Conversations
Conversations is the record of every interaction your assistants have had — web chats, phone calls and SMS — merged into one list. Use it to read transcripts, see how visitors rated the answers, spot turns that failed, and check whether a conversation met the goal you set.
The page header
Three controls sit beside the title:
| Control | What it does |
|---|---|
| The active pill | How many chat conversations are live right now. The dot is lit while at least one is. |
| The assistant menu | Scopes the whole page to one assistant. All Assistants is the default. |
| The refresh button | Reloads the current page of results without losing your filters. |
Conversations that go quiet are closed for you: when the list loads, any chat still marked active after 30 minutes without a new message is marked ended.
Channels
The chips across the top of the filter card pick a channel: All, Chats and Voice, plus Calls and SMS when phone and SMS are switched on for your workspace — or when you already have call or SMS conversations on record. Each channel chip carries a count when it has any conversations, so you can see the split without switching. The counts respect the other filters you have set.
Search
The Search contacts, messages, phone numbers… box searches as you type, with a short pause before the results refresh. What it matches depends on the channel:
| Channel | Matched on |
|---|---|
| Chat | Session ID, external user ID, and the text of any message in the conversation |
| Call | Caller number and the call summary |
| SMS | The to and from numbers and the message body |
Because chat message text is searched, you can find a conversation by something the visitor or the assistant actually said.
Filters
Filters opens a panel under the search box. When any filter is set, a badge beside the button counts them — search, date range, feedback and channel each count as one.
Date range
Pick one of All Time, Today, Last 7 Days, Last 30 Days, or Custom.
The presets are anchored to your own clock: Today runs from midnight to the end of today, Last 7 Days and Last 30 Days go back that many days from the start of today up to the end of today.
Custom reveals From and To date-and-time fields. Editing either one applies immediately and switches the range to custom. You can set just one end — a From with no To means "everything since then".
Feedback
Filter by how visitors rated the assistant's answers: All, Liked, Disliked, Has feedback, or None.
Note
Feedback ratings exist only on web chat, so choosing anything other than All here restricts the list to chat conversations. Calls and SMS have no per-message rating to filter on.
The filter footer
The bottom of the panel shows how many of the total conversations you are looking at, and offers Clear all to reset search, dates, feedback and channel in one click.
The summary row
Four stats sit between the filters and the table. They recompute against whatever filters are active, and they are hidden while results are loading or when there are none.
| Stat | What it means |
|---|---|
| Total | Conversations matching the current filters, with the chat / call / SMS split underneath |
| Active | Chat conversations that are live right now |
| Avg duration | Average length of ended chat conversations |
| Satisfaction | Share of message ratings that were thumbs-up, with the number of votes below |
Note
Active, Avg duration and Satisfaction are computed from web chat only — calls and SMS contribute to Total and to the channel counts, but not to those three. For call and SMS volume over time, see Analytics.
Average duration is capped at one minute per message, so a chat that was left open overnight without being closed cannot skew the figure.
The table
Each row is one conversation. Click anywhere on the row — or focus it and press Enter — to open it; the View button at the end of the row does the same thing. On a narrow screen the table becomes a list of cards with the same information.
| Column | Shows |
|---|---|
| Contact | A channel icon, the contact's name, and a preview of the last thing said. Visitors with no name read as Anonymous Visitor; where the conversation is linked to a contact, the contact's name is used. |
| Assistant | Which assistant handled it |
| Msgs | Number of messages (turns, for calls) |
| Status | Active or In progress with a live dot, Ended, Sent / Delivered for SMS, Voicemail, or a red badge for Failed, Error, No answer and Busy |
| Feedback | Thumbs-up and thumbs-down counts for that conversation, or a dash when nobody rated it |
| Duration | How long it ran — 45s, 3m, 1h 5m, 2d |
| Started | When it began, in your local time |
Sorting
Msgs, Status, Duration and Started are sortable — click the header to sort, click it again to reverse. Started, newest first, is the default. Sorting is applied across all matching conversations, not just the page you are on.
Pagination
Twenty conversations per page. When there is more than one page, a pager appears below the table showing which rows you are seeing out of the total. Changing a filter always returns you to page one.
Reading a conversation
Opening a conversation shows the full transcript in a panel over the page.
The header
The header names the contact and summarises the conversation: the message count (turns, for a call), the duration, when it started, and — depending on channel — the call disposition or whether the SMS was inbound or outbound. A Call or SMS chip marks non-chat conversations.
The resolution badge
If the assistant has a Goal set on its General settings tab, every conversation is graded against that goal when it ends, and the result appears as a badge in the header:
| Badge | Meaning |
|---|---|
| Goal achieved | The conversation met the goal you described |
| Goal not met | It ended without meeting the goal |
| No goal | No goal is configured, so nothing was graded |
| Grading error | The grader could not evaluate this conversation |
Graded conversations also carry a score out of 5 next to the label — Goal achieved · 4/5. Hover
the badge to read the grader's one- or two-line explanation of why it decided that.
Grading is what powers the resolution figures in Analytics. Until you write a goal, every conversation grades as No goal and those figures stay empty.
The transcript
Messages run in order, visitor on one side and assistant on the other, each with the time it was sent. Date separators — Today, Yesterday, or the date — break up conversations that span more than one day. Call transcripts label each turn Contact or AI Assistant.
Two markers are worth looking for:
- Liked / Disliked — a visitor rated that answer with the thumbs in the widget. When the rating was a thumbs-down and the visitor left a comment, the comment is shown under the message.
- Failed turn — the assistant did not answer that turn. The tag carries a short error code where one is available, and the conversation's status becomes Error. These are turns the visitor experienced as a failure, so they are the first thing to look at when someone reports the assistant "stopped working".
Exporting
The download button in the panel header offers Export as TXT and Export as JSON. Both save the conversation you are reading to a file named after its ID and start date.
- TXT is a readable transcript: visitor and session details, then each message prefixed with its time and speaker.
- JSON is the structured record — visitor name, session ID, start time, duration, message count, status, and the messages with roles and timestamps.
The footer
The foot of the panel shows the identifier for the channel you are looking at: the start of the session ID for a chat, the direction, number and duration for a call, or the number and delivery status for an SMS. Close dismisses the panel.
Using this page to improve an assistant
The fastest loop is to read the conversations that went wrong and change one thing at a time.
- Start with the thumbs-down. Open Filters, set feedback to Disliked, and read those transcripts. The comment attached to a thumbs-down usually says exactly what the visitor wanted instead.
- Then look at failed turns. Filter the channel to Chats and scan for the red Error status. A run of failed turns is a configuration or service problem, not a wording problem.
- Then read the ones marked Goal not met. The grader's reasoning tells you where the conversation stalled — a question the assistant could not answer, a hand-off that never happened, a detail it never asked for.
- Fix one cause. Missing knowledge goes in the assistant's documents; wrong tone or a missing question goes in its instructions; a goal it can never reach needs a better goal.
- Come back and compare. Set the date range to Last 7 Days and check Satisfaction and the balance of Goal achieved against Goal not met before and after the change.
Search is the other half of the loop: when support tells you "people keep asking about refunds", search the word and read what the assistant actually says today.
For volume, cost and outcome trends across all of this, go to Analytics. For the account-wide record of what your assistants did — including events that are not conversations — see Activity log.