Why most satisfaction surveys get ignored
A customer who just resolved their problem doesn't feel like filling out a fifteen-question form with Likert scales. The reason so many CSAT surveys get such low response rates isn't that people don't care about giving you feedback — it's that the cost of giving it is higher than the perceived benefit of doing so. The first rule of good CSAT measurement is that it has to cost the customer almost nothing to respond.
The right question is usually just one
The standard that works best in support is a single short-scale question — "How satisfied were you with this support?" on a 1-to-5 scale, or a straight thumbs-up/thumbs-down — right after the conversation closes. You don't need more than that to get a useful signal. If you want to understand the "why" behind a bad rating, offer an optional text field, but genuinely optional.
Timing matters as much as the question
Asking too soon gives you a rating of the conversation, not of the outcome. Asking too late drops the response rate to nearly zero. The right moment is immediately after the case gets marked resolved, on the same channel where the conversation happened — a short WhatsApp message right after closing the chat has a much better chance of a response than a separate email the next day.
Don't survey everything, all the time
Sending a survey after every single conversation, without exception, wears the customer down and ends up lowering the quality of the responses you do get. It makes more sense to sample — survey a portion of conversations, or prioritize surveying after cases that had some complexity or length, where the feedback is more informative.
What to do with a bad rating
A low rating with no follow-up is a piece of data that's worthless operationally. What actually gives CSAT real value is having a defined process for what happens next: does someone review low ratings within 24 hours? is there a way to reach back out to that customer if the comment suggests the problem actually wasn't resolved?
CSAT as a thermometer, not a final verdict
An aggregate CSAT number is useful as a trend over time, but it's dangerous to treat it as the single verdict on the quality of your support. It's worth looking at it broken down — by channel, by type of query, by whether the AI or a human resolved it — because a healthy average can be hiding a channel that systematically leaves people unsatisfied.