A live chat satisfaction rating is the quick score a visitor gives a conversation after it ends. It's useful when you treat each rating as a pointer to a transcript, not as a grade. For a small team, the routine is simple: ask once, right after the chat closes; read every low rating alongside its conversation; look for patterns by topic and time of day; and change one thing at a time.
Below: when and how to ask, how to read a small number of ratings honestly, what to do after a low score, and a 30-minute weekly review.
What a chat satisfaction rating does (and doesn't) tell you
A rating captures how the visitor felt at the end of the chat. That makes it good at spotting friction: long waits, a reply that missed the point, a chatbot that went in circles, an answer that sounded curt.
It doesn't tell you everything:
- It isn't proof the problem was solved. A visitor can rate a friendly chat highly and still not get what they needed. Another can rate a correct answer poorly because it was bad news ("we don't ship to your country").
- It doesn't tell you why. The score points you to the conversation. The transcript tells you what happened.
- It reflects the people who answered. Many visitors close the window without rating, and the ones who do are often the very pleased and the very annoyed. Treat the average as a rough signal.
The term you'll often see for this is CSAT, short for customer satisfaction score. It's usually reported as the share of ratings that were positive (satisfied ratings divided by all ratings, times 100) or as an average score. Use whichever your tool shows, and keep using the same one so months stay comparable.
When to ask for a rating
Ask once, at the end of the conversation, while the experience is fresh. A rating request that pops up mid-chat interrupts, and one sent days later by email tends to get fewer, vaguer answers.
A few rules keep the request welcome:
- Close the chat when it's actually done. Confirm the outcome first ("Glad that's sorted! Anything else I can help with?"), wait for the reply, then close. Closing on a visitor who still has a question is an easy way to earn a low score.
- Don't ask twice. One request per conversation is plenty.
- Make it optional. A rating should never stand between the visitor and leaving.
- Keep the ask tiny. One tap or one click. If your tool offers a comment field, keep it optional.
What to say before you close
You don't need a special script, but a short closing line makes the rating feel like part of the conversation rather than a survey. Pick one that fits your tone:
Thanks for chatting, [name]. If you have a second, a quick rating helps us improve. Have a great [day / evening].
To recap: [what you'll do] by [when]. Thanks for your patience, and feel free to rate this chat so we know how we did.
Avoid "Please rate me 5 stars!" and similar nudges. Asking for a high score flatters the number and hides the chats you most need to see. For more closing lines, see the closing section of our live chat scripts.
How to read ratings when you only get a few
Small businesses rarely have enough ratings for the average to mean much on its own. If you get a dozen ratings a month, one unhappy visitor can move the average noticeably. That's normal, and it's why the number should start a review, not end one.
Three habits keep you honest:
- Always look at the count. "Average down this month" means little if it's based on six ratings. Write the number of ratings next to the score every time.
- Read every low rating. With small volumes this is quick, and it's where nearly all the useful lessons are.
- Compare with yourself, over longer periods. Look at a rolling quarter as well as the month. Industry averages mix different businesses, channels and definitions, so they're a poor yardstick for a small team. Our guide to live chat metrics explains why we compare with our own trend instead.
Look at ratings in groups
Once you have a few months of ratings, sort them into groups and look for the group that stands out:
| Group by | What it can reveal |
|---|---|
| Topic (pricing, delivery, technical help) | One subject your answers or website explain badly |
| Time of day or weekday | Times when replies are slow because nobody's covering |
| Bot only vs handed to a person | Whether the chatbot or the handoff is the weak point |
| Teammate | Who might need help with a type of chat (coach privately, never as a league table) |
If you tag conversations by topic, this sorting takes minutes in a spreadsheet. If you don't tag yet, start with your three or four most common topics.
What low ratings usually point to
Low ratings tend to cluster around a few causes. Reading the transcript usually makes it obvious which one you're looking at.
| What you see in the chat | Likely cause | What to change |
|---|---|---|
| Visitor waited, then left or wrote "hello??" | Slow first reply, or nobody online | Adjust coverage, set honest office hours, improve the offline message |
| Visitor repeated what they'd told the bot | Agent didn't read the context | Read the collected answers before replying; see our chatbot to human handoff checklist |
| Visitor asked for a person early and couldn't find one | Bot stood in the way | Keep a "Talk to a person" option visible at every step |
| Short, copy-pasted replies to a detailed question | Script used without adapting it | Personalize the first line; answer the actual question |
| Correct answer, unwelcome news | The policy, not the chat | Explain the reason and offer an alternative; record it as policy feedback |
| Chat closed before the visitor confirmed | Rushed close | Confirm the outcome and wait for a reply before closing |
| Wrong or invented answer from an AI assistant | Missing or unclear content | Fix the source content and make sure the AI hands off when unsure |
If the first row keeps coming up, especially for chats that arrive outside your hours, our guide to after-hours live chat has offline message templates and a coverage checklist. The last row matters if you use AI answers: each wrong answer is a content gap, and our guide to stopping AI chatbot hallucinations walks through the fixes.
What to do after a low rating
A single low rating is a case to read, not a verdict. Here's a simple sequence that works for a team of one or ten.
- Read the whole conversation, including what the chatbot collected and any private notes.
- Decide whether the visitor's problem is still open. If it is, that comes first.
- Follow up if you have a way to reach them and there's something useful to say. Keep it short and specific.
- Note the cause in a line: wait time, missing context, policy, wrong answer, rushed close.
- Fix the cause, not the score. Update the saved reply, the bot step, the website page or the office hours.
- Talk about it as a team if it's a pattern, not to single someone out.
Follow-up templates
Only follow up when the visitor left contact details and you can actually help. Never ask them to change their rating.
The problem is still open
Hi [name], this is [your name] from [business]. I read back through our chat from [day] and I don't think we fully answered your question about [topic]. Here's what I found: [answer]. If that doesn't solve it, reply here and I'll take it from there.
They waited too long
Hi [name], sorry you had to wait when you messaged us on [day]. That's on us, not you. Your question was [topic], and the answer is [answer]. If anything else is unclear, just reply.
The answer was bad news
Hi [name], thanks for your honest feedback on our chat. I know [policy] wasn't what you hoped to hear. The reason is [short reason]. What we can do instead is [alternative]. If that would help, reply and I'll set it up.
If you keep writing the same follow-up, the real fix is upstream: a clearer page, a better bot answer, or a change in coverage.
A 30-minute weekly ratings review
Put it in the calendar, ideally the same slot each week.
- Note this week's number of ratings and the score, next to last week's.
- Open every low rating and read the conversation.
- Write one line per low rating: topic and likely cause.
- Skim a few high ratings too. They show what to keep doing and make good examples for new teammates.
- Read a couple of unrated chats, especially ones where the visitor left without a reply. Ratings aren't the only signal.
- Check whether any open problems still need a follow-up.
- Pick one change for the coming week and write it down.
- Next week, check whether that topic's ratings improved before changing something else.
Once a month, roll the weekly notes up into your monthly scorecard and look at the groups in the table above.
How to do this in RetroChat
On the Business and Autopilot plans, visitors can rate a conversation after your team closes it, so closing chats when they're truly done is part of the routine. The Reports page shows Satisfaction from visitor ratings for the last 7, 30 or 90 days or any custom range, compared with the same length of time just before, and Team performance shows satisfaction for each teammate alongside chats replied, closed and first response time. Turn on the weekly report email to get the key numbers every Monday, which fits the weekly review above.
For the deeper look, the chat CSV export includes the rating for each conversation, along with tags, the assignee and the page, so you can group ratings by topic or teammate in a spreadsheet. Everything the chatbot collected is shown next to each chat, and private notes let you record the likely cause for the team. See Inbox and team for how reports and exports work, and Plans and billing for what each plan includes.
If you're still setting chat up, start with our guide to adding live chat to your website, and our guide to a shared inbox for live chat covers the daily routine that keeps ratings healthy in the first place.
Start listening to your chats
Ratings are worth collecting only if someone reads the chats behind them. Ask once at the end, read every low one, fix one cause a week, and the score usually follows. You can try the ratings and reports in a free 14-day RetroChat trial, or book a demo to see them with your own questions.
Frequently asked questions
What is a good live chat satisfaction score?
There's no reliable universal number. Published averages mix industries, team sizes, rating scales and definitions, and many come from vendors without a published method. Compare your score with your own previous months, and pay more attention to the low ratings and their causes than to the average.
How do you calculate chat satisfaction?
The most common method is the share of positive ratings: the number of satisfied ratings divided by the total number of ratings, times 100. Some tools show an average score instead. Either works, as long as you use the same method every month and note how many ratings it's based on.
When should you ask customers to rate a chat?
At the end of the conversation, once the outcome is confirmed and the chat is closed. Ask once, keep it to a single tap, and make it optional. Asking mid-chat interrupts the visitor, and asking days later tends to get fewer and vaguer answers.
How do I respond to a bad live chat rating?
Read the whole conversation first, then check whether the visitor's problem is still open. If it is and you can reach them, follow up with a short, specific message that solves it, without asking them to change their rating. Then note the cause and fix it at the source.
Why are my chat ratings low even when the answers are correct?
Usually because of how the chat felt rather than what it said: a long wait, a rushed close, having to repeat information, or an answer that was bad news without an explanation or an alternative. Read a few of those transcripts side by side and look for the common moment where things went wrong.