The quantification of qual: why open-ends are becoming your most valuable data
By Centico Research · 3 min read
For years, the open-ended question lived a strange double life. Researchers loved writing them — they are where respondents say the things no rating scale can capture. And researchers quietly dreaded them, because turning a few thousand messy sentences into something countable meant days of manual coding, or worse, skimming a sample and hoping it was representative.
So open-ends got treated as a garnish. A few quotes for the appendix. A word cloud if the deadline was tight. The real analysis happened in the closed questions, where the numbers already lived.
That trade-off is disappearing — and it changes what your most valuable data actually is.
From a sample to a census
The old constraint was human attention. A coder could only read so many verbatims in a day, so teams read a slice and generalised. AI removes that ceiling. It is now practical to read and code every single response in a study — not a sample of them — in the time it once took just to brief a coder. When you code the whole base instead of a fraction, small but important themes stop slipping through. The complaint mentioned by 4% of respondents — the one that turns out to be your real churn driver — finally shows up.
Longer answers, more signal
Something else is shifting at the same time. As surveys become more conversational, respondents simply write more. Follow-up prompts and chat-style questions are pulling answers two to three times longer than a flat text box ever did. More text used to mean more work. Now it means more signal — provided you can structure it without flattening the nuance. That last part is the catch.
Why a human still reads the output
It is tempting to point an AI at a column of verbatims, accept whatever themes come back, and call it coded. We don't, and neither should you. Models are fast and consistent, but they miss sarcasm, fold two ideas into one tidy label, and quietly drift when a theme appears that wasn't in the original frame. The job is not to replace the coder — it is to let the machine do the first, exhausting pass at full speed, then have an experienced person confirm the frame, catch the edge cases, and sign off. Speed from the model, judgement from the human. That combination is what makes the output trustworthy enough to put in front of a client.
The practical takeaway is simple. The open-ends you have been treating as colour are now the part of your study most likely to contain something you did not already know. The teams that win in 2026 are not the ones collecting more data — they are the ones finally reading all the data they already have.
— Centico Research
