Before you upload a Word, Excel or PowerPoint file to a chat assistant
Last updated: 19 October 2026
Short answer: when you attach a .docx, .xlsx or .pptx to ChatGPT, Gemini, Copilot or Claude, the service receives the file, not the page you see. That includes comments and their authors, tracked changes with the deleted text, the author and company in the properties, and the paths of linked files. Remove what the assistant does not need before you upload, and check the service's training setting, which is a separate question from whether it keeps your chats.
This page is about what goes with a file and where the services describe their settings. What each kind of hidden data is, and where it lives in the file, is in hidden data types in Word, Excel and PowerPoint.
What goes with the file
A .docx, .xlsx or .pptx is a ZIP package of parts. Uploading it sends every part, including the ones Office does not show on the page:
- Comments and their authors, including resolved ones (comments, ink and annotations).
- Tracked changes: deleted text stays in the file until the change is accepted or rejected, with the name of whoever made it (accept or reject tracked changes).
- Properties: author, last modified by, company and manager (the author's name in Word).
- Links to other files: an Excel link or a linked object records where the source file is, which can include a user name or a share name (Excel links and full paths).
- Hidden sheets, rows and slides, and speaker notes (hidden sheets, notes and hidden slides).
Which of these parts a service reads when it answers is up to the service. We found no vendor page that lists which parts of an Office file it reads, and that can change. Assume the whole file was received.
Where the services describe training
These are the vendors' own pages, read on 19 October 2026. Setting names and menus change; look at the setting in your own account before you upload. Several pages do not say what the setting is when an account is new, so we do not state defaults for personal accounts.
| Service | Personal accounts | Work accounts |
|---|---|---|
| ChatGPT | Settings › Data controls › Improve the model for everyone. OpenAI's Data controls in ChatGPT: "When Improve the model for everyone is off, your new conversations won't be used to train OpenAI models. They can still appear in your chat history." The same page adds that if you give feedback, such as a thumbs up or down, "the entire conversation associated with that feedback may be used to train OpenAI models". A temporary chat stays out of your history unless you save it, and OpenAI says it may keep a copy for no more than 30 days. | The same page: "By default, OpenAI does not use content from ChatGPT Business, Enterprise, Edu, or ChatGPT for Healthcare workspaces to train its models." |
| Gemini | The Keep Activity setting. Google's Gemini Apps Privacy Hub says that when it is on, your chats and "what you share with Gemini (like files, videos, screens, and photos)" are saved in your Activity, that "a subset of chats are reviewed by human reviewers", and: "Please don't enter confidential information that you wouldn't want a reviewer to see or Google to use to improve our services, including machine-learning technologies." | Google's privacy hub for Gemini in Workspace says your content is not reviewed by people or used for model training outside your domain without permission. |
| Copilot | Microsoft's Copilot privacy controls describe Privacy › Training on conversation activity (and Training on voice conversations), say opting out covers future conversations, and say the setting "will not exclude your conversations from being used for other general product or system improvements". That page applies only to the older Copilot app. Microsoft's page for the updated app, available since 18 August 2026, does not describe a training setting, so look under Privacy in your own Copilot settings. | Microsoft's data, privacy and security page for work accounts: "Prompts, responses, and data accessed through Microsoft Graph aren't used to train foundation LLMs, including those used by Microsoft Copilot." |
| Claude | Model Improvement in Privacy Settings. Anthropic's Is my data used for model training?: "Your Incognito chats are not used to improve Claude, even if you have enabled Model Improvement in your Privacy Settings." | Anthropic's page for commercial customers: "By default, we will not use your inputs or outputs from our commercial products", naming Claude for Work and the Anthropic API. |
Training is one use among several. The same pages describe keeping chats in your history, review by people, and use to protect the service, and a training setting does not switch those off. If a file should not leave your organisation at all, a setting is not the answer; do not upload it.
Cleaning the file first, in the browser
- Work on a copy. Keep the original for your own records.
- Ask what the assistant needs. If it needs one table or two paragraphs, paste those, or save a copy with only that sheet or section. Less in the file means less to clean.
- See what is in the copy. Drop it into ScrubDrop, which takes .docx, .xlsx or .pptx files. It reads the file in your browser; the file is not uploaded. The list shows each kind of hidden data it found, with counts and short samples.
- Remove what goes. The free version takes one file at a time and removes properties, comments and tracked changes; tracked changes are accepted, so read the result if some changes should not stand. ScrubDrop Pro also removes links to other files, hidden sheets and slides, speaker notes, embedded objects and the other structural kinds, and can reject tracked changes instead.
- Let it re-check. ScrubDrop reads the cleaned file again with separate parsers and offers it only if nothing of the chosen categories remains and the text you can read matches the original, after the tracked changes are accepted or rejected as you chose.
- Open the cleaned copy, then check the service's training setting, then upload.
What cleaning does not change
- What is on the page. If the text names a client, the assistant receives the name. Cleaning removes what is out of sight, not what you can read.
- White or tiny text, and cells outside the print area. ScrubDrop lists them but does not remove them, in either version.
- The file name. The service sees the name you upload; ScrubDrop adds "-scrubbed" to it and keeps the rest.
- What the service does next. Retention, review and training are set by the service and your account, as above.
Questions
Will a chat assistant see the comments and tracked changes in my file?
They are in the file it receives. Which parts of a .docx, .xlsx or .pptx a service reads is up to the service, and we found no vendor page that lists them. Remove what the assistant does not need before you upload.
If I turn off training, is my file not kept?
Not necessarily. The vendors describe training and keeping chats as separate things. OpenAI says that with Improve the model for everyone off, new conversations are not used for training but can still appear in your chat history, and that giving feedback on a response can make that conversation available for training; Google says that even with Keep Activity off it still uses your chats to respond to you and to protect its users.
Does ScrubDrop send my file anywhere?
No. ScrubDrop reads and cleans the file in your browser, and the file is not uploaded. What you then upload to a chat assistant is your choice.
Sources: the OpenAI, Google, Microsoft and Anthropic pages linked on this page were read on 19 October 2026. ScrubDrop is not affiliated with any of them.
See what your file carries before you upload it, in your browser: one file at a time, the supported kinds of hidden data it finds are listed, and properties, comments and tracked changes are removed. Check a file.