Keep it short. The scenario has done the work of putting someone in a moment; the question just has to ask them what they do next, the way a person would hear it.
Write it as someone would hear it
Second person, present tense, plain:
When you check out, will you leave a tip for your driver?
Not:
What is the likelihood that the respondent will leave a gratuity?
The second version asks a twin to step outside itself and estimate a probability about a person. Twins answer that kind of question the way anyone would — with a guess about people in general, which is exactly what we are trying not to measure.
Ask about behaviour, not opinion
The sharpest questions ask what someone does. The weakest ask what they think.
| Weaker | Stronger |
|---|---|
| How important is price to you? | Which one do you put in your basket? |
| Would you consider switching? | Do you switch, or stay? |
| How do you feel about the fee? | Do you complete the order? |
Behaviour questions produce answers you can act on, and they are the ones validation can actually check.
One decision at a time
A question that contains two decisions gets you a distribution over a muddle.
Will you order again, and would you recommend it?
Those are two simulations. Run them separately, on the same population and date, and you can look at both answers side by side and know which is which.
Match the question to the options
The question and the action space have to fit together. If the question is "will you tip?" the options are about tipping. If the options include "abandon the order", the question has to be wide enough to admit that — "what do you do?" rather than "how much do you tip?".
Mismatches here are a common cause of a strange-looking result: twins are being asked one thing and forced to answer with options about another.
Do not smuggle in the answer
Questions that name a consideration make it salient, and salience moves results.
Given how high the fees already are, will you leave a tip?
That will produce fewer tips, and the drop is your wording, not your customers. If the fee level is a real part of the situation, it belongs in the environment as a fact everyone gets, not in the question as an emphasis.