The Tool Desk
Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →Outbyte Driver Updater FREEFix the driver behind crashes, sound loss and screen glitchesFind Drivers →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
Turn off your AI service’s model-training control before you submit research, and verify the setting in the account and app you will use. This usually excludes new chats from future model improvement; it does not erase older content, stop every form of processing or safety review, or override feedback and administrator rules.
What changing a training setting actually controls
AI assistants separate several activities that are often described loosely as “using your data.” A training or model-improvement switch controls one activity: whether eligible content may be used to improve future models. It is not a universal confidentiality switch.
| Decision point | Question to answer |
|---|---|
| Scope | Does the control cover text prompts only, or also uploads, coding sessions, voice, images and context from connected apps? |
| Time direction | Does it affect content sent after the change, or does it also change the treatment of earlier chats? |
| Retention | How long can content remain available for service delivery, abuse prevention, feedback handling or other operational purposes? |
| Human and safety handling | Can authorized personnel or service providers review content to protect the service, users or the public? |
| Feedback exceptions | Can a thumbs-up, thumbs-down or other feedback submission attach the conversation to model-improvement work? |
| Account authority | Is the switch yours, or can a school, employer or administrator set the policy? |
Policies and menu labels change. The instructions below reflect the providers’ documentation checked on October 13, 2026. Open the linked help page and confirm the switch is still present immediately before sending sensitive material.
Recommended Free Tools
ChatGPT: turn off “Improve the model for everyone”
Steps for a personal Free, Plus or Pro account
- Open ChatGPT and select your profile menu.
- Choose Settings, then Data Controls.
- Turn off Improve the model for everyone.
- Leave the setting off in the account that will actually receive your research. If you use more than one account, repeat the check after switching accounts or devices.
OpenAI says conversations and tasks sent after you opt out are not used to train its models. The setting is account-level and you do not have to delete your existing chat history to change it. Read the current Data Controls FAQ and model-improvement data guidance for the account-specific terms.
#1 Best Overall
The feedback exception
For personal Free, Plus and Pro users, OpenAI says that submitting feedback on a response, including a thumbs-up or thumbs-down, can make the conversation associated with that feedback available for model improvement even after you opted out. Do not send feedback on a conversation you need excluded. If you must report a problem, first check what information the report form will attach.
What the switch does not promise
Opting out concerns model training; it does not mean the service performs no processing. The service still has to receive and handle a prompt to answer it, and other security, abuse-prevention, legal or operational practices can apply. Treat the control as a limit on training use, not as a guarantee that nobody can ever access or review the material.
Claude: disable Model Improvement or use an Incognito chat
Change the consumer privacy setting
- Sign in to your consumer Claude account.
- Open the privacy or account settings area.
- Find the Model Improvement control and turn it off.
- Start a new chat or coding session only after confirming the control is off.
Anthropic says turning off Model Improvement means new chats and coding sessions are not used for future model training. The exact location and wording can change, so verify the live control using Anthropic’s model-improvement settings instructions.
Incognito chats
Anthropic also says Incognito chats are not used to improve Claude, even when the main Model Improvement setting is enabled. Use Incognito when you need a per-conversation option, but still read the current retention and safety terms before uploading confidential source material. The distinction between model improvement and ordinary service processing remains; the training exclusion is not a promise of zero handling.
For the provider’s explanation of consumer data use, consult the training-use FAQ. Business and organizational accounts can have different contractual controls, so do not assume the consumer setting applies to them.
Gemini Apps: switch off Keep Activity in your Google Account
Steps for a personal Google account
- Open your Google Account and go to Gemini Apps Activity.
- Find Keep Activity.
- Turn it off, then confirm the choice if Google displays a confirmation dialog.
- Return to the activity page and check that the status is still off before submitting your research.
Google says turning Keep Activity off stops future chats from appearing in Activity and from being used to train Google’s AI models, unless you submit feedback. This setting is separate from other Google services and account controls. Some connected apps may become unavailable when it is off.
The 72-hour retention limit
Turning Keep Activity off does not make a chat disappear immediately. Google says qualifying Gemini chats are retained with the account for 72 hours so it can provide the service, process feedback and protect Google, users and the public. That retention period is a Gemini Apps policy detail, not a general rule for other AI assistants.
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 minutePC Slower Than It Used to Be?
A free scan shows the junk files, broken settings and background clutter dragging Windows down - then fixes them in one click.Free scan · Windows 10 & 11Feedback, saved activity and service-provider review
If Keep Activity remains on, saved chats and shared files may be used to improve Google services, including training models. Google also states that activity reviewed by service providers is disconnected from the account and retained for three years. Submitting feedback can create an exception to the Keep Activity opt-out, so avoid feedback on a conversation that must stay outside training use.
Connected Google content
Google’s Gemini Apps Privacy Hub says Gemini does not train generative models directly on an entire Gmail inbox, Drive, Contacts, Calendar or Photos library. However, summaries, excerpts, generated media and inferences from relevant connected content can help answer a prompt and can be used to train models when the applicable activity settings permit it. Do not interpret the statement as “connected-app data is never used.” Also remember that Gemini settings do not control processing by every other Google service.
Voice and Live controls
Audio and Gemini Live video or screenshare improvement controls are distinct from Keep Activity. If your research session includes voice, video or screen sharing, inspect those improvement settings separately rather than assuming the text-activity switch covers them.
Microsoft Copilot: check conversation and voice training controls
Individual-account procedure
- Open Copilot and select your profile.
- Choose Privacy or Privacy settings.
- Look for Training on conversation activity and turn it off if you do not want eligible conversations used for model training.
- If you use voice, find Training on voice conversations and set it independently.
- Save the choices and reopen the privacy panel to verify them.
Microsoft presents these controls for individual accounts, but the path and available switches can differ among web, mobile, desktop and app versions. Use Microsoft’s current Copilot privacy controls page and follow the labels shown in your interface. A training opt-out can leave personalization enabled, so do not confuse the two settings.
Do not apply consumer instructions to work deployments
Microsoft 365 Copilot, work or school accounts and organization-managed deployments can operate under separate terms and administrator policies. The individual-account switches above are not evidence that your employer or school’s Copilot deployment treats prompts the same way. Use the approved organizational account and ask the administrator which data, retention and training terms apply before submitting internal research.
Personal, business and school accounts are different
Account authority is as important as the visible toggle. A personal account normally lets you change its consumer privacy setting. A business or school tenant may enforce retention, logging, access and training rules centrally; an end user may be unable to override them. An administrator can also change a policy after you checked it.
- Use the organization’s approved AI account for organizational research rather than a personal account.
- Ask whether prompts, uploaded files, voice, generated outputs and connected data are logged or retained.
- Confirm whether the tenant’s contract excludes customer content from model training and what exceptions apply.
- Ask how to request deletion, export or an audit if the material is regulated or commercially sensitive.
Never infer business privacy terms from a consumer help page. The same product name can represent materially different controls depending on the account and administrator.
Training opt-out is not deletion or complete confidentiality
It usually applies forward, not backward
Each provider describes its opt-out primarily for new content after the change. Turning a switch off does not automatically erase previous conversations, saved files or activity records. If earlier material must be removed, use the service’s separate history or account-deletion controls and verify what deletion means for backups, legal retention and safety records.
Quick wins for a faster PC:
Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Clear out junk files and repair common Windows errorsFree Scan →Service delivery still requires processing
An assistant must process a prompt to generate an answer. Providers can also process content for security, abuse prevention, troubleshooting, legal compliance and service integrity. A training exclusion narrows one downstream use; it does not prevent all automated or human handling allowed by the provider’s terms.
Best Value
Feedback can override the practical result
OpenAI, Google and the other providers describe feedback pathways separately from ordinary chat handling. If a provider says feedback can make an opted-out conversation eligible for improvement work, do not submit that feedback on sensitive research. Use a redacted example or a different, non-sensitive conversation when possible.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Prepare research before you paste or upload it
- Classify the material. Remove personal identifiers, unpublished findings, credentials, contract terms and source passages that you are not authorized to disclose.
- Choose the account. For employer, client or school work, use the approved tenant and obtain its data-handling rules.
- Set the control first. Do not paste the material and then look for a privacy switch; many controls apply only to future content.
- Check every modality. Inspect text, file-upload, coding, voice, image, video and connected-app settings that your session will use.
- Use the minimum necessary data. Summarize or redact source material instead of uploading complete documents when the task does not require them.
- Do not provide feedback on the sensitive conversation. If you need to evaluate an answer, copy a sanitized excerpt into a separate test chat.
- Record the configuration. Note the account, date, setting name and app version so you can reproduce the decision for an audit.
Quick comparison of the four services
| Service and account covered | Control to check | What the provider says about new content | Important qualification |
|---|---|---|---|
| ChatGPT personal Free, Plus or Pro | Settings → Data Controls → Improve the model for everyone | New conversations and tasks after opting out are not used to train OpenAI models. | Feedback on a response can make the associated conversation available for improvement. |
| Claude consumer | Privacy settings → Model Improvement; Incognito is a separate option | New chats and coding sessions are not used for future training when Model Improvement is off; Incognito chats are not used to improve Claude. | Menu labels can change, and organizational terms may differ. |
| Gemini Apps personal Google account | Google Account → Gemini Apps Activity → Keep Activity | With Keep Activity off, future chats do not appear in Activity or train Google’s AI models unless feedback is submitted. | Chats are retained for 72 hours; connected-app, audio and Live controls have additional details. |
| Microsoft Copilot individual account | Profile → Privacy; conversation and voice training switches | Use the available controls to opt out of training on conversation and voice activity. | Paths vary by app; personalization may remain on, and work or school deployments are governed separately. |
When the setting is missing or appears to fail
You cannot find the named switch
Confirm that you are in the intended account, not a different profile or tenant. Update the app or open the provider’s web interface, then search the current privacy help page for the exact control name. Providers can rename or relocate settings by region, subscription and platform.
The switch turned itself back on
Check whether an administrator controls the account, whether you changed accounts, or whether a different app has its own privacy panel. Recheck after signing out and back in. Do not submit sensitive research until the status is visibly off in the active account.
Do these 3 things before closing this tab:
1Clear out junk files and repair common Windows errors2Fix the driver behind crashes, sound loss and screen glitches3Repair Windows errors before they cause bigger problemsYou already submitted the material
Changing the setting now may protect future content but does not retroactively change the treatment of the earlier submission. Stop sending additional copies, avoid feedback on the conversation, review the provider’s deletion and privacy-request process, and notify your organization’s data owner if the material was regulated or confidential.
A connected app or voice feature stopped working
Some controls intentionally disable integrations or have separate audio, video or screenshare settings. Decide whether the feature is necessary; if it is, use redacted input or an approved organizational account rather than silently re-enabling broad activity collection.
A practical rule for sensitive research
Set the training control before the first prompt, use the narrowest account and modality available, redact anything that identifies people or unpublished work, and treat every answer as processed data rather than a private notebook. Verify the setting again after an account, app or policy change. These steps reduce training exposure, but only the provider’s current terms and your organization’s rules can define the remaining handling.
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.
