iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
Yes, you can stop several major consumer AI chat services from using new conversations to improve their models, but “don’t train on my conversations” does not mean “nothing is stored, processed, or seen.” ChatGPT, Claude, and Gemini each provide a training or model-improvement control. Each also keeps separate rules for safety checks, feedback, temporary chats, connected apps, and other data handling. Use the provider-specific setting below, then verify what remains allowed for your account, region, and features.
What a training opt-out actually changes
Model training is only one purpose for which a chat provider may process your input. A service can stop using a new conversation to improve a general-purpose model while still processing it to generate a reply, detect abuse, enforce usage rules, maintain a feature you enabled, or investigate a safety signal. A provider may also treat feedback, temporary chats, uploaded files, voice, and connected-app content differently from ordinary text conversations.
That is why “not used for training” should be read narrowly: the conversation is excluded from the provider’s stated model-improvement process under the conditions in its policy. It is not a promise of deletion, zero retention, end-to-end encryption, or zero human access. None of the policies compared here supplies one common retention period that can be fairly compared across services.
Recommended Free Tools
| Question | Why it is separate from training |
|---|---|
| Is model improvement enabled? | This is the setting that determines whether eligible conversations may help train or improve models. |
| Is the conversation stored in history? | History, account activity, and model-improvement eligibility are different controls. |
| Can safety systems inspect it? | Providers may analyze chats for abuse prevention, policy enforcement, or security even after an opt-out. |
| What happens if I send feedback? | A feedback action can submit the associated conversation for improvement, overriding the practical protection you expected from a general opt-out. |
| Are temporary or incognito chats different? | These modes have their own rules. They may change training eligibility without eliminating all processing. |
| What about connected apps and files? | Prompts, responses, excerpts, summaries, generated media, or inferences drawn from another service can fall under separate connected-data terms. |
Comparison: the three documented consumer options
The table uses only controls and exceptions described in the providers’ current consumer guidance. A default marked “not stated” means the cited page does not establish a universal default for every account, geography, or product variant.
#1 Best Overall
| Service | Opt-out or privacy control | What the control covers | Temporary/incognito behavior | Important exception |
|---|---|---|---|---|
| ChatGPT | Improve the model for everyone in Settings > Data controls, or the privacy portal’s “Do not train on my content” choice. | When off, new conversations are not used to train OpenAI models. Whether the setting is on by default is not established in the cited guidance. | Temporary Chat is not used for training, does not appear in history, and does not create or use memories. | Submitting feedback may make the entire associated conversation available for model improvement even after opting out. |
| Claude | Anthropic’s Model Improvement preference for consumer Free, Pro, and Max accounts. | The preference controls whether conversations are used to improve Claude; the cited guidance does not establish one default for every account. | Incognito chats are not used to improve Claude even when Model Improvement is enabled. | Conversations flagged for safety review may be analyzed to improve policy enforcement and train safeguards models. |
| Gemini Apps | Keep Activity in Gemini Apps Activity controls whether future activity is used to improve Google AI. | Turning it off stops the stated future-activity use for improvement, but chats are still processed to answer you and protect Google, users, and the public. | Temporary Chat has separate model-improvement and human-review caveats; it is not a blanket “never processed” mode. | Human review can still occur for service protection and safety purposes. Connected-app data has additional rules when Keep Activity is on. |
For the underlying wording, see OpenAI’s data-use guidance, Anthropic’s model-training guidance, and Google’s Gemini Apps Privacy Hub. Policies and labels can change, so check the live control immediately before entering sensitive information.
ChatGPT: turn off “Improve the model for everyone”
What the setting does
For an individual-plan account, open Settings > Data controls and turn off Improve the model for everyone. OpenAI also describes a privacy-portal option called Do not train on my content. With the setting off, new conversations are not used to train OpenAI models. The wording is prospective: it applies to new conversations after the control is changed, rather than promising that every older conversation is retroactively removed from all improvement workflows.
How to change it
- Sign in to ChatGPT.
- Open your account menu and choose Settings.
- Select Data controls.
- Switch Improve the model for everyone off.
- Return to the control later if you change your preference; settings and account behavior can change with product updates.
The privacy-portal route is an alternative if you manage the preference outside the chat interface. Use the exact current label shown in the portal rather than assuming that a history or memory control also changes training eligibility.
Free tools Windows power users keep installed
One-click scans. No signup required.
The feedback exception
OpenAI says that if you submit feedback on a response, the entire conversation associated with that feedback may be used for model improvement even when you had opted out. Treat thumbs-up, thumbs-down, written comments, and similar feedback as a separate data-submission decision. If a conversation contains confidential material, do not attach feedback unless you accept that exception.
When Temporary Chat is the better fit
OpenAI says Temporary Chats do not appear in history, do not create or use memories, and are not used for model training. That addresses history, memory, and training at once for this mode, but it should not be generalized into a claim that no processing occurs. The service still has to process the request to produce a response and apply its other operational and safety rules.
Claude: use the Model Improvement preference and Incognito chats
What Anthropic documents
Anthropic’s consumer guidance covers Free, Pro, and Max plans and describes a Model Improvement choice. Use the privacy controls in your Claude account and locate that preference before starting a conversation you do not want used to improve Claude. The guidance does not establish one universal default for all users, so inspect the control instead of relying on what another account or older screenshot shows.
Incognito mode
Anthropic says Incognito chats are not used to improve Claude even when the Model Improvement preference is enabled. Choose Incognito for a conversation that should have this narrower improvement treatment. Incognito is not equivalent to a promise that the conversation cannot be processed for safety or service operation.
The safety-review carve-out
Anthropic states that conversations flagged for safety review may be used or analyzed to improve detection and enforcement of its Usage Policy, including training models for its safeguards team. Therefore, “Claude never reviews chats” is inaccurate. The model-improvement preference and the safety-review process serve different purposes; disabling one does not disable the other.
A practical Claude workflow
- Open your Claude account’s privacy or data settings.
- Find the control labeled Model Improvement and read its current explanation.
- Disable it before sending a new conversation if you do not want eligible chats used to improve Claude.
- For an especially sensitive session, start an Incognito chat and confirm that the interface indicates the mode.
- Avoid assuming that an Incognito session is exempt from safety analysis if Anthropic flags it.
Gemini Apps: understand Keep Activity, Temporary Chat, and human review
Keep Activity controls future improvement use
Google’s signed-in Gemini Apps guidance says Keep Activity controls whether future activity is used to improve Google AI. Manage it from Gemini Apps Activity. Turning Keep Activity off does not stop Gemini from processing a prompt: Google says chats are still used to provide responses and help protect Google, its users, and the public. That protection can include human reviewers.
Google distinguishes signed-in and signed-out use, and its Gemini Apps Privacy Hub supplements the broader Google Privacy Policy. The hub was listed as updated September 24, 2026; the associated privacy notice lists June 29, 2026. Those dates matter because labels, defaults, and regional treatment can change.
How to change Keep Activity
- Open Gemini Apps while signed in.
- Open the account or activity controls and choose Gemini Apps Activity.
- Find Keep Activity.
- Turn it off, then read the confirmation describing what happens to future activity.
- Use Temporary Chat when you need its separate handling, and reread the current notice because Temporary Chat has its own improvement and review caveats.
Why “off” does not mean “unseen”
Google says that with Keep Activity off, conversations remain available for generating replies and for protection of Google, users, and the public. Human review may still occur for those purposes. Google also says Temporary Chats are not used to improve Google AI with help from human reviewers, but the same notice does not say Temporary Chats are never processed. Keep the distinctions intact: no improvement use, no human review for that improvement purpose, and no processing at all are three different claims.
Crashes, No Sound, or Screen Glitches?
Random freezes, missing sound and display glitches usually trace back to one bad driver. Find and replace yours safely.Free scan · under a minuteWindows Errors? Fix Them Before They Spread
Repair common Windows errors and clear accumulated junk for a smoother, more stable PC - no reinstall needed.Free scan · no reinstallConnected apps can change the risk calculation
Gemini’s connected-app feature deserves separate attention because the material entering a conversation may be much broader than the text you type. Google says Gemini Apps activity that includes connected-app data can be used to improve Google services, including training generative AI models, when Keep Activity is on.
The potentially relevant material can include prompts and responses plus summaries, excerpts, generated media, and inferences from connected services. Google says a subset of Connected Apps data may be reviewed by people. It describes steps such as disconnecting interactions from the Google Account and reducing personal information before reviewers inspect them, subject to stated exceptions.
Google’s own Connected Apps guidance tells users: Please don’t connect apps to Gemini if they contain personal or confidential information that you wouldn’t want to be used for training generative AI models.
That is specific advice about connected-app data, not a universal legal rule and not a statement about ChatGPT or Claude.
Before connecting an app, ask:
- Could the app expose messages, health details, financial records, work documents, or private contact information?
- Will Gemini receive derived summaries or inferences even if the original document is not displayed in the chat?
- Is Keep Activity currently on for the account and region?
- Can you disconnect the app after the task, and what activity remains available under the notice?
Read Google’s current explanation at About personalization with Connected Apps before enabling a connection.
Choose the right control for your situation
You want ordinary chats excluded from model improvement
- ChatGPT: turn off Improve the model for everyone before starting new conversations.
- Claude: disable the Model Improvement preference.
- Gemini: turn off Keep Activity for future activity.
These actions address model improvement only. They do not erase an existing history, prevent all safety processing, or change what happens when you submit feedback.
Best Value
You want a one-off sensitive session
Use ChatGPT Temporary Chat or Claude Incognito, because both providers explicitly describe those modes as excluded from their respective training or improvement process. Gemini Temporary Chat has separate caveats, so read the current in-product notice rather than assuming it has identical behavior.
You may need to report a bad answer
On ChatGPT, feedback can make the entire associated conversation available for model improvement despite an opt-out. If feedback is necessary, remove confidential details first when the product allows it, or accept that the feedback exception applies.
You plan to connect another service
Review connected-app permissions separately. Gemini can use connected-app material for service improvement and model training when Keep Activity is on, and some of that data may be reviewed by people. Do not apply the privacy intuition for a short typed prompt to a mailbox, drive, calendar, or other connected source.
The Tool Desk
Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →A verification checklist before sharing sensitive information
- Identify the account type. Consumer, work, school, signed-in, and signed-out experiences may have different controls.
- Open the live data setting. Confirm the exact label and current state immediately before the session.
- Check the scope. Determine whether the control applies to future chats only, as ChatGPT’s wording does.
- Choose a temporary mode where appropriate. Read its own notice instead of treating it as a universal no-processing mode.
- Do not send feedback casually. In ChatGPT, feedback may expose the full associated conversation for improvement.
- Review integrations. Disconnect apps you do not need and avoid linking sources containing information you would not want used for model training.
- Separate safety from training. Expect abuse prevention, policy enforcement, or security processing to continue where the provider says it does.
- Recheck after updates. Settings names, defaults, product modes, and regional behavior are volatile.
What this comparison cannot establish
The official pages do not provide a common retention-period measurement, a comparable percentage of chats reviewed, or evidence for ranking one service as the universally “most private.” They also do not establish current consumer practices for every other AI chat product or for self-hosted and local models. The defensible conclusion is narrower: these three services document controls that can reduce model-improvement use, while retaining distinct safety, feedback, temporary-chat, and connected-data exceptions.
If your requirement is that a provider must never receive or retain plaintext data, a training opt-out is the wrong control to evaluate. You would need a separate architecture and policy review covering hosting, encryption, access, logging, and deletion—questions not answered by the model-improvement settings described here.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
