Twins must finish on exactly one of your options. They cannot answer off the menu and cannot decline. That constraint is what makes results addable into a distribution — and it is also why an option you left out does not simply go unmeasured. It gets absorbed into the options you did list.
The failure that matters
Suppose you ask whether people switch to a cheaper rival, and you offer two options: switch, or stay. For part of your population the honest answer is "I would not buy either — I would skip it this week." Those people cannot say that. They will be pushed into switching or staying, and your percentages will move by however many of them there were.
Nothing in the output flags this. The distribution looks clean. The result is wrong in a direction you cannot see.
So the single most useful discipline in setting up a simulation is asking: is there a real thing a real person would do that is not on this list?
Checking for a missing branch
Two ways, both cheap.
Run it with a catch-all. Add "something else" or "none of these" and run it once. If the catch-all takes a meaningful share, you are missing a real branch — open a few of those twins and read what they say they would actually do. Then rerun with the option named properly.
Read the second choices. Every twin also records what it would do if its first choice were unavailable. When second choices pile onto one option, or when twins visibly strain against the options in their reasons, the space is too narrow.
Writing the options
Make them mutually exclusive. "Buy the large pack" and "buy on promotion" can both be true, and a twin forced to pick one is guessing.
Make them concrete actions. "Leave a tip" and "no tip" are things people do. "Positive sentiment" and "negative sentiment" are not.
Keep them parallel. Options written at different levels of detail attract attention unevenly. Three plain options and one long, carefully argued one will not perform equally, and the difference is your writing.
Two at minimum, and few is better. Two to five options is the usable range. Beyond that twins are choosing from a list rather than deciding, and the shares get thin enough that the differences between them stop meaning anything.
Do not include an option nobody could pick. Adding an obviously bad option to "make the choice realistic" tells you nothing and takes share from the option nearest it.
Order
You may notice that a result never depends on the order you typed the options in. Presentation order is varied across the population, because a list presented the same way to everyone would let a position preference build up into what looks like a finding. The order you enter is for your convenience; the twins do not all see it that way.
When none of the options is good
Sometimes the honest answer for most of your population is that they would not do any of this. That is a finding, not a problem — but only if they can say it. Give them the option, and let the size of it tell you.