How to Use Chat Settings for Better Privacy: Claude, ChatGPT, Gemini, Perplexity, Copilot
GPTPrompts.AI Editorial
Tested privacy settings on each tool's current 2026 interface. Cross-referenced with each company's official privacy policy. Verified May 2026. · Last updated May 16, 2026
The honest 2026 guide to AI chat privacy. The exact settings to change on each tool, what they do, and a 10-step audit you can finish in 20 minutes.
The direct answer
Turn off model training, set retention to 3 months, use Temporary Chats for anything sensitive.
Three moves cover 80% of the privacy benefit on every major AI chat tool: (1) turn off model training on ChatGPT, Perplexity, and Copilot, (2) set Gemini Apps Activity to 3-month auto-delete (default is 18 months), (3) use Temporary Chats (ChatGPT) or Incognito mode (Perplexity) for anything you would not paste into a Google search box. The 10-step audit at the end of this page walks through every setting in 20 minutes.
Affiliate link, we may earn a commission at no extra cost to you.
How we tested this
How we tested AI chat privacy settings
We tested every setting in this guide on the current 2026 interface of each tool, signed in across consumer, Pro, and enterprise accounts where applicable. We cross-referenced behavior with each company's published privacy policy and Trust Center.
Privacy settings change. Companies frequently rename menus, move toggles, and update default behaviors. We capture the location and intent of each setting; the exact path may shift slightly. Always verify against the current interface and the company's privacy documentation if anything looks different.
We deliberately focused on settings that meaningfully change data behavior. We did not include cosmetic options or marketing-only settings. Every step in the 10-step audit changes what data leaves your account and how long it persists.
Section 1
Privacy settings tool by tool
Each tool has different defaults and different settings to change. Walk through your most-used tools first; the rest can wait for your next quarterly audit.
ChatGPT
Default behavior
On the free tier, OpenAI may use your conversations to improve their models unless you opt out. ChatGPT Plus, Team, and Enterprise tiers have stronger defaults. Team and Enterprise do not train on customer data at all. Memory is on by default on Plus and stores facts across conversations.
Key settings to change
Improve the model for everyone (turn OFF)
Stops OpenAI from using your prompts and responses for training new models. Set once and applies to future conversations.
Where: Settings → Data Controls → Improve the model for everyone
Chat history and training (review)
Controls whether your chat history is saved at all. If you turn history off, new chats are not saved in your account and not used for training.
Where: Settings → Data Controls → Chat history
Temporary Chats (use for sensitive topics)
Conversations marked Temporary do not appear in history, are not used for training, and disappear after 30 days. Use for anything you would not paste into a search bar.
Where: Click the chat title at the top, toggle Temporary
Memory (turn OFF if you want zero persistence)
Memory lets ChatGPT remember facts about you across conversations. Useful for personalization, risky if you share a device or want clean separation.
Where: Settings → Personalization → Memory
Custom Instructions (review what you have shared)
Custom Instructions are persistent context you typed in once and apply to every conversation. Audit annually; people forget what they wrote 6 months ago.
Where: Settings → Personalization → Custom Instructions
Export your data
Download a complete export of your conversations, custom GPTs, and account data for review or backup. Useful before deleting an account.
Where: Settings → Data Controls → Export Data
Red flags to watch
- If you use a free or Plus account at work for company-confidential information, you are likely violating your employer's policy. Switch to ChatGPT Team or Enterprise.
- Memory entries persist until you delete them. Review every 90 days.
Best for
Individuals who use ChatGPT daily. Turn OFF model training, USE Temporary Chats for anything sensitive.
Claude
Default behavior
Anthropic does NOT use Pro, Team, or Enterprise customer conversations for training by default. Free tier policy varies and has changed over time. Anthropic explicitly states inputs and outputs from paid tiers are not used to train models unless you opt in.
Key settings to change
Privacy & data controls (review)
Verify whether your conversations are being used for model training. Pro and above default to no training; free tier requires checking the current policy.
Where: Settings → Privacy or Account
Delete individual conversations
You can delete individual chats from your history. Deleted conversations are removed from your account view, though backend retention may follow legal hold periods.
Where: Hover over a chat in the sidebar and click delete
Project privacy
If you use Projects (Claude Pro feature) to upload files or share context with the model, those files are accessible to Claude within that project. Audit which files you uploaded.
Where: Project settings → Files
Shared chat links (review)
If you have created shareable chat links, those URLs make the chat publicly viewable by anyone with the link. Audit your shared chats list.
Where: Settings → Shared Links or via the share icon at the top of any chat
Delete account (final option)
Deletes all your Claude account data. Subject to backend retention for legal compliance reasons.
Where: Settings → Account → Delete Account
Red flags to watch
- Shared chat links are public. Anyone with the URL can view the conversation. Never share a chat that contains private information.
- Project files persist in the project until you remove them. Even if you delete a conversation that referenced the file, the file itself remains.
Best for
Writers, researchers, and developers who want strong defaults out of the box. Audit Shared Links and Project Files.
Gemini
Default behavior
Gemini Apps Activity is ON by default and saves your conversations to your Google Account for up to 18 months unless changed. Saved activity may be reviewed by humans and used to improve Google products. You can turn this off completely.
Key settings to change
Gemini Apps Activity (set to off or auto-delete)
Controls whether Gemini saves your conversations to your Google Account. When off, conversations are not saved and cannot be reviewed by humans.
Where: myactivity.google.com/product/gemini → Gemini Apps Activity
Auto-delete activity (3, 18, or 36 months)
If you keep activity on, set auto-delete to 3 months so old conversations are cleared automatically. The default is 18 months, which is longer than most people need.
Where: myactivity.google.com/product/gemini → Auto-delete
Workspace data sharing (review for Google AI Pro / Ultra)
When you use Gemini inside Workspace apps (Docs, Gmail, Drive), Google applies stricter data handling for paid Workspace tiers. Verify your business or consumer status affects the right tier.
Where: Google Workspace admin console or account settings
Delete specific Gemini conversations
You can delete individual conversations from your activity log. Deletion is generally final after a short retention window.
Where: myactivity.google.com/product/gemini → click on conversation → delete
Web & App Activity (broader Google account)
Gemini activity is part of your broader Google account activity. Audit Web & App Activity separately if you use Gemini inside Search.
Where: myaccount.google.com/data-and-privacy
Red flags to watch
- Gemini Apps Activity ON means humans may review your conversations as part of Google's product improvement process. Anything sensitive should go through a paid Workspace tier with stricter handling.
- Default 18-month retention is much longer than most users want. Set to 3 months unless you have a reason.
Best for
Heavy Google Workspace users. Use the Workspace integration for business; lower the personal retention window to 3 months.
Perplexity
Default behavior
Perplexity by default may use your queries to improve their service and AI models. Pro accounts have stronger controls. Conversations and AI Profile data persist until you delete them. Incognito mode is available for one-off private searches.
Key settings to change
AI Data Retention (turn off training)
Controls whether Perplexity uses your conversations to train AI models. Available in the privacy settings on most accounts.
Where: Settings → Privacy or AI Data Retention
AI Profile (audit personal context)
Perplexity builds a profile of your interests and preferences to personalize answers. Review what is stored, and clear it if you do not want personalization.
Where: Settings → AI Profile or Personalization
Incognito search (per-query privacy)
Searches in Incognito mode are not saved to your history and not used for personalization. Use for one-off sensitive queries.
Where: Toggle at the search bar or via the menu
Spaces (review what is shared)
If you have created Spaces (Perplexity's project workspaces), conversations and uploaded files within a Space are accessible to anyone you invited. Audit collaborators on every Space.
Where: Spaces → individual Space settings
Delete account
Deletes your Perplexity account and associated history. Subject to backend retention for legal compliance.
Where: Settings → Account → Delete Account
Red flags to watch
- AI Profile builds a detailed model of your interests over time. If you share an account or use Perplexity at work, audit and clear regularly.
- Spaces are collaborative by design. Anyone you invite has access to the full conversation history in that Space.
Best for
Researchers and analysts. Use Incognito for sensitive queries; review your AI Profile every 90 days.
Copilot
Default behavior
Microsoft Copilot has different privacy posture by tier. Free consumer Copilot may use your prompts for improvement and personalization. Copilot Pro and especially Microsoft 365 Copilot (business and enterprise) have much stronger defaults including Commercial Data Protection that prevents your prompts from being used for training.
Key settings to change
Commercial Data Protection (verify enabled)
Available with a signed-in Microsoft Entra (Azure AD) work or school account. When enabled, your conversations are not used to train AI models. Look for the green shield icon in the Copilot interface.
Where: Sign in with your work/school account; verify shield icon
Chat history (manage)
Controls whether your Copilot conversations are saved across sessions. Saved history personalizes results but expands the data footprint.
Where: Copilot Settings → Chat History
Personalization data (review)
Microsoft Copilot can use your account context (email, calendar, files in M365) to personalize responses. Audit which apps are connected.
Where: Settings → Personalization or Connected Apps
Microsoft Privacy Dashboard
Centralized view of all data Microsoft has collected via your account, including Copilot interactions. Use this to delete activity in bulk.
Where: account.microsoft.com/privacy
Work vs personal account separation
Keep your work and personal Microsoft accounts strictly separate. Work Copilot follows your employer's data policies; personal Copilot follows consumer terms. Mixing them is the easiest way to create compliance issues.
Where: Sign in to Copilot with the correct account; verify in the top-right
Red flags to watch
- Free consumer Copilot has weaker defaults than Pro or Microsoft 365 Copilot. If you handle work data, use the work account version with Commercial Data Protection.
- Connected apps (calendar, email, OneDrive) expand the privacy surface significantly. Audit them quarterly.
Best for
Microsoft 365 users in business or school environments. Always verify the green shield (Commercial Data Protection) before sharing anything sensitive.
Section 2
The 20-minute privacy audit (10 steps)
Run through this once now, then set a quarterly calendar reminder. Total time: 15 to 20 minutes if you have accounts on all 5 tools.
Open every AI chat tool you use and check which account you are signed in with
Work email, personal email, family Google account, employer Microsoft account. Confusion here causes 80% of accidental data leaks.
Turn OFF model training on ChatGPT, Perplexity, and Copilot
Default behavior on free tiers is to use your prompts for training. Three clicks each to turn off. Persistent across future conversations.
Set Gemini Apps Activity to OFF or 3-month auto-delete
Default 18 months is longer than almost anyone needs. Visit myactivity.google.com/product/gemini.
Review ChatGPT Memory and Custom Instructions
Settings → Personalization. Delete anything sensitive, professional, or out of date. Most users have stale entries from 6+ months ago.
Audit Claude shared chat links
Anyone with a shared link can view that conversation. Settings → Shared Links. Revoke any that contain private information.
Audit Perplexity AI Profile and Spaces
Clear personalization data if you share devices or use Perplexity at work. Audit collaborators on every Space.
Verify Copilot Commercial Data Protection at work
Green shield icon must be visible in your Copilot interface when using work data. If not, you may be sending data to consumer Copilot.
Use Temporary or Incognito chats for sensitive topics
ChatGPT Temporary Chats, Perplexity Incognito mode. Anything you would not paste into a Google search bar should use these modes.
Set a quarterly calendar reminder to re-audit
Privacy settings drift. New features get added with default-on settings. Companies update their data policies. 4 audits per year keeps the drift manageable.
Read your employer's AI policy if you use AI at work
Many employers have policies you may not be aware of. Some prohibit pasting any customer data into AI chat tools regardless of tier. Verify before assuming you are compliant.
The fastest way to keep this fresh
Set a recurring calendar event for the first of January, April, July, and October. Title: "AI privacy audit, 20 minutes." Each session, run through the 10 steps above. Privacy settings drift, new features get added with default-on settings, and companies update their data policies. Quarterly audits keep the drift manageable.
The full audit takes about 15 to 20 minutes once you know the menus. Most of the value comes from steps 2, 3, and 7 (model training off, Gemini retention to 3 months, verify Copilot Commercial Data Protection at work).
How we'd actually handle AI chat privacy today
Honest opinion from running daily audits on a real account set.
The realistic threat model for most individuals using AI chat tools is not government surveillance or corporate espionage. It is the accidental over-share: pasting a draft message that mentions a colleague, asking a medical question on a shared family Google account, or feeding a confidential contract into a free-tier tool while working from a coffee shop. The defenses for that threat model are simple and free: turn off model training, use Temporary Chats for anything sensitive, separate work and personal accounts cleanly.
The two settings that almost nobody changes but absolutely should: ChatGPT's "Improve the model for everyone" toggle and Gemini Apps Activity retention. The first one keeps your conversations out of OpenAI's training corpus. The second one shortens the default 18-month conversation retention to 3 months. Both are off-by-one-toggle. Both should be the first changes you make.
For employer use cases, the single biggest mistake is using free or consumer-tier AI for work data. ChatGPT free, free Copilot, free Perplexity, and Gemini on a personal Google account all have weaker privacy posture than the paid or enterprise tiers. If you handle anything that includes customer names, internal documents, or trade secrets, you should be on ChatGPT Team or Enterprise, Claude Team or Enterprise, Microsoft 365 Copilot with Commercial Data Protection, or running a local model. Anything else is a policy violation waiting to happen.
The third biggest gap we see in real audits: shared chat links left public on Claude and ChatGPT. People share a chat once with a colleague and forget the link is publicly viewable. Audit your shared chats every quarter and revoke anything that contains personal or professional context.
Finally, on personal habits. Use Temporary Chats and Incognito mode like you would use private browsing windows: for one-off sensitive queries that should not live in your history. Anything you would not want a colleague, spouse, or future employer to see in your chat history belongs in a Temporary or Incognito conversation. The friction is low; the protection is meaningful.
Verdict: the right privacy posture for your situation
Honest recommendations by threat model. No paranoia.
If you use AI chat tools casually for everyday tasks
Run the 10-step audit once, set quarterly reminders, done
Most people are over-thinking privacy and under-doing the basic settings. Turn off model training on the 3 tools that allow it, shorten Gemini retention, use Temporary Chats for anything sensitive. Total time to set up: 20 minutes. Total cost: $0. This covers 80% of realistic privacy concerns.
If you use AI chat tools for work content
Move to paid enterprise tier with stronger defaults
ChatGPT Team or Enterprise, Claude Team or Enterprise, or Microsoft 365 Copilot with Commercial Data Protection. All three guarantee no training on customer data and add admin controls. Cost: $20-30/seat/month. Worth it for anyone who handles customer names, internal documents, financial data, or strategic content. Read your employer's AI policy before assuming you are compliant.
If you handle health, legal, or financial information
Local models for sensitive cases, paid AI for general work
For the highest-sensitivity work (patient records, attorney-client privileged content, M&A data), run an open-source model locally (Llama, Mistral, DeepSeek). The data never leaves your machine. For lower-sensitivity work, use paid enterprise AI with audit logs. Hybrid approach is the practical answer for regulated industries.
If you are a journalist, activist, or political target
Local models + signed-out browsers + minimal metadata
Threat model includes lawful access by states, leaks, or targeted attacks. Use Llama or Mistral running locally for anything sensitive. Use commercial AI tools only on a clean, signed-out browser profile for non-sensitive queries. Avoid linking AI tools to email or calendar. This is the most defensive posture and the most friction; only worth it for genuine high-risk threat models.
Where we would NOT spend time
Worrying about training data after turning off model training
Once you have turned off the model training toggle and use paid enterprise tiers for sensitive work, additional handwringing about whether your data is being seen is mostly anxiety, not action. Trust the major providers' published policies on the settings they expose. Focus your effort on the actual leak surface: shared chat links, connected apps, and mixing work and personal accounts. Those are the gaps that cause real incidents.
Want the privacy audit as a downloadable checklist?
We packaged the 10-step audit into a printable PDF checklist plus a Notion template you can clone. Pin one in your bookmarks, run it once a quarter, never lose track of your settings drift.
Section 3
FAQ: AI chat privacy
What people ask most often about AI chat data, training, and retention.
Does ChatGPT use my conversations to train its models?
What is the most private AI chat tool I can use?
Can my employer see my AI chat history?
What is the difference between Temporary Chat and Incognito mode?
How long do AI companies keep my conversations?
Should I be worried about AI tools reading my private messages?
Does turning off chat history actually delete past conversations?
Is it safe to paste work emails or company documents into ChatGPT?
What is Commercial Data Protection in Microsoft Copilot?
How often should I audit my AI chat privacy settings?
Keep reading
More AI guides
- Industry Guides
AI Tools for Medical Office Managers
US guide to AI tools for medical office managers covering SOPs, scheduling scripts, patient education drafts, and privacy
Read guide → - AI Tools & Apps
AI Character Chat Prompts
Prompts for AI character chats including fictional characters, historical figures, fantasy RPG, and movie character roleplay
Read guide → - Industry Guides
Best AI Tools for Healthcare Admin
Buying guide for healthcare admin AI tools covering scheduling, SOPs, patient communication, and privacy
Read guide → - AI Tools & Apps
AI Chat Prompts
Prompts for AI chatbot conversations including brainstorming, debate practice, interview simulation, and creative conversations
Read guide → - AI Model Prompts
Midjourney Prompts
Stunning image generation with Midjourney prompt mastery
Read guide → - AI Model Prompts
ChatGPT Prompts
Write clearer ChatGPT prompts with goals, context, examples, and practical checks
Read guide →