Millions of people now type personal questions into AI chatbots every day — medical symptoms, work documents, business plans. Yet most users have never opened the privacy settings of the chatbot they use. This practical guide to AI chatbot privacy answers the key question: how private are these conversations, really? The short answer is that your chats are not private by default. Depending on the chatbot and your settings, conversations may be stored, reviewed, and used to improve AI models. The good news is that ChatGPT, Google’s Gemini and Anthropic’s Claude each offer settings that reduce what is collected and kept. This guide explains what each one collects, walks through the settings that matter, compares temporary chat modes, and ends with a checklist of eight privacy checks you can run today. Company privacy policies change over time, so treat this as general information rather than legal advice, and always check the official documentation for the latest details.
What AI chatbots collect about you
Before changing any setting, it helps to know what each chatbot gathers. The three major services publish official documentation, and the picture differs meaningfully between them.
ChatGPT (OpenAI). On personal workspaces — Free, Plus and Pro plans — “Improve the model for everyone” is switched on by default, meaning new conversations can be used to train future models. Switching it off keeps your chat history but stops new conversations being used for training. ChatGPT also has a memory feature that remembers details across conversations — a separate setting, so changing one does not change the other.
Gemini (Google). For users aged 18 and older, Gemini Apps Activity is on by default and stored for up to 18 months. Gemini also collects location data continuously — your general area from IP addresses and saved home or work addresses, plus precise location with permission. Google states that it does not train generative AI models directly on your Gmail, Drive, Contacts, Calendar, Wallet or payments, or Photos media, though summaries generated during a conversation may be used to answer your prompts.
Claude (Anthropic). Anthropic collects identity and contact information, payment information, and your Inputs and Outputs — the prompts, files and images you send, and the model’s responses. The policy notes that personal data included in your Inputs may be reproduced in Outputs.
The differences at a glance:
| ChatGPT | Gemini | Claude | |
|---|---|---|---|
| Training use default | On by default for personal workspaces; can be switched off | Future chats not used to improve models when Keep Activity is off and no feedback is submitted | Uses Inputs/Outputs per privacy policy; users can make privacy rights requests |
| Activity/history retention | Chat history kept unless deleted or account removed | Activity stored up to 18 months by default; adjustable to 3 or 36 months | Inputs and Outputs collected under the consumer privacy policy |
| Notable extra collection | Cross-conversation memory (separate setting) | Continuous location data collection | Identity, contact and payment information |
| Account deletion | Available from settings on personal workspaces; permanent | Managed through Google account controls | Privacy rights requests via privacy@anthropic.com |
ChatGPT: the settings to check
OpenAI’s data controls live under Settings, then Data controls, as explained in the official Help Center article. There are four areas worth reviewing.
1. The training switch. “Improve the model for everyone” is enabled by default on personal workspaces. Switch it off if you do not want your conversations used for training. Your chat history stays available — only the training use stops for new conversations.
2. Memory. Memory lets ChatGPT remember details across chats. It is a separate control from the training switch, so turning off training does not turn off memory. Review the memory settings separately and delete stored memories you no longer want kept.
3. Temporary chats. When you need extra discretion for a single conversation, start a temporary chat. Temporary chats are not saved to your history, do not create new memories, and are not used for training. One caveat: the company says copies may be retained for up to 30 days for safety purposes, according to secondary privacy reviews of the official documentation — so treat temporary chats as short-term, not zero-trace.
4. Account deletion. Signed-in users on Free, Go, Plus and Pro plans can delete their account from settings, and the deletion is permanent. Self-service deletion is not available on managed workspaces, such as those run by an employer or school, where the organisation controls the account.
Gemini: the settings to check
Google explains Gemini’s data handling in the Gemini Apps Privacy Hub. Four settings matter most.
1. Gemini Apps Activity. Activity collection is on by default for users 18 and older. You can turn it off, and you can change how long activity is stored: the default is 18 months, adjustable to 3 or 36 months. If you have used Gemini for a long time with defaults on, consider shortening the retention period and deleting old activity.
2. Keep Activity and feedback. If Keep Activity is off and you submit no feedback, Google says your future chats are not used to improve its models. This is the key combination for reducing training use on Gemini.
3. Temporary chats. Gemini’s temporary chats work differently from ChatGPT’s. They are kept with your account for 72 hours and are not used to train Google’s AI models. That is a defined, short window — useful, but still a retention period rather than instant deletion.
4. Location. Gemini always collects location data — your general area from your IP address and saved home or work addresses, and precise location when you grant permission. Review which location access you have allowed and whether the chatbot genuinely needs it.
One reassuring detail from the Privacy Hub: Google does not train generative AI models directly on your Gmail, Drive, Contacts, Calendar, Wallet or payments, or Photos media. However, summaries or excerpts generated during a conversation may be used to answer your prompts — so quoting a private email into a chat still puts that text into the conversation.
Claude: the settings to check
Anthropic publishes its practices in its privacy policy. Claude’s documentation is more policy-focused than settings-focused, but three points matter.
1. Know what counts as your data. Anthropic collects identity and contact data, payment information, and your Inputs and Outputs — your prompts, uploaded files and images, and the model’s responses. Assume everything you type is stored as part of that record.
2. Watch what you include. The policy explicitly states that personal data included in Inputs may be reproduced in Outputs. That means sensitive details you paste in can resurface in the model’s replies within the conversation. Keep confidential material out of prompts.
3. Exercise your privacy rights. Anthropic provides a privacy rights request channel at privacy@anthropic.com. The policy also notes that it does not knowingly collect data from people under 18, and that commercial or enterprise customer data is governed by customer agreements rather than the consumer policy — so workplace Claude accounts follow different rules from personal ones.
Temporary chats, compared
Both ChatGPT and Gemini offer a temporary chat mode, but they are not identical:
- ChatGPT: not saved to history, creates no memories, not used for training. Copies may be retained for up to 30 days for safety purposes (per secondary privacy reviews citing the official docs).
- Gemini: retained with your account for 72 hours, and not used to train Google’s AI models.
Gemini’s window is shorter and precisely defined; ChatGPT’s covers more behaviour but carries the longer safety-retention caveat. Neither mode guarantees instant deletion — use them for individual sensitive conversations, not for truly confidential material.
What you should never paste into an AI chatbot
Settings help, but habits help more. A useful rule of thumb: if you would hesitate to publish it on a website, do not paste it into a chatbot.
- Passwords, PINs and access codes. These can be stored, reviewed, or reproduced in conversation history. Use a password manager instead.
- Customer or employee personal data. Names, addresses, ID numbers and contact details belonging to other people are not yours to share — and may breach data-protection rules where you work.
- Medical and financial details. Diagnoses, prescriptions, account numbers and tax information deserve systems with defined retention and access controls, not a general chatbot.
- Unreleased work. Draft manuscripts, business plans, product designs and source code can be retained or reproduced in outputs. Treat a chatbot like a public forum for anything commercially sensitive.
- Someone else’s private communications. Forwarding another person’s emails or messages into a chat shares their data without their consent.
Eight AI chatbot privacy settings to check today
This checklist brings the guide together into actions you can complete in one sitting:
- Open ChatGPT’s Settings, then Data controls, and decide whether “Improve the model for everyone” should stay on.
- In the same area, review ChatGPT’s memory settings separately — remember that memory and training are independent controls.
- Check your Gemini Apps Activity setting and retention period: consider switching from the 18-month default to 3 months.
- On Gemini, confirm Keep Activity is off if you do not want future chats used to improve models.
- Learn where the temporary chat option lives in each chatbot you use, so it is available when a sensitive question comes up.
- Review the location permissions you have granted to Gemini and remove any that are not needed.
- Re-check each chatbot’s privacy policy or privacy hub yearly — policies change.
- Decide whether your most sensitive use case belongs on a chatbot at all, or on a model running on your own computer (see below).
The most private option of all: run the model yourself
Every setting above reduces collection, but every cloud chatbot still sends your prompts to someone else’s servers. If you want conversations that never leave your machine, the strongest option is running an AI model locally on your own computer: no account, no provider-held activity history, no training pipeline receiving your text.
Our guide How to Run AI Models on Your Computer with Ollama: A Beginner’s Guide walks through the setup step by step. Local models are less powerful than the largest cloud systems, but for drafting, brainstorming and private documents they keep everything in-house. For context on what today’s large models do with data, see also Google’s Gemini 4 Carbon: Everything We Know About the New AI Model.
A quick caution
Privacy settings, retention periods and policies change — sometimes without much notice. The details above reflect the official documentation at the time of research, but each company’s current policy always takes precedence. This guide is general information for everyday users, not legal advice. If you handle other people’s data at work, check your employer’s policies and the applicable data-protection rules before choosing settings.
Summary
AI chatbots are not private by default: ChatGPT may use new conversations for training unless you switch it off, Gemini stores activity for up to 18 months alongside location data, and Claude records your inputs and outputs under its privacy policy. The settings that matter most are the training and activity toggles, memory controls, temporary chat modes and location permissions — plus the habit of never pasting passwords, personal data, medical details or unreleased work into a prompt. Work through the eight checks above, and for the strongest privacy of all, consider running a model locally on your own computer.
Do AI chatbots use my chats to train their models?
In ChatGPT, “Improve the model for everyone” is on by default for personal workspaces, so new conversations can be used for training unless you switch it off. In Gemini, future chats are not used to improve models when Keep Activity is off and no feedback is submitted. Temporary chats in both services are not used for training.
Does deleting a chat permanently delete my data?
Not necessarily. Deleting a chat removes it from your visible history, but companies may retain copies for limited periods — Gemini keeps temporary chats for 72 hours, and secondary privacy reviews report ChatGPT may keep copies of temporary chats for up to 30 days for safety. Deleting your entire account is stronger: on ChatGPT’s personal plans it can be done from settings and is permanent. When in doubt, check each company’s current privacy documentation.
What should I never paste into an AI chatbot?
Never paste passwords or access codes, other people’s personal data, medical or financial details, unreleased work such as manuscripts or business plans, or someone else’s private messages. Anthropic’s policy notes that personal data in your inputs may be reproduced in the model’s outputs, and once text leaves your device you cannot fully control where it is stored. If you would hesitate to publish something on a website, keep it out of chatbot prompts.
Can I use an AI chatbot without creating an account?
Some services allow limited use without signing in, but the privacy settings described in this guide — training switches, activity retention, memory controls — generally require an account to manage. The most private way to use AI without an account is to run a model locally on your own computer, where no provider account is involved and your prompts never leave your machine. Our Ollama guide explains how to set that up.
Ahesanur Asun is the Founder and Editor of UptoFeed, where he publishes updates on
AI news, artificial intelligence tools, and emerging technology. He focuses on explaining complex AI developments in a clear and practical way so readers can quickly understand and use the latest innovations.
