Copilot, ChatGPT and the =AI() style add-ins use a general chat model: you describe what you want, and the model writes an answer. AI Categorizer uses a decision model (Jev, from TypeSafe AI, run through Cloudflare) that is asked one question per row, which of your categories fits, and can only answer with one of them or Other.
Where the difference shows #
| Chat model in Excel or a chat window | AI Categorizer | |
|---|---|---|
| What comes back | Text it writes. It can write a category that is not in your list, or reword one of yours. | One of your categories, exactly as you wrote it, or Other. It cannot invent a category that is not in your list. |
| Doubtful rows | Review and confidence fields depend on the tool and how you set up the task. | A confidence and a status on every row; rows below your threshold are marked Review. |
| Long lists | File handling, row limits and batching depend on the tool and plan; check that every row was processed. | Up to 50,000 rows per run, written into new columns next to your data. Repeats and rows that already have a result are free. |
| Same data, next month | Answers can differ from one chat to the next. | Your setup is remembered for the workbook, and rows that already have a result keep it, so last month’s answers do not change. |
| What leaves your workbook | Whatever you paste or select, sent to that provider. | Only the columns you tick, your categories and your question, sent to Cloudflare and TypeSafe AI. Not to OpenAI. |
| Cost | Usually a Copilot license or a chat subscription. | Pay per row with credits that never expire. No subscription. |
Where a chat model is the better tool #
- You do not have a category list yet and want ideas for one. AI Categorizer cannot write text, so it cannot suggest categories.
- You want a summary, a rewrite, or a free-text answer about each row.
- You have twenty rows and a chat window already open.
Where AI Categorizer is the better tool #
- You already have the list, such as a chart of accounts or a topic list, and every row must end up in it.
- The rows are many and come back every month.
- A wrong category costs money or a wrong conclusion, so you want to see which rows to check.
It still makes mistakes: it can pick the wrong one of your categories. That is why every row has a confidence and a status. How accurate is it? shows where it gets it wrong, with two samples you can check yourself.