All guides

How to Protect Your Privacy When Using ChatGPT and Other AI

Protecting your privacy when you use ChatGPT or any other AI app comes down to three things: switch off training, clean up your shared links, and watch what you type in. Each one takes a couple of minutes, and this guide walks you through all three for five of the most popular AI apps.

I do all three in every AI app I use, before anything important goes in. It's the first thing I tell any friends, family or small business owner to do when they start using AI.

Step 1: Switch off training

Many AI apps can use your chats to train their future models unless you switch that setting off. On a free or personal account, it's usually switched on until you change it. This is the biggest privacy setting most people have never touched.

What "training on your chats" actually means

Model training: the company can use what you type (and what the AI says back) to help build the next version of its AI. Some of those chats can also be read by people the company employs to check quality.

That's a problem when your chats include a customer's name, a quote, your pricing or anything about your family. Turning training off means your new chats aren't used to build future models. You can still use the app exactly as before.

Before you start: what the switch does and doesn't do

The setting is useful, but it isn't a magic delete button. Worth knowing:

  • It only works going forward. Chats you've already had may have been used, and that can't be undone.
  • Your chats are still stored for a while. Most companies keep them for a short time to run the service and check for misuse.
  • Thumbs up or down can count as opting in. In some apps, rating a reply sends that conversation to the company even with training off. Skip the thumbs if it matters.
  • Menus move. These apps rename and shift settings often. If a label below doesn't match, search the app's settings for "training", "improve" or "privacy".

How to turn off training in ChatGPT

  1. Open your account menu (your name or picture).
  2. Go to Settings, then Data controls.
  3. Switch off Improve the model for everyone.

The setting follows your account, so doing it once covers the website and the phone app. Your chat history stays put. OpenAI's own steps are in its data controls help article.

Using OpenAI's dots? The same switch covers them. With Improve the model for everyone turned off, your dots' conversations, the actions they take and the apps you connect to them aren't used for training on a personal plan. OpenAI explains this in its dots privacy FAQ. Dots can also connect to your other apps, so only connect the ones you're happy for it to read.

How to turn off training in Google Gemini

  1. Go to myactivity.google.com/product/gemini while signed in.
  2. Find Keep Activity at the top and turn it off.

The trade-off: with Keep Activity off, Gemini stops saving your chat history too. Google still holds new chats for up to 72 hours to run the service, and chats that were already reviewed by people can be kept for up to three years. That's from Google's Gemini privacy help page. If you use Gemini through a work Google account, your admin may control this setting.

How to turn off training in Microsoft Copilot

  1. Sign in on copilot.com or in the Copilot app.
  2. Open your profile, then Privacy.
  3. Switch off both training toggles: the one for your conversations (text) and the one for voice.

There are two switches, so turning off only one leaves the other on. Microsoft says the opt-out covers future conversations only. Its current steps are on the Copilot privacy controls page. If you use Copilot with a work or school Microsoft 365 account, those chats aren't used for training in the first place.

How to turn off training in Claude

  1. Click your name at the bottom left, then Settings.
  2. Go to Privacy.
  3. Switch off Help improve Claude.

It applies to your whole account on every device. Incognito chats are never used for training, whatever this setting says. Anthropic explains it in its privacy centre.

How to turn off training in Grok

Grok has two places to check, because it lives inside X as well as in its own app.

  1. Grok app or grok.com: go to Settings, then Data Controls (called Data on the web), and switch off Improve the model.
  2. On X: go to Settings and privacy, then Privacy and safety, then the Grok data sharing section, and untick the box that allows your posts and Grok chats to be used for training.

xAI confirms in its consumer FAQ that once you opt out, new conversations aren't used to train its models.

What about Meta AI?

Meta AI is the assistant built into Facebook, Instagram, WhatsApp and Messenger. As of October 2026, I couldn't find a switch for Australian users that keeps Meta AI chats out of training. Meta has also said publicly that it uses Australians' public Facebook and Instagram posts to train its AI.

My rule: treat Meta AI like a public noticeboard. Ask it general questions, and keep anything about you, your customers, your money or your family out of it.

Step 2: Clean up your shared links

When you press Share on a chat, the app creates a public link that anyone with the link can read. In July 2026, some shared Claude chats turned up in Google search results. Chats that were never shared weren't exposed, but it's a good reminder to clean up old links.

  • ChatGPT: go to Settings, then Data controls, then Shared links, then Manage, and delete any you no longer need.
  • Claude: go to Settings, then Privacy, then Shared chats, and unshare anything old.

If you've shared chats from any other app, look in its settings for "shared" or "public links" and do the same. And before you share a chat in future, read it through as if a stranger will see it.

Step 3: Watch what you type in

Settings only go so far. What you type still matters. Here's what never goes into my AI, whatever the settings say:

  1. Bank details. No bank names, account numbers or logins. I use nicknames like "the home loan" instead.
  2. My kids' names. It's "my son" and "my daughter", and it still knows exactly who I mean.
  3. Where we live. No home address, no school name. "Home" and "school" are enough.
  4. Any password. Ever. When AI needs access to something, it gets a guest pass I can switch off any time.
  5. Anything that matters, until training is off. Which is what this guide is for.

Set up your AI properly from day one

The words privacy and AI can feel scary, I get it. But an AI system that's set up properly from day one, and knows your business, who you serve and how you write, is one of the fastest ways to create more time, money and freedom in your business.

If you're interested to see what an AI assistant can do for your business, book a free call here. And follow me on Instagram at @automatewithalicia, where I share helpful tips and show you how AI can reduce the admin load in your business today.

Settings checked on 24 October 2026. These apps change their menus often, so if something has moved, search the app's settings for "training" or "privacy".

Not ready to talk? The quiz shows you which admin job to hand over first.