Multiple choice is the right format when you already know what the answers are and want to count them, and the wrong one when you are still finding out. That distinction decides most of the value: options written before you have read fifty open answers are a list of your own assumptions, and every respondent who picks the nearest one is quietly confirming them for you.
Earn the options before you offer them
Ask openly first, read the answers, and let the repeated ones become your options. Done that way the choices are the customer's own categories and the counts mean something. Written from the inside instead, the list reflects how your team talks about the product, and the results tell you how well your customers can guess at your vocabulary.
Make the options exclusive and complete
Two choices that overlap split the same answer in half and both look smaller than they are. A list with no other option forces everybody into the nearest wrong box, and the resulting tidiness is entirely artificial. Always include an other with a short free-text field, and read what people write in it, because that is where next round's options come from.
Order changes the answer
People pick early options more often on a screen and later ones more often when read aloud, and grouping related choices together nudges people towards the group. Randomising the order where the tool allows it removes most of this, and where it does not, at least keep the order fixed between rounds so the bias is constant rather than moving.
Single or multiple selection, decided deliberately
Letting people pick several answers gives a richer picture and makes the counts stop adding to a hundred, which is fine as long as everybody reading the results knows it. Forcing one choice is cleaner to read and throws away detail. Neither is wrong; presenting a multi-select result as if it were single-choice is.
Questions people ask about multiple choice survey questions
When should a customer survey use multiple choice?
Once you know what the common answers are from having asked openly. Before that, open questions tell you far more.
How many options should a question have?
As few as cover the real answers, plus an other. Long lists cost attention and encourage people to pick the first plausible option.
Should we randomise the order?
Where the tool allows it, yes. Where it does not, keep the order identical between rounds so at least the bias does not move.
Is an other option always needed?
In a customer survey, effectively yes, and the free text under it is where you learn what the next version should offer.