Ten. Below ten complete responses in a period, RawQuotes names no insight at all. This post explains what counts as one complete response, what you see while you are under ten, and why we would rather show you nothing than a guess.
What counts as one complete response
One person who reached the end of your survey and answered at least one question marked "Count for insights". If they answered five of ten questions, that is one complete response. If they answered all ten, it is still one. We count responses, not the answers inside them.
Feedback from other sources counts the same way. One file row, one pasted block or one item sent through the API is one complete response. The ten are counted across every source in the project together, in the period you are looking at.
What you see under ten
No insights. Instead you see how many complete responses you have, how many are needed, how many are missing, and a link to the raw responses, so you can read them yourself in the meantime. If you are looking at a period that is still open, the panel also offers to switch to the last finished one.
Why not name something from six answers
Grouping a handful of answers produces something that looks convincing and means nothing. With six answers, one person with a strong opinion is a sixth of everything you have. An insight named from that looks exactly like one named from six hundred: the same sentence, the same card. You would have no way to tell them apart, so we don’t draw the first one.
Ten is not a statistical law. It is the point below which we would rather not draw a conclusion, and it is the same on every plan. Paying more does not lower it.
A second threshold: three
An insight also needs at least three responses behind it in the period you are looking at. With fewer, it is not listed. Filter by one source, and an insight with fewer than three responses from that source is marked “Below threshold for this source”: you see how many of the three it has from that source, but no quotes.
What this costs you
There will be weeks when RawQuotes shows you nothing. A new survey on a quiet page can sit under ten for a while. We think that is better than an insight you can’t trust, and the raw answers are there to read until then.
Once there are ten, every insight comes with the answers it was counted from. How to read them critically, in RawQuotes or any other tool: how to check AI survey analysis against raw answers.