How Text Analytics Works: From Open-Ends to Categories¶

Open-ended questions get you the richest answers in a survey and the least analyzable ones. A thousand free-text responses can't go straight into a bar chart - someone has to read them, decide what themes actually show up, and tag each response against those themes before any of it becomes a chart at all. Done by hand, that's the part of research that eats the most hours for the least visible progress. Text Analytics is built to do the first pass of that work for you, while keeping you in control of the result.

The Two-Step Workflow¶

Text Analytics splits the work into two distinct steps, on purpose - generating categories and applying them are different judgment calls, and conflating them is how you end up with categories nobody actually reviewed.

Step one: generate categories. Point Text Analytics at an open-ended question and it reads through the responses to draft a set of categories that reflect the themes actually present in what people wrote - not a generic, predefined list applied to every question regardless of content. This step produces a starting set of categories for you to react to, not a finished taxonomy.

Step two: classify responses. Once you're happy with the category set, Text Analytics goes back through every response and assigns each one to the category (or categories) it best matches. This is the step that turns free text into something you can actually chart, cross-tab, or include in a report.

You're Always in the Loop¶

Nothing in this workflow is a black box you have to accept as-is. The category set Text Analytics drafts is a starting point you can rename, merge, split, delete, or add to before classification runs - see Editing and Refining Categories in Text Analytics for how. And once responses are classified, individual assignments can be reviewed and corrected if something's clearly off, the same way you'd spot-check any first-pass coding a research assistant handed you.

Results generated by AI are visibly marked as such throughout the interface, so it's always clear which parts of your category set or classifications came from the automated pass versus what you've since edited by hand.

Confidence and Quality¶

Classification results can include a confidence indicator alongside each assignment, giving you a signal for where to focus manual review rather than treating every classified response as equally certain. This is meant to guide spot-checking, not to replace it - treat a low-confidence assignment as a flag worth a second look, not an error to assume is wrong.

What Happens to the Original Text¶

Classifying a response never discards or rewrites what the respondent actually wrote - the original verbatim stays attached to every response, category assignment included, so you (or anyone reviewing your report later) can always trace a chart back to the actual words behind it.

FAQ¶

Does Text Analytics work on any question type?
It's built for open-ended text responses - questions with predefined answer options (single-select, scale, and similar) don't need it, since they're already structured.

How long does classification take?
It runs in the background rather than requiring you to wait on the page - you'll be able to continue working elsewhere while it completes, and come back once it's done.

Can I reclassify responses if I change the category set later?
Yes - editing your categories and re-running classification applies your updated categories to the response set again.

Does using Text Analytics cost credits?
Yes - both generating categories and classifying responses are credit-based actions. You'll see an estimate before committing to a run.

Can I use my own category list instead of the AI-generated one?
Yes - the AI-generated set is a starting point, not a requirement. You're free to build your category list manually instead, or use the generated set purely as a first draft and edit it into something entirely your own.

Next: once you have categories you're happy with, see Editing and Refining Categories in Text Analytics for how to fine-tune them before classifying.