How do I analyse Excel and CSV data without burning AI tokens or credits?

What actually costs tokens?

Every character in your file. A model reads a spreadsheet by ingesting it — every cell becomes tokens — and every follow-up carries the file again. A long CSV costs by row count; a wide Excel sheet costs more, because each extra column is paid for on every row.

Every plan meters this: ChatGPT and Claude with rolling windows and weekly caps, Gemini with compute limits, Julius and Excel's Copilot with credits. Different names, same meter — the bigger the file, the more each question costs. A seemingly simple question can come with a shocking price tag.

How do you spend fewer tokens in a chat tool?

Send less of the file:

• Delete unused columns. Width costs the most — an ignored column is paid for on every row.
• Paste headers and a few rows instead of attaching the whole workbook.
• Ask for the method, not the answer. A formula the model writes once, you rerun for free.
• Start a fresh conversation per task, so old uploads are not carried along.

These stretch an allowance; they don't remove the meter. The expensive part is making the model read your data at all.

What if you want to keep your data private?

Paperswift removes that step. Your Excel or CSV file opens in your browser and is not uploaded. The model's only job is turning your question into a query, and all it sees is your column names and types — never the values in any row. The query runs on your machine; the answer is a chart with the logic readable underneath. More on the privacy page and in how it works.

A question sends the same short schema whether the file has ten rows or ten million, so questions are unlimited on every tier, including free.

What do you give up?

Whatever needs the model to see your rows. Paperswift counts, sums, groups, filters, and charts — but it can't summarise a comments column. For that kind of thinking, use an assistant or a notebook tool like Julius; the token bill is the price of that depth.

Which approach should you choose?

• Trim and upload when the model must see the content — free text to summarise, a messy export to untangle.
• Paperswift when the questions are structured — totals, trends, breakdowns — about an export you already have. Questions are free and unlimited; the paid tier adds taking results out (downloads, saved charts, exports) — see pricing.