One question at a time, genuinely one
«How satisfied are you with the speed and courtesy of the service?» is two questions glued together. Someone who found it fast but rude has no score to give, so they invent one: the number you get measures neither speed nor courtesy.
The warning sign is the conjunction: every «and» inside a question deserves suspicion. If the sentence contains two things that can be judged, that's two questions — or a matrix, which is how you ask four without asking four.
Don't lead the answer
«How much did you like the new design?» assumes they liked it and scores higher than «What do you think of the new design?». The same goes for adjectives in the intro: «our recently improved service» shifts the answers before the question even arrives.
The neutral version is nearly always uglier to read, and that's fine. A survey isn't marketing copy: if the question sounds flat, it's probably correct.
Options must cover everyone, and not overlap
Two symmetrical mistakes. Not covering everyone: age bands «18–25, 26–35, 36–50» exclude minors and the over-50s, who then pick at random or leave. Overlapping: «0–5, 5–10» doesn't say where someone with exactly 5 goes.
The fix is adding an exit — «other», «don't know», «prefer not to say» — whenever the list might not be enough. It costs one option and saves the responses of everyone who would otherwise make something up.
Ask about behaviour, not intentions
«Would you use it?» and «What would you pay?» collect polite answers that predict nothing: stating an intention costs nothing, and people are optimistic about themselves. «How do you solve it today?» and «How much time does it cost you each week?» collect facts.
It's the difference between a survey that confirms what you hoped and one that tells you something. In practice: wherever you were about to write a conditional, try rewriting the question in the past tense.
One question per question, no leading adjectives, options that cover everyone — and ask what happened, not what they would do.