Choosing Between AI-Generated and Manually Built Categories¶
Text Analytics doesn't require you to start from an AI-generated draft - you can build a category set entirely by hand and classify against it, the same as you would an AI-generated one. Which starting point makes more sense depends on what you're walking into the project already knowing.
When AI-Generated Categories Make Sense¶
If you're exploring a question with no strong preconception of what themes will come up, letting Text Analytics draft categories from the actual responses is usually the faster path - you get real, data-driven candidates instead of guessing at a codebook before you've read anything.
When Building Manually Makes Sense¶
If you already have a fixed taxonomy - a standard coding scheme your team uses across every wave of a tracking study, or a client-specified set of categories you need to apply exactly - building the category set by hand and skipping generation entirely keeps you from having to edit a generated draft down to match something you already know you need.
A Hybrid Middle Ground¶
These aren't exclusive. A common approach: generate categories to see what the AI surfaces, and use that as a sanity check against your own planned taxonomy - if the generated set surfaces something you hadn't planned to code for, that's worth knowing before you finalize your own manually built set. See Using AI-Generated Categories as a Starting Point for a Custom Taxonomy for more on this approach.
What Doesn't Change Either Way¶
Regardless of how the categories came to exist, classification, confidence scoring, manual review, and export all work identically - Text Analytics doesn't treat an AI-generated category set differently from a manually built one once classification runs.
Consistency Across a Tracking Study¶
If you're running the same survey question repeatedly across waves and want directly comparable results wave over wave, manually building (or carefully locking down) your category set is usually the safer choice - letting AI generate a fresh set each wave risks producing categories that don't line up cleanly with prior waves, since each generation run analyzes only that wave's responses. See Comparing Category Sets Across Two Survey Waves for more on this specific situation.
FAQ¶
Can I build categories manually and still get AI classification?
Yes - classification runs against whatever category set exists, whether it was generated or built by hand.
Does building categories manually cost credits?
No - only category generation and classification are credit-based. Manually creating categories yourself doesn't consume credits.
Can I mix a manually built category with AI-generated ones in the same set?
Yes - once a category set exists, you can add your own categories to it alongside any AI-generated ones, and edit either kind the same way.
Is manually built classification as accurate as AI-generated?
The category set itself (generated or manual) doesn't affect classification accuracy - classification quality depends on how clearly your categories are defined, not on how they were created.
Next: if you're leaning toward building manually for a repeated study, see Comparing Category Sets Across Two Survey Waves.