Reviewing Conversations

Every conversation your bot has — good, bad, or abandoned — lands in the Activity tab. This is where you find out what people actually ask, spot the conversations that went wrong, and feed what you learn back into training. By the end of this article you'll have a repeatable weekly review routine.

Open any bot and click Activity in the tab bar.

The usage strip

If the bot has had billable activity this period, a summary strip sits at the top: total billable interactions, split into customer chats and assist suggestions. It's a quick sanity check on how this bot contributes to your plan allowance — the full picture lives on the Usage page.

Triage tiles: start here

Above the conversation list, four clickable tiles show live counts and apply a filter preset in one click:

A good habit: click Negative first (these are your clearest failures), then Unresolved (these are your ambiguous ones). Click All conversations to clear the preset.

Filters and search

The filter card below the tiles gives you finer control:

• Search conversations... — free-text search across conversations.
• Satisfaction — All, Positive, Neutral, or Negative.
• Status — All, Resolved, or Unresolved.
• Date range — a calendar picker for any period.

Active filters appear as removable badges underneath, with a Clear all button. Below the list you can change the page size (5, 10, 20, or 50 per page) and page through results with Previous / Next.

Reading a conversation row

Each row is a compact summary of one conversation:

• Who — the visitor, with a country flag when location is known. Hover the name for session details.
• When — relative time ("2 hours ago") plus a message count.
• What — an AI-generated one-line summary of the conversation, so you can scan dozens without opening each one.
• Badges — a Positive or Negative satisfaction badge (neutral conversations show no badge — that's the default state, not a signal), and an Unresolved badge once a conversation has sat unresolved for more than 24 hours.

Rows that deserve attention are highlighted: negative conversations get a red edge, and conversations unresolved for over a day get an amber one. If your list is mostly plain rows, things are going well.

Satisfaction and resolution, defined

• Satisfaction comes from the thumbs up / down control on bot answers (requires Collect Feedback enabled in your widget settings). Positive, neutral, or negative per conversation.
• Resolution records whether the conversation ended with the user's issue resolved without human intervention. It drives the Resolution Rate on the bot Dashboard.

Opening the full thread

Click anywhere on a row to expand it. You'll see the complete message thread — user messages on the right, bot replies on the left, with the bot's formatting and any source links it cited rendered exactly as the visitor saw them.

Reading full threads is the single best way to judge answer quality. Summaries tell you what was asked; the thread tells you how well it was answered.

Session details

HelpJet records session context with each conversation. Hover the visitor's name on any row to see the captured details: IP address, country, city, device type, browser, and operating system.

Behind the scenes, each session also records the domain the widget was loaded on, the visitor's language and timezone, session duration and response times, and — when present — the referrer and UTM parameters (source, medium, campaign) the visitor arrived with.

Open in Test: replay a conversation

Every row has an Open in Test button. It opens that exact conversation in the Train & Test Bot tab, where you can re-ask the same questions against your current training and see whether the bot does better now.

This is the fastest fix-verify loop in HelpJet: find a bad answer in Activity, fix the training, replay the conversation, confirm the new answer. See Train & Test Mode.

Workflow: turning conversations into training improvements

Here's a review routine that takes 15–20 minutes a week:

1. Open Activity and click the Negative tile. Expand each conversation and note why the answer failed — wrong information, missing information, or a question outside the bot's scope.
2. Click Unresolved and skim the summaries. Look for repeated topics: three unresolved conversations about the same thing is a training gap, not bad luck.
3. For each gap, decide the fix:
   • Missing information → add a source that covers it in the Sources tab. See Training Sources Overview.
   • Wrong or outdated information → find the source the bot cited (the source links in the thread tell you which one) and update or refresh it.
   • Out of scope → add guidance to your custom prompt, e.g. how to hand off to email support. See Brand & Voice.
4. Use Open in Test on the worst conversations to verify each fix in Train & Test Mode before calling it done.

The bots that get good reviews aren't the ones that launched with perfect training — they're the ones whose operators read the Activity tab and closed the gaps.

Common questions

Why don't most conversations have a satisfaction badge?

Only positive and negative ratings show a badge. Most visitors never rate at all, and those conversations count as neutral — showing a badge on every one would bury the rows that matter.

Can I delete a conversation?

Not from the Activity tab. Conversation history is retained for analytics and billing accuracy.

Do my own test conversations show up here?

Conversations from the Train & Test Bot tab and dashboard preview appear in your data. Use the search and date filters to separate them from real visitor traffic if needed.

A conversation is marked Unresolved but the user got their answer. Why?

Resolution is inferred from the conversation — if a user got their answer and left without confirming, it can stay unresolved. Treat the unresolved list as "worth a look," not a failure count.