The count isn't the problem, the time is
Nobody opening a survey counts the questions: they form an impression of how long it'll take. Five 1–5 scales take twenty seconds; three open questions require thinking, and thinking is what nobody wants to do in a post office queue. A ten-question multiple-choice survey is, in practice, shorter than a four-question open one.
So the useful rule isn't «seven questions maximum» but «two minutes maximum». And two minutes can be measured: open the preview and actually fill it in, without skipping, the way someone who didn't write it would.
Every extra question also costs you the earlier answers
This is the part people underestimate. Someone who abandons at question eight doesn't leave you seven useful answers and one gap: they leave an incomplete response, which has to be handled separately in the stats and often can't be used at all. The extra question doesn't just cost itself, it costs retroactively.
Which is why «do I really need this?» should be asked backwards: not «would I like to know?» — the answer to that is always yes — but «what would I do differently depending on the answer?». If nothing changes, that question is occupying the space of a response.
How to find out which one to cut
No guessing needed: if the survey has run once, the per-question drop-off tells you exactly where people stop. It's a single column in the Responses tab, and it almost always points at one specific question — usually the first open one, or the first that asks something personal.
If the survey hasn't gone out yet, the honest substitute is sending it to five people and asking where they got bored. Five is enough: the breaking points of a questionnaire are not subtle.
The exception that matters
All of this holds when the person answering owes you nothing: customers, subscribers, the public. If they have to fill it in anyway — a job application, an order, a mandatory internal form — length stops being the main constraint and clarity becomes more important. There, twenty questions are fine, as long as it's clear what they're for.
Don't count questions: time your own survey by actually filling it in, and cut anything that wouldn't change a decision of yours.