Using AI-Generated Categories as a Starting Point for a Custom Taxonomy¶

There's a false choice that shows up whenever AI-generated output is on the table: use it as-is, or throw it out and build from scratch. With Text Analytics' category generation, neither extreme is necessary - the generated set is fully editable, which means it works well as raw material for a taxonomy you shape yourself, not just a finished product to accept or reject.

Why Start from a Draft Instead of a Blank Page¶

Building a category set from nothing means reading a meaningful chunk of responses yourself just to get a sense of what themes exist before you can even start naming categories. A generated set skips that step - even an imperfect one gives you real candidate themes, pulled from your actual responses, to react to. Reacting to a draft is almost always faster than generating ideas cold.

Treat It Like a Colleague's First Pass¶

The useful mental model: read the generated categories the way you'd read a research assistant's first attempt at a codebook. Some categories will be exactly right. Some will be close but need a better name. Some will be too broad or too narrow. And some real themes may be missing entirely. None of that means the draft failed - it means it did its job of giving you something concrete to work from.

Building Your Own Taxonomy On Top¶

From there, use the full range of editing tools (see Editing and Refining Categories in Text Analytics) to shape the set into something that reflects how you and your team actually think about the data - rename categories into your team's own language, merge ones that are really the same theme, split ones that are bundling distinct ideas, add categories the AI missed, and delete ones that don't hold up.

If you have a pre-existing taxonomy you use across multiple studies - standard categories your team always codes for - you can also use the generated set purely as inspiration for what else might be present, while manually building your final categories to match your established taxonomy.

When to Skip the Draft Entirely¶

If you already have a fixed, well-established coding scheme you need to apply exactly as-is, building it manually from the start may be more direct than editing a generated draft down to match it. Generation is most useful when you're open to what the data itself suggests, not just applying a predetermined scheme.

FAQ¶

Does editing the generated categories cost credits?
No - editing is free. Credits apply to the generation and classification runs themselves.

Can I delete every generated category and start over manually within the same set?
Yes - deletion and manual addition are both available, so you can clear out a generated set entirely and build your own from the same starting point if that ends up being faster than editing individual categories.

Will my edited category names get overwritten if classification runs again?
No - classification uses whatever category names and definitions currently exist at the time it runs, including your edits.

Is there a way to save a taxonomy I've built for reuse on future surveys?
Categories are built per question within a survey - check your Text Analytics workflow for whether a category set can be reused as a starting point on a different survey's question.

Ready to start editing? See Editing and Refining Categories in Text Analytics for the full set of tools available.