What Text Analytics Actually Analyzes: Your Full Response Set, Not a Sample¶
A reasonable worry before trusting an AI-generated category set: did it actually consider everything respondents wrote, or extrapolate from a slice of it? This matters because a category set built from a partial read can miss a real theme that only shows up in the responses it never looked at.
Category Generation Looks at the Full Response Set¶
When you generate categories for an open-ended question, Text Analytics analyzes the complete set of responses to that question - not a sample, and not a subset. The categories it proposes are drawn from themes actually present across everything respondents wrote for that specific question.
Classification Covers Every Response¶
The classification step - assigning each response to a category from your approved set - runs against every response to the question, not a representative sample projected onto the rest. Every individual response gets its own classification, which is also what makes response-level review and correction possible afterward (see Reviewing and Correcting Individual AI Classifications) - there's no response that was never actually looked at.
Per-Question, Not Cross-Question¶
Text Analytics generates and classifies categories one open-ended question at a time. It doesn't pool responses across multiple different questions into a single category set - each question gets its own independent analysis, which keeps a category set specific to what that particular question actually asked.
What This Means for Small vs. Large Response Sets¶
Because the analysis covers everything rather than sampling, the categories you get reflect your actual data regardless of how many responses you have - a smaller response set won't be padded out with a genericized category list, and a larger one won't be summarized down to only its most common themes at the expense of smaller but real ones. That said, a very small response set naturally produces a thinner set of categories simply because there's less variety in what was actually said - see How Many Responses Before Category Generation Is Worth Running? for more on that practical threshold.
FAQ¶
If I add more responses later, does the existing category set update automatically?
No - categories reflect the responses that existed when you generated them. If you add meaningfully more responses later, re-running category generation (see Running Multiple Classification Passes on the Same Question) lets you check whether new themes have emerged.
Does Text Analytics treat longer responses differently from short ones?
Both are analyzed as part of the same full-response-set read - there's no separate handling based on response length, though a very short or low-content response (like "good" or "n/a") naturally gives the analysis less to work with than a detailed one.
Can I run Text Analytics on a subset of responses on purpose - like just one segment?
Category generation and classification run on all of a question's responses at once - filtering to a specific segment beforehand isn't part of the standard workflow. If you need segment-specific analysis, cross-tabbing the finished categories against a segmenting question in Reporting (see Cross-Tabs and Combined Survey Analysis in Reporting) is the way to see that breakdown after classification.
Next: for how the two-step process itself works end to end, see How Text Analytics Works: From Open-Ends to Categories.