The conversations list

What it is

The Conversations tab is one chatbot's inbox: every session that produced at least one message, from every channel, with the tools to slice it. Empty sessions — someone opened the chat and left without typing — never appear.

It has four parts, top to bottom:

1. Summary cards — review progress plus totals for the currently filtered set.
2. The filter bar — search and nine facets.
3. The table — nine columns, one row per conversation.
4. Pagination — 10, 25, or 50 rows per page.

Conversations are per chatbot. There is no combined inbox across all your chatbots; if you run three, you check three tabs. That is a consequence of each chatbot owning its own knowledge and audience.

When you would use it

• Daily, as a review queue: what came in, what needs a human's eyes.
• After a change, to see whether answers actually improved.
• When a customer refers to "what your chatbot told me".

Where to find it

Open Dashboard → Chatbots → your chatbot → Conversations.

The summary cards

Four cards sit above the filters, and they recalculate against whatever filter is active — so the numbers always agree with the rows underneath.

• Review progress — a completion bar showing reviewed versus awaiting review. Clicking a segment filters the list to it, which is the fastest way into your queue.
• Total conversations in the current filter.
• Total messages in the current filter.
• Average messages per conversation — a rough health signal. Consistently 1–2 usually means people ask once and leave; consistently high can mean the chatbot is not resolving things on the first try.

The columns

Column	What it tells you
Conversation	The opening message — enough to recognise the thread
Channel	Where the person chatted (see below)
Visitor	Who they are, if your site told us. You can rename this column
Messages	How many messages were exchanged
Answered	How many of those got an answer, as a share
Avg. confidence	How well-grounded the answers were, as a bar (hidden on narrow screens)
Last activity	When it was last touched
Review	Reviewed, or needs review — your team's flag
(actions)	View details, and the review toggle

Clicking a row opens the conversation in the Test Chatbot drawer so you can continue it as the owner. To read the full transcript page instead, use View details in the row's actions menu. See Read a conversation.

The Channel column

Channel is where the human was sitting, not what the chatbot knows. Possible values:

• Website widget — the chat bubble on your site.
• Shared link — the hosted /agent/{handle} page, including iframe embeds of it.
• Test chat — you or a teammate, from the dashboard. These are real stored conversations, so your own testing shows up here.
• Unknown — the origin could not be determined.
• Messaging apps — Telegram, WhatsApp, Messenger, Instagram, Slack, Discord, LINE, Viber, WeChat, SMS, where connected and available.

Filtering by channel is the quickest way to answer "is anyone actually using the widget?" — filter to Website widget and look at the count.

Every filter, and what it is for

Search with a scope selector:

• All content — both sides of the conversation.
• User messages — only what visitors wrote. Use this to find how people phrase a problem.
• AI responses — only what the chatbot said. Use this to find every thread where it said something you now regret.

Answer status — All, Answered, Not answered, Partial answer. Not answered is your knowledge-gap list.

Confidence — All, under 50%, 50–70%, over 70%. Low-confidence answers are the ones most likely to be subtly wrong even when they were not flagged.

Feedback — All, Positive, Negative, No feedback. See Conversation feedback.

Review status — All, Reviewed, Needs review.

Channel — as above.

Visitor — a user ID or name your own site passed in. See Name visitors on the widget.

Date range — All time, Today, Last 7 days, Last 30 days, Last 90 days, or a custom from/to.

Sort — Newest first, Oldest first, Most messages.

Every filter is written into the URL, so a filtered view is a link you can bookmark or send to a teammate, and the browser's back button steps through your filter history.

Reviewed — what it actually means

Reviewed is an internal flag for your team, entirely separate from the visitor's thumbs up or down. It means "a human has looked at this". Visitors never see it and it does not change anything about the chatbot's behaviour.

Two ways to set it:

• Per row, with the toggle in the table.
• Mark all as reviewed, the button above the table. It appears only when something is awaiting review, and it clears every unreviewed conversation on this chatbot — not just the page you are looking at, and not just your current filter. Use it when you have genuinely finished the queue, not to tidy up a filtered view.

A workable routine: filter to Needs review plus Negative feedback, work that list, then filter to Needs review plus Not answered and work that one. Everything else can usually be bulk-cleared.

Steps

1. Open the chatbot → Conversations.
2. Read the summary cards to see how much is waiting.
3. Click the awaiting-review segment of the review card, or set the filters you want.
4. Open a row — View details for the full transcript.
5. Fix whatever the thread reveals: a knowledge gap goes to Knowledge, a behaviour gap goes to Settings → Instructions.
6. Toggle Reviewed on the thread.
7. Adjust rows per page (10, 25, 50) if you are working through a long queue.

What you will see

A filter bar, summary cards, and a table. Active filters show as removable chips with a Clear filters control. When nothing matches, the empty state says so and offers to clear the filters, which is different from the "no conversations yet" state you get on a brand-new chatbot.

If your deployment sets a retention window, a notice at the top tells you how many days conversations are kept before they are deleted automatically. Take that seriously — there is no bulk export, so anything you need beyond that window has to be copied out.

Limits and plan notes

You need permission to view conversations; changing review flags needs permission to manage them. Viewers can read but not toggle. See Roles and permissions.

Reading the list costs nothing. Continuing a thread through Open full chat does spend message credits, because each turn is a real answer.

Common problems

The list is empty but I just tested the chatbot.

Give it a moment — a conversation appears once there is a reply to store. A turn that failed before any reply was generated does not create one.

I can see conversations from a chatbot I did not expect.

You have another chatbot open, or you switched account. Conversations are strictly per chatbot and per account.

My own test chats are cluttering the list.

Filter the Channel column to exclude Test chat by selecting a specific other channel, or filter by date range to skip your setup day.

I marked all as reviewed and cleared threads I had not read.

Mark all as reviewed covers every unreviewed conversation on the chatbot, ignoring filters. The flag is reversible per row from the table or the transcript, but there is no bulk undo.

Can I see every chatbot's conversations on one page?

No. Open each chatbot. Customers in the sidebar gives you an account-wide view of people, but the threads themselves stay under their chatbot. See Customers list.