate response | | - | - | - | | "What does your methodology involve?" | General exploration | Explain the methodology and invite further questions. | | "Could this approach work for my organization?" | Interest in relevance | Explore the visitor's situation without assuming readiness to buy. | | "Can you help us implement this next month?" | Potential service interest | Explain the next step and offer human contact when appropriate. | | "I'd like to book a consultation." | Explicit request for human follow-up | Direct the visitor to the available booking or contact option. | > [!WARNING] > > Do not treat a visitor's identity, a single question, or the length of a conversation as proof that they are a qualified lead. Confirm intent through what the visitor actually expresses and through an appropriate human conversation. ## 3. Review the information available in Agentsy To review conversation insights: 1. Sign in to your Agentsy workspace. 2. Open the relevant AI Experience. 3. Navigate to the available conversations, visitors, or analytics area in your workspace. 4. Review the conversation and engagement information your account makes available. 5. Note recurring topics, unanswered questions, and explicit requests for next steps. The exact menu names and available information may vary with your workspace configuration. If a report or metric described elsewhere is not visible, do not assume it is enabled for your account. > [!TIP] > > Start by reviewing a small sample of recent conversations and writing down three recurring themes. A simple review is often more useful than collecting metrics without a clear question in mind. ## 4. Turn observations into improvements Use a simple review cycle: **1. Identify a pattern** Find a recurring question, confusion point, or request that appears in available conversations. **2. Check the underlying cause** Determine whether the issue comes from missing knowledge, unclear system instructions, website messaging, or a genuine limitation of the service. **3. Make one targeted improvement** Update the relevant knowledge document, refine the assistant's instructions, clarify landing page content, or improve the human handoff. **4. Test the change** Ask the assistant realistic questions that reproduce the original problem. Check for accurate, useful, and appropriately bounded answers. **5. Review subsequent conversations** Look for signs that the experience is becoming clearer or more useful. Avoid claiming improvement based on too little evidence. ## 5. Use insights to improve human follow-up When a visitor explicitly asks to speak with someone, the conversation may provide useful context for that next step, subject to the information and permissions available to your organization. For example, a visitor might explain that they are comparing two service approaches and want help choosing between them. A human advisor can use that expressed need to prepare a more relevant discussion. Good follow-up should: * Respect the visitor's request and communication preferences. * Focus on the problem or goal they voluntarily shared. * Avoid assumptions about budget, authority, urgency, or readiness. * Follow your organization's privacy policies and applicable law. The AI Experience should support better human conversations, not replace the judgment needed to understand a prospective client. ## 6. Protect visitor privacy Conversation content can include personal or commercially sensitive information. Review and use it responsibly. * Access conversation information only when you are authorized to do so. * Avoid copying sensitive content into unrelated documents or tools. * Do not encourage visitors to share passwords, payment card details, medical records, or other unnecessary sensitive information. * Use aggregated or de-identified themes when individual identity is not needed. * Follow applicable privacy notices, retention policies, and legal requirements. > [!WARNING] > > A conversation being available for review does not automatically authorize every possible use of its contents. Follow your organization's privacy commitments and applicable data-protection requirements. ## Example: Insights for an insurance education experience Suppose visitors repeatedly ask about the difference between a premium, deductible, and out-of-pocket maximum. **Observation:** Visitors need clearer explanations of foundational terminology. **Action:** Improve the educational knowledge documents with concise definitions and worked examples. **Test:** Ask the assistant to explain these concepts in plain language, including what it cannot determine without policy-specific information. **Human follow-up:** If a visitor wants to discuss their personal coverage options, guide them to the appropriate licensed professional rather than requesting sensitive details in chat. This approach turns visitor questions into better educational content without assuming that every visitor is ready to purchase insurance. ## Conversation review worksheet Use this lightweight template during a periodic review: | Question | Notes | | - | - | | Which topics appeared repeatedly? | | | What did visitors seem to find confusing? | | | Which questions could the assistant not answer reliably? | | | Did anyone explicitly request human assistance? | | | Which knowledge or instruction changes would help? | | | What will we test after making changes? | | ## Common mistakes to avoid * **Counting every registration as a qualified lead.** Registration and commercial intent are different. * **Assuming long conversations mean success.** A long exchange may indicate engagement, confusion, or both. * **Overreacting to one unusual question.** Look for recurring patterns and meaningful context. * **Treating AI-generated interpretations as facts.** Verify conclusions against the visitor's actual words. * **Ignoring knowledge gaps.** Use unanswered questions to improve the experience. * **Using conversation data without considering privacy.** Collect and review only what is appropriate. ## Related guides * [Conversation Experience](/conversations-and-leads/conversation-experience) * [User Registration](/conversations-and-leads/user-registration) * [Lead Identification](/conversations-and-leads/lead-identification) * [Understanding User Intent](/conversations-and-leads/understanding-user-intent) * [Human Handoff](/conversations-and-leads/human-handoff) * [Update Your Knowledge](/knowledge-management/update-your-knowledge) * [Troubleshoot AI Responses](/knowledge-management/troubleshoot-responses) This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.