Uploading an Excel file to an online chatbot is convenient, but it means sending your data to servers you do not control. Before you do, it helps to understand what happens to that data, reduce it to the minimum and, when needed, pick an alternative. Here is a practical method, current as of November 2026, that you can apply without technical skills.
What happens to your data when you upload it to an online service?
It depends on the service and the type of account. You can think of three levels, from riskiest to most protected:
- Public chatbots (personal account). The official Gemini help, for example, explains that a subset of chats may be read by human reviewers and that, with "Keep Activity" on, chats may be used to train the models. It also asks you not to enter confidential information you would not want a reviewer to see.
- Work tools under a business agreement. Microsoft states that prompts, responses and data accessed through Microsoft Graph are not used to train Copilot's foundation models. Google states that Workspace does not use customer data to train models without the customer's prior permission or instruction. Your company's rules and file permissions still apply.
- AI running locally on your computer. The model runs on your PC and the data does not pass through an outside service, as long as you really use a local model (see below).
What should you check before uploading a file?
- The account type: personal or work/school. The protections can be very different.
- Privacy settings: look in the service for the options on activity history and on using conversations to improve models, and read the official page before uploading anything.
- Your company policy: ask IT or the privacy contact which AI tools are approved. If there is a list, use only that.
- Sharing: Google warns that if a file is shared externally, content generated by Gemini inside it may be visible to people outside your domain.
How do you reduce the data before using it?
The principle is data minimization: the GDPR requires personal data to be adequate, relevant and limited to what is necessary for the purpose. Apply it to the sheet before sending it anywhere:
- Remove identifying data: names, tax IDs, emails, phone numbers, addresses, IBANs.
- Replace them with codes: "Customer 001" instead of the name. Keep the lookup table in a separate file that you never upload.
- Use only the columns you need for the question you want to ask.
- Work on an extract: a few dozen rows are enough to see whether a formula works.
Be careful: removing the name is not always enough. A combination of date, department and amount can still make a person identifiable. When in doubt, do not upload.
What alternatives are there to public chatbots?
| Option | Where the data is processed | What to check |
|---|---|---|
| Ollama or LM Studio (local AI) | On your computer, with a local model. LM Studio states that it can run completely offline. | With Ollama some models are "cloud" (name ending in :cloud) and run on Ollama's servers: choose a local model. |
| Copilot in Excel (Microsoft 365) | In the Microsoft service, under its business commitments. | Availability depends on the subscription and your organization's settings. |
| Gemini in Google Sheets | In the Google service, with the protections of your plan. | Plan type, file sharing and administrator rules. |
Local AI needs a capable computer: for the example model in Ollama's guide (about 7.2 GB), 8 GB of VRAM or unified memory on a Mac is recommended. A small model may make mistakes more often than a large one, so checking the results is still essential. To try it, after installing Ollama:
ollama run gemma4:e2bHow do you ask AI for help without sharing the real data?
The safest method is to describe the structure of the sheet instead of pasting the data:
- Open the sheet and note the headers, the type of each column (text, number, date) and the row range.
- Write one or two made-up examples, never real rows.
- Explain the goal in one sentence.
- Ask the AI for the formula or the steps, with an explanation of each part.
- Paste the formula into your sheet, locally, and test it.
I have an Excel sheet with these columns: A = Customer code (text, example C-001), B = Order date (date), C = Department (text), D = Amount in euros (number), E = Status (Paid or Unpaid). Headers are in row 1 and the data runs from row 2 to row 5000. I am not giving you the real data. I want the total of the "Unpaid" amounts for each department in a chosen month. Write me the formula, explain each part and tell me how to check it on 5 test rows.How do you check that the results are right?
- Test the formula on a few rows whose result you already know, calculated by hand.
- Compare with a second method: a pivot table or a total with SUMIF.
- Do not trust numbers given "off the top of its head": AI can get calculations wrong, so ask for the formula and let Excel run it.
What changes for people at work
In practice: use the tools your company has approved, reduce the data to the minimum and prefer describing the structure over uploading the file. This article is not legal advice: for real cases follow internal rules and ask whoever handles privacy in your organization. As a general reference there is Regulation (EU) 2016/679 (GDPR).
Frequently asked questions
If I only remove the names, can I upload the file?
Not always. Other fields can make a person identifiable. Assess the risk, cut the columns and, when in doubt, use an extract or just a description of the structure.
Is local AI always private?
A local model processes the data on your computer, but make sure you are not using a cloud model by mistake. With Ollama, models whose name ends in :cloud run on the service's servers.
Can I trust the AI's calculations?
No, always check them. Google's official pages warn that Gemini may give inaccurate information. Verify with formulas, pivot tables or a manual calculation on a sample.
Official sources
- Ollama: quickstart
- LM Studio: documentation
- Microsoft: data, privacy and security for Copilot
- Google: generative AI privacy in Workspace
- Google: Gemini Apps privacy help
- Regulation (EU) 2016/679 (GDPR) on EUR-Lex
This article was prepared with the help of artificial intelligence tools based on the official documentation, and is reviewed before publication.













