Categorize
Categorize reads a column of free text, groups rows that say similar things, and names each group, so you can see the themes without deciding them in advance.
Who can see this: anyone who can open the project. Running it adds a step, so you need to be able to edit the project.
- The column: the text Categorize analyzed.
- The granularity chips: how broad or specific the categories should be.
- The new column: where the categories will go.
When to use it
Section titled “When to use it”Use Categorize when you have lots of text and no fixed list of categories: survey answers, support tickets, feature requests, notes. It finds the groups for you.
If you already know the categories you want, use the Researcher and name them.
Ask for it
Section titled “Ask for it”Ask in project chat. For example:
Find the main themes in these notes.
Categorize picks the text column to work from. It needs at least 20 rows with text.
Choose how detailed
Section titled “Choose how detailed”Before it labels anything, Categorize looks at the patterns in the text and says how many rows it analyzed. Then it offers three levels, each as a chip:
| Level | Starting label | What it gives |
|---|---|---|
| Broad | ~10-15 broad themes | High-level groupings for a broad overview |
| Medium | ~30-80 topic areas | Balanced categorization with moderate detail |
| Specific | Specific categories | Fine-grained categories for detailed analysis |
The labels and descriptions on the chips are written for your data, so yours may read differently. The chat also says “Categories will be added as a new column called {column}.”
Pick a level, or Cancel. On the free plan, a table over 1,000 rows also offers Upgrade for more rows.
While it runs
Section titled “While it runs”Categorize runs in the background as a Categorize step and reports its progress in five phases, such as “Discovering clusters… 4/5”. A large table can take a few minutes.
What you get
Section titled “What you get”The same table with new columns:
| Column | What it holds |
|---|---|
| {column} | The category for the row |
| {column}_confidence | How well the row fits its category |
| {column}_estimated | Whether the row was ambiguous and assigned to the nearest category |
| {column}_detail | A finer label, only with the specific level |
Rows with no text are labeled “Other / Mixed”. The chat sums up the result, including how many categories it found and the five largest.
If more than a quarter of the rows were ambiguous, the summary says so and points to the _estimated column.
Limits
Section titled “Limits”| Free plan | Paid plans | |
|---|---|---|
| Rows | First 1,000 | Up to 100,000 |
| Minimum | 20 rows with text | 20 rows with text |
Categorize caps the number of categories. If the first pass finds too many, it widens the groups automatically.
Credits
Section titled “Credits”Asking uses a credit for the question, as usual. Categorize doesn’t add a separate charge. Your plan may differ; see How credits are counted on Monitor usage.
Good to know
Section titled “Good to know”- Ask for Categorize in chat, so it starts from its own preview. A plan step that skips the preview stops with “Categorize requires configuration from the preview flow.”
- Category names are written by AI, so running it again, including on a schedule, can label the groups differently.
- Look at the broad themes first, then run again with a more specific level where you need detail.
Related
Section titled “Related”- Researcher: sort rows into categories you name
- Chat
- Projects