Is It Safe to Put Personal Information into Gemini? | Human Review of Conversations, the Setting That Stops Training, and Masking Before You Send

Drafting Gmail replies, summarizing documents in Drive, tidying up spreadsheets. Gemini, which you can use straight from your Google account, is an AI you reach for as a natural extension of everyday work. That also means you can end up pasting text that contains customers' names and contact details without noticing.

This article is for people who use Gemini at work, and for people in IT and general affairs who set the company's rules for using generative AI. We go through, in order, how the information you enter is handled, when people review conversations, how to stop training, and how to mask information when you really need to use it.

On how Gemini handles data, we only describe what Google's official help said as of November 2, 2026. Terms and setting names can change, so check the latest official pages before you use it.

Bottom line: don't enter it on a personal account, and if you must, mask it first

  • On a personal Google account, Keep Activity is on by default (for users 18 and over). While it's on, conversations are used to improve services, including training generative AI models
  • To improve the service, human reviewers read some conversations. Reviewed conversations are kept for up to three years and are not deleted even if you delete your activity
  • If you turn Keep Activity off, conversations from then on are not used for training. However, conversations are kept in your account for 72 hours
  • With work or school accounts (eligible Google Workspace editions), people do not review your conversations or uploaded files, and they are not used for training without permission
  • In practice, the most reliable approach is to mask names, contact details, and ID numbers before sending

How Gemini handles the information you enter

Personal Google accounts

According to Google's Gemini Apps Privacy Hub, when Keep Activity is on, Google uses activity data such as your conversations to provide, develop, and improve its services (including training generative AI models). This setting is on by default for accounts of users 18 and over.

Google itself advises: "Please don't enter confidential information that you wouldn't want a reviewer to see or Google to use to improve our products, services, and machine-learning technologies."

Human reviewers may read your conversations

The thing to know in particular about Gemini is that human reviewers read some conversations to improve Google's services. Reviewers include trained service providers working for Google.

  • Conversations are disconnected from your account before they reach reviewers
  • Reviewed conversations are kept for up to three years
  • Reviewed conversations and related data are not deleted even if you later delete your activity

Even when disconnected from your account, if a name or phone number is written in the body of the conversation, a person will see it. It may no longer be clear whose conversation it is, but the personal information written in it stays as is. You need to decide whether to enter it on the assumption that it can't be deleted later.

Work and school accounts (Google Workspace)

When you use Gemini with a work or school Google account, different data processing terms may apply. According to the Generative AI in Google Workspace Privacy Hub, when you use the Gemini app on an eligible edition, human reviewers do not read your chats or uploaded files, and they are not used to train generative AI models without your permission. Google also states that your interactions with Gemini are not disclosed outside your organization.

If your company uses Gemini, using it with a work account is the baseline. If you are also signed in to a personal account in the same browser, it's easy to mix up which account you have Gemini open in. Check the account shown on the screen before you use it.

That said, this too is a promise about "not training" and "no human review." What you enter is processed on Google's servers, so you are still handing personal data to an outside service.

The setting that stops training: turn off Keep Activity

If you use a personal account, first check the following setting (in a browser on a computer).

  1. Open gemini.google.com
  2. Open Activity from Settings & help
  3. Click "On" at the top, and choose "Turn off" or "Turn off and delete activity"

Once it's off, conversations from then on don't appear in your activity and are not used to train AI models (unless you send feedback yourself). However, conversations are kept in your account for 72 hours to provide the service and process feedback. Also, while it's off, you can't use connections with other apps such as Google Workspace.

If you leave it on, you can choose an auto-delete period for your activity. The default is 18 months, and you can change it to 3 months, 36 months, or never auto-delete. However, reviewed conversations are kept for up to three years regardless of this period.

Temporary chat

If you want to leave the setting as is and just not keep a particular conversation, you can use a temporary chat. Temporary chats are not used to train AI models, and human reviewers don't read them. However, they are kept in your account for 72 hours.

The Personal Information Protection Commission's alert

On June 2, 2023, Japan's Personal Information Protection Commission issued an alert on the use of generative AI services. It asks that if you enter prompts containing personal information, you thoroughly confirm whether doing so is within the scope of the purpose of use, and whether the provider uses that data for machine learning.

The same thinking applies to Gemini: on a personal account with Keep Activity left on, what you enter is used for training. The alert is covered in detail in Is It Safe to Put Personal Information into ChatGPT?.

Situations in Gemini that tend to lead to data leaks

  • Pasting the contents of Gmail or Drive in bulk: email threads and tables in shared drives are full of names and contact details that even the person pasting them doesn't notice. The more you paste at once, the more you miss
  • Handling work documents on a personal account: on a personal account, conversations are subject to training and human review. Handle work documents on your work account
  • Attaching files or images: photos of business cards and invoice PDFs also show names, addresses, and bank account numbers

The categories of information you shouldn't enter (information that identifies a person such as names, My Number IDs and card numbers, sensitive personal information, passwords and API keys) are the same as for ChatGPT. How to treat each category is summarized in the table in the ChatGPT article.

Real leak incidents with generative AI and their causes are collected in Generative AI Data Leak Cases.

What to decide as a company

  • Use Gemini with work accounts: don't handle work information on personal Google accounts
  • If personal accounts remain, align the settings: decide whether to turn Keep Activity off or use temporary chats, and have everyone check
  • Send things in a form you wouldn't mind a person reading: reviewed conversations can't be deleted later. Set a procedure for masking before sending

How to mask personal information before sending

Masking works the same in ChatGPT and Gemini. Replace names with "Person A" and email addresses with "(email)" before pasting, or ask in a masked form such as "a thank-you email to [Name]," and fill in the final details on your end.

With many entries, replacing by hand leads to misses. A browser extension that checks the content the moment you paste helps reduce oversights. When adopting one, check who provides it and whether it sends the content outside for inspection.

Checking before sending, inside the browser: Samurai App Data Guard

Samurai App's Chrome extension, "Samurai App Data Guard," checks the content before it is sent when you paste text or attach a file in Gemini. It checks the same way on other generative AI sites such as ChatGPT, Claude, and Copilot.

  • What it finds: names, email addresses, phone numbers, addresses, dates of birth, My Number IDs, credit card numbers, bank account numbers, API keys and tokens, and passwords. It finds names using clues such as table headers and labels like "Name:"
  • What you can choose: cancel, send as is, send partly masked, or send masked. When masked, each match is replaced with its category name, such as [NAME] or [EMAIL]
  • The check stays in the browser: it uses no AI model and checks with fixed rules. It never sends the content outside for inspection

If you mask before sending, no personal information remains in conversations that human reviewers read or that are used for training. The flow from pasting a customer list to sending it masked is shown in the video on the PII detection page (45 seconds, demonstrated in ChatGPT). For details, see the PII detection page.

Sources

Related featurePII detection and masking before data reaches AIPII detection finds email addresses, phone numbers, My Number IDs, API keys and more in the files and text you send to AI, right in the browser, and lets you send a masked copy. No model is used and nothing is sent out for the check.

More in this category

Back to AI and security articles

Everyday work comes together, connects, and becomes the company's strength.

Contact usOpen Samurai App

See the demoContact us