I have two points here.

Firstly, you should always use such clear wording that your respondents understand the question exactly in the way you need it. For example, for me a question asking to rate the promptness of a bank's service is certainly not about knowledge management and not about reliability but about the speed of service. If your target population understands it differently, you have to rephrase the question.

Secondly, each of your questions should be about a single issue. Examples in best practices typically focus on (erroneously) asking multiple questions in one question. But if you're really asking a single question about a single issue and your audience understands it in several different ways, then the problem is in wording. Don't delete the question, just rephrase it and re-test on the pilot sample.

I hope this is not a let down.  But I agree with Alexander Tarvid.  

It is difficult to measure something that has multiple meanings, whose thought may have more than one interpretation, or that the respondent can not make complete sense.  It is important to have one to many correspondence in designing the instrument.  For example: beauty means factor 1, factor 2 and factor 3.  Having many to many correspondence is a faulty design.  For example beauty means factor 1, factor 2, factor 3, and intelligence means factor 1, factor 3, factor 5, and factor 6.

Research is rigorous and not done in one sweep.  It is very important to pretest the instrument so that all the issues you may encounter during the actual testing can be averted.  

Now, to help you address your problem.  You can discard your initial run.  Make that a learning curve.  Come up with pretest.  Then revise your instrument based on feedbacks of respondents and your observation on their difficulty.  Then conduct actual test.  

Good luck to your study

Just to illustrate the point, let's talk about research on systems success. In this case, success could mean use, user satisfaction, and utility. These are surrogates (sometimes called constructs) of success. For each surrogate, you can have several measures (proxies) like access numbers, unique access, Likert scale, perceptions...

When a measure correctly represents constructs, which in turn also represent the primary variable to be tested we say it have a high internal validity. This representativeness gives us trustworthiness in research results.

Thus, you cannot ask "Do you think the system is a success?". You should ask: Would you use it?; Do you like it?; and Is it useful? So you will have to break your question, as indicated above, in as many questions as 'prompt' could mean.

A pre-test is made to discover these shortcomings.