Your website's conversations help identify what visitors need and where the published content falls short. Review them as observed site interactions, with the selected website and date range kept clear.
Open the correct inbox#
Select the site in the dashboard and open Chat. The Conversations screen lists the title, status, number of messages, and date. Choose View to inspect a conversation; Previous and Next let you move between conversations.
For a hosted site, the route includes its slug, such as /site/example/chat. This site boundary matters when one organization manages several websites. A conversation list for another site is not evidence that an expected submission was lost.
Opening a conversation marks it reviewed. Contact information, when recorded, can be shown in the conversation view. A conversation from a supported contact form can include its submitted fields and contact details. Do not assume every chat visitor supplies contact information.
Turn questions into useful changes#
Review a bounded sample and note repeated topics, unclear wording, missing facts, and questions the available pages cannot answer. Separate the visitor's question from the assistant's response when summarizing demand.
For each proposed improvement, record the existing page or new page needed, the reviewed source fact, and a short success test. For example, several questions about service area may justify one clear service-area answer rather than several nearly identical pages.
Where the conversation view offers the chat-to-CMS draft panel, use it to prepare a draft in the intended collection. Review the generated text and save/preview it before publication. A generated draft should not be treated as a verified business claim.
Export through a connected agent#
The get_chats tool can retrieve conversations in cursor-paginated batches. Set the organization, optional site, and conversation creation-time bounds deliberately. Use include_messages: false to inventory conversations cheaply, or include ordered transcripts when analysis requires them.
Keep the same filters while following next_cursor. Process every required batch before describing an analysis as complete. Creation-time filters select conversations by when they began; they are not a filter for every individual message date.
MCP exports exclude contact information and internal message metadata. This differs from the authorized dashboard view. Do not promise a contact export based on transcript availability.
Interpret the evidence#
These records describe conversations in your own website chat. They do not reveal independent conversations in external AI services or prove a broader market trend. Combine recurring questions with support and customer research, then publish reviewed content and test the revised answers.