Convi captures two distinct kinds of shopper feedback — per-message thumbs (was this answer good?) and conversation-level CSAT (was the whole chat good?). This article explains where each comes from, where to find them, and how to use both.
In this article
The two feedback signals
Per-message thumbs
Conversation-level CSAT
Where feedback shows up
The Analyze → Overview view
The Conversations inbox view
Filtering by feedback
Turning feedback on or off
Acting on negative feedback
The two feedback signals
Signal | What it rates | When it's captured |
Message thumbs up / down | A single AI message | Shopper clicks the thumb under any AI message during the chat |
Conversation CSAT | The whole conversation | After the conversation resolves, Convi shows a rating prompt (Auto-end conversation) |
They're independent. A shopper can thumb-down one assistant message and still rate the conversation positively, or thumb-up several messages and still rate the overall chat negatively if the final outcome wasn't satisfying.
Per-message thumbs
During a chat, the shopper can click 👍 or 👎 under any individual assistant message. When they do, the message stores their feedback. In the Conversations inbox, you'll see a badge under that message:
Positive feedback by user — green, thumbs-up icon.
Negative feedback by user — amber, thumbs-down icon.
These badges are also a signal Convi itself uses — repeated thumbs-down on similar questions surfaces as a knowledge gap in Analyze → Recommendations.
Conversation-level CSAT
After a conversation ends, Convi can show a rating prompt — a title, a thumbs-up / thumbs-down choice, and (on thumbs-down) a free-text "Negative reason" field. The copy of this prompt is configurable per language under Settings → Language → {language} → Auto-end conversation:
Title — the heading on the prompt.
Negative reason — the label on the free-text follow-up.
Negative placeholder — placeholder in the textarea.
Negative button — the submit-button label.
After submission — toast/message shown after the shopper submits feedback.
To turn the prompt on or off, open Publish → Widgets → Bubble → Customer satisfaction feedback — "Collect customer satisfaction feedback through a like or dislike rating after resolving the user's inquiry."
Where feedback shows up
Both signals surface in two places:
Surface | What you see |
Analyze → Overview | Aggregate CSAT chart + aggregate thumbs-up / thumbs-down counts |
Conversations inbox | Per-conversation satisfaction on the User data card; per-message thumbs in the transcript |
You don't have to export anything. Both views update in near-real-time as feedback comes in.
The Analyze → Overview view
Two cards near the bottom of the Analyze → Overview page:
Customer satisfaction over time — a line chart of daily average CSAT (1–5) across rated conversations. "Average customer satisfaction rating (1–5) per day across rated conversations."
Message feedback — a side-by-side card showing total Thumbs up and Thumbs down counts and percentages for the window, with a vs previous period badge on each.
Use the chart to see trends (is quality climbing or slipping?) and the message-feedback card to see absolute volumes (how many negative messages did we serve this week?).
The Conversations inbox view
In Conversations, the right-hand User data card shows the conversation's satisfaction:
User's satisfaction — Positive, Negative, or Unknown.
If the shopper left a negative free-text reason, a small info icon next to the value reveals it in a tooltip.
Inside the transcript itself, per-message thumbs render as badges directly under the AI message that was rated.
Filtering by feedback
In the Conversations inbox filter bar you can filter by Customer feedback:
Upvote — only conversations the shopper rated positively.
Downvote — only conversations the shopper rated negatively.
Unknown — conversations with no rating.
This is the fastest way to triage problems. Filter to Downvote every Monday, open each conversation, and look at the assistant's answers. Each one is a specific improvement opportunity.
Turning feedback on or off
Per-message thumbs are part of the widget. To toggle the customer-satisfaction conversation-level prompt, use Publish → Widgets → Bubble → Customer satisfaction feedback.
Auto-end conversation copy (title / negative reason / button / after-submission message) is configured per locale under Settings → Language → {language} → Auto-end conversation.
ℹ️ Note: Don't disable feedback to inflate your CSAT. Negative feedback is the most valuable signal your assistant produces. Read it; act on it.
Acting on negative feedback
A short triage loop you can run weekly:
Conversations → Filters → Customer feedback → Downvote.
Open each conversation and read the assistant's responses.
For each problem, decide:
The assistant didn't know something → Train → Content → Add knowledge (or use Analyze → Recommendations → Add answer if the gap is already on the list).
The assistant knew but answered weirdly → check Train → Rules — write a rule for that situation.
The assistant should have handed over → check Train → Abilities → Handover to human scenarios.
Re-test the question in Evaluate → Playground to confirm the fix.
Most "bad CSAT" weeks resolve into 3–5 specific gaps. Closing those usually fixes the trend.