STAYCHARTED BLOG · GOOGLE SHEETS
How to categorize Google Sheets rows with AI using your own categories
Automatically categorize Google Sheets rows using your labeled examples. Train a model, fill new rows, check confidence and review uncertain answers.
Start with the decisions already in your sheet
A spreadsheet often contains two jobs: facts someone collected and a category someone chose. The message is the fact; Billing, Access or Feature Request is the decision. If you repeatedly make that second decision, reviewed rows can become training examples.
Imagine a customer-feedback sheet with Message and Category columns. Your team sends refund requests to Billing even when the message also mentions signing in. A model needs examples of that convention, not just a dictionary definition of “billing.”
| Text in a row | Your team's category | What teaches the distinction |
|---|---|---|
| I cannot sign in to request my refund | Billing | The refund is the work to route |
| My verification code never arrives | Access | The request is about account access |
| Could you add a dark theme? | Feature Request | The customer wants a product change |
Use the final reviewed category. Do not train on a mixture of initial guesses and corrected answers.
How to categorize new rows in Google Sheets
- Prepare examples. Put one record on each row, with the text fields and the category your team approved. Give columns clear names and resolve contradictory labels.
- Install the add-on. Open StayCharted AI Model Trainer in Google Workspace Marketplace. Sign in to your StayCharted workspace from the add-on.
- Choose what the model reads. Select the useful input columns and the answer column. Leave out IDs or information that would not be available for a new row.
- Check and train. Review the data checks, train the model and examine its held-out results. Look at errors by category, not only the overall score.
- Fill new rows. Use the model on the new records and select the output column. The add-on writes answers and confidence into empty cells; keep reviewed labels separate from new suggestions.
- Review and improve. Check uncertain answers and record corrections for the next training version. On Essentials and Business, the Review Queue supports this review process.
Keep a copy of the original sheet for comparison. Use the same input columns and category meanings when filling as you used for training.
When to use a trained model instead of an AI formula
Google Sheets also supports categorization through Gemini. Google's AI in Sheets announcement describes categorizing and summarizing rows. That can be a useful starting point for a one-off task.
A trained model is worth comparing when you repeatedly apply the same categories, already have reviewed examples, and want measured results before relying on the answers. Test both approaches on the same untouched rows. Compare mistakes and review effort as well as speed.
The point is not that a prompt cannot classify. It is whether your team's past decisions improve the results for your task.
Choose useful columns and keep data access narrow
More columns do not automatically make a better model. A message and product type may help; a random row ID usually does not. A field filled only after a human decides the category can leak the answer into training and produce an unrealistic score.
The Sheets add-on uses the current-spreadsheet permission. Selected column contents are sent to StayCharted when you check data, train or fill. Read how Google Sheets data is handled before including customer information.
Start with a small, reviewed batch and compare its answers with what your team would have chosen. Expand only when the categories you care about perform well.
Common questions
Can AI categorize Google Sheets rows into my own categories?
Yes. With labeled examples, StayCharted learns the categories your team uses and applies them to new rows. Check held-out results before relying on the output.
Do I need to write a formula or code?
The StayCharted Google Sheets add-on provides the training and filling workflow without code. You choose the input and answer columns.
Does it overwrite my existing labels?
The add-on writes answers and confidence into empty cells in the selected output area. Keep your approved labels separate from suggestions and retain an original copy.
Can I use Gemini instead?
Yes. Compare a prompt-based approach and a trained model on the same held-out rows. A trained model is particularly useful when your reviewed examples encode team-specific category rules.