Previewing and testing
The Preview tab walks your survey exactly as a respondent will: real questions, real logic, real validation, real page transitions.
Preview responses are real responses, kept separate
A preview genuinely submits. It creates a response, it can be completed, and you can open it afterwards and read it back — which is the point, because that’s how you check what a given path actually recorded.
Those responses are marked as previews and kept out of your results and limits. In the responses list they carry a Preview badge, and they’re excluded from counts. So preview as often as you like; you won’t have to clean anything up afterwards.
Preview runs your working copy
Preview always shows the survey as it stands right now, not any version pinned to a distribution. That’s what makes it useful for checking edits before they go out.
It’s also the usual explanation when a respondent describes something you can’t reproduce: they’re on an earlier pinned version. Check the version number on their distribution. See Editing a live survey.
What to actually test
Clicking straight through proves very little. The parts that break silently:
- Every logic path. Take the yes branch, then the no branch, then the one nobody will pick. A broken rule doesn’t error — it asks the wrong person the wrong question and you find out at analysis.
- Screen-outs and early endings. Confirm the respondent lands somewhere sensible, not in an unexplained stop.
- Validation. Try to get past a required question. Enter something malformed in a format-checked field and read the message as a respondent would.
- Piping. Check the piped answer appears where you expect and reads grammatically in the sentence around it.
- Groups. Add two or three repeats, then read the response back and confirm each set is stored separately.
- On a phone. Matrices and rank questions are where desktop-designed surveys fall down.
Before a real launch
Run a small pilot through an actual link distribution with people who aren’t you — the questions you’ve reread thirty times are exactly the ones whose ambiguity you can no longer see. Then expire that distribution and create a fresh one for the real thing.