Discord Ticket Feedback: How Star Ratings Work, and How to Read Them

Most support tools ask how it went by sending an email the next day. A Discord ticket bot cannot do that, because it has no email address for anybody and never will. What it has instead is the one message it is already sending: the close notice that lands in the opener's direct messages with their transcript attached.

So the rating rides along on that message. That decision, taken because there was nowhere else to put it, is what makes a Discord satisfaction score behave unlike every other one you have used.

How a member rates a closed ticket

Nothing happens while the ticket is open. The prompt exists only at the end, built into the close rather than sent afterwards.

The direct message is the same one that carries the transcript link, the server's name and the reason the ticket was closed. Beside that link sits a button labelled Leave Feedback. Pressing it opens a form titled How did we do?, with a star rating from one to five and an optional box underneath asking whether there is anything to add. Submitting returns a private thank you and greys the button out.

Two things are checked before it saves, and both are checked on our side rather than trusted from the click. The person pressing the button has to be the member who opened that ticket, so a forwarded message cannot be rated by somebody else. And the rating has to be the first one for that ticket, which the database enforces directly, so there is no version of this where a second answer quietly overwrites the first. The whole exchange appears in whatever language the server is set to, because the prompt is part of the bot's interface rather than a survey tool bolted on beside it.

Why it arrives in a DM, and why there is only one shot

A ticket bot deletes the channel when the ticket closes. That is the design, and what a close actually does covers it properly, but the consequence for feedback is worth stating: at the exact moment you would like to ask how it went, the room the conversation happened in no longer exists.

The opener's direct messages are the only surface left. That gives you one delivery, and you cannot repeat it.

The right hand column is not a worse version of the left. Some of it is better: asking while the answer is still in someone's head beats asking on Thursday about Monday, and the reply arrives without the member leaving the app. The rest is a real cost, and the honest way to hold both is that a Discord rating is a verdict rather than a conversation. You get the number, you get whatever they chose to type, and then it is closed.

Who can read the ratings, and who cannot

Ratings are visible to Manage Server, and to nobody else.

In Discord, /ticket feedbacks opens a private viewer: total and average in the header, then the stars, the ticket, the member, the note and the date, five to a page, filterable to a single star count. The dashboard shows the same ratings on the web, with the spread across all five and a link to the ticket each one came from.

That gate matters more than it looks, because of who it excludes. A staff member whose role can claim and close tickets cannot read the ratings left on their own work. Neither can the member: their rating is deliberately withheld from the ticket history they are shown, so what they wrote is never quoted back at them.

Ratings are not anonymous to an administrator. Each one is stored against the member who opened the ticket and displayed that way. That is the trade, and it is the right way round: a rating you cannot trace back to a conversation tells you a mood and nothing else. It does mean the notes are candid text about named people, which is why reading them is an administrator's job rather than a staff wide feed.

What the average hides

That alone should stop you quoting the mean at anybody, and there are two more.

Ratings are self selected. Only people who open the DM, notice the button and choose to press it are in your sample, which is not the population that opened tickets. Small servers feel this hardest: at thirty ratings, two annoyed members move the average further than a month of good work does.

The bigger gap is what the record does not contain.

There is no field saying whether a human answered that ticket or whether the AI closed it alone, and roughly half of all tickets across every server running this bot close with no human involved at any point. So a share of your satisfaction score is not a measurement of your team at all. It is a measurement of answers your staff taught the AI months ago, landing in tickets nobody read.

That figure is live at /api/stats/global rather than typed into this page. Your own split will be different, and separating the two halves of your score means opening the tickets behind the low ratings rather than reading the summary, which is what you should be doing anyway.

What a one star is actually worth

More than the four stars around it, and not because bad news is more interesting.

A low rating with a note is the only unprompted writing a member does about your support after the conversation has ended, when there is nothing left to gain by being polite. Both viewers let you filter to one and two stars for this reason, and the dashboard marks those cards so they are hard to skim past.

The useful move is always the same. Open the ticket the rating came from, read the transcript, and work out which of two things went wrong: the answer was wrong, or the answer was right and took too long to arrive. Those have completely different fixes and the star does not tell them apart.

If the AI got it wrong and a staff member put it right in the same ticket, that ticket is worth feeding back in. /ai improve turns that conversation into training, so the next member asking gets the staff answer instead of the mistake, and how the AI's memory works explains what it keeps and what it does not.

There is a catch, and it bites on the tickets you most want to fix. That action refuses a ticket the AI resolved on its own, and it refuses a ticket with no staff reply in it. A one star on a ticket the AI closed alone is both at once, so there is nothing in the conversation to learn from: the only substantial text in it is the wrong answer. Those you fix by teaching the correct answer yourself, in your own words, which takes about two minutes and is the highest value thing a bad rating can point you at.

What ticket feedback does not do

Each of these is a boundary rather than a roadmap item, and better known now than discovered halfway through building a process around it.

There is no reminder and no second attempt. One button on one message. If it goes unread, that ticket has no rating and never will.

There is no way to reach somebody who has DMs closed, or who has left the server. They see nothing, and they are not counted as a non response anywhere, so your rating count is quietly a count of people who could receive a DM.

There is no per staff member score. The rating belongs to the ticket, not to whoever claimed it. You can see who was in a given ticket, but nothing totals it per person, and given how many tickets more than one person touches, that is probably honest rather than missing.

There is no survey builder and no export. One fixed scale, one optional box, read where they are stored.

And there is no editing, by anyone. A rating is final for the member who left it and final for you. Ratings are kept up to 24 months and then purged, which on the free plan outlasts the 90 day transcript behind them, so an old note can survive the conversation it was about.

Is a satisfaction score worth tracking in a Discord server?

Sometimes it is real signal and sometimes it is decoration. The difference is mostly volume, and what you intend to do with it.

The version that works is unglamorous. Once a week, filter to one and two stars, open the three or four tickets behind them, fix whatever they point at, and ignore the average entirely. That beats watching a number a preselected fifth star was going to flatter anyway. If you want the figures that genuinely move, support analytics is the honest guide to which ones mean something.