Free tools Windows power users keep installed
One-click scans. No signup required.
AI apps must process your prompt or uploaded file to generate a response. Depending on the product, account type and settings, the same data may also be checked for safety, used to personalize your experience, sent for human review, retained for a stated period, or used to improve models. Those purposes are separate: turning off model training does not automatically delete chat history, memory, feedback records or provider backups.
This guide explains the current published policies for ChatGPT consumer plans, Claude consumer products and Gemini Apps. Policies, controls and retention periods can change by plan, region, workspace and date, so treat the provider’s linked settings and notices as the final authority for your account.
As an Amazon Associate I earn from qualifying purchases.
What happens when you submit a prompt or file?
Your text, image, document, spreadsheet or other upload is transmitted to the provider’s systems so the service can interpret it and return an answer. Gemini’s privacy notice expressly lists prompts, shared files, generated content, device and app information, connected-app data and related activity among the information it may collect (Gemini Apps Privacy Hub). Other providers describe comparable processing in their product terms, although the exact categories differ.
Do these 3 things before closing this tab:
1Clear out junk files and repair common Windows errors2Scan for outdated or missing drivers - takes under a minute3Repair Windows errors before they cause bigger problemsThat basic processing is only the first step. A provider may apply several additional purposes, each with its own control or exception:
#1 Best Overall
- Response generation: The service analyzes the input, retrieves any permitted context and runs a model to produce the requested output.
- Security and abuse prevention: Automated systems, and sometimes trained reviewers, may inspect content for fraud, malware, self-harm, illegal activity, prompt injection or other safety risks.
- Model improvement: Eligible conversations may be used to evaluate or train future models. This is not the same as keeping a chat in your history.
- Personalization: A product may save memories, preferences or conversation context to make later answers more relevant.
- Human review: A limited sample, a safety-flagged conversation or a conversation attached to feedback may be examined by people or service providers under the provider’s procedures.
- Connected services: If you authorize a connector, the app can process material returned by that service. The connector permission and the AI app’s own data settings are separate decisions.
No published document establishes a universal percentage of prompts reviewed by humans or a universal probability that a particular upload will enter training. Avoid claims such as “the app trains on everything you type”; eligibility depends on the product, plan, setting, feedback action and applicable exception.
Training, retention, deletion and memory are different controls
These terms are often used together, but they answer different questions:
| Control or process | What it answers | What it does not guarantee |
|---|---|---|
| Model improvement or training | Whether eligible content may help evaluate or improve models. | It does not erase a saved chat, remove a memory or undo a feedback submission. |
| Chat history | Whether you can see and manage a conversation in your account. | Deleting history may not remove safety, legal, billing or human-review records kept under a separate policy. |
| Temporary or Incognito mode | Whether a conversation is excluded from history, memory or model-improvement use under that product’s rules. | It does not mean instant deletion; safety retention and other exceptions can still apply. |
| Memory or personalization | Whether details can be saved and reused in later conversations. | Turning memory off does not necessarily delete the original conversation. |
| Feedback | Whether a thumbs-up, thumbs-down or written report attaches the conversation to a quality or safety investigation. | Opting out of training does not always override feedback handling. |
| Connected-app permission | Whether the AI can retrieve data from another service. | Revoking access later may not erase content already copied into a prompt or answer. |
OpenAI explicitly says that turning off model improvement does not delete saved chats. Anthropic and Google likewise describe separate retention and review conditions. Always check each setting independently rather than treating a single “privacy” switch as a complete deletion command.
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 →ChatGPT: consumer, temporary and business/API rules
Individual Free, Plus and Pro plans
OpenAI says content from individual Free, Plus and Pro accounts may be used to improve models. You can turn off Improve the model for everyone in Settings > Data controls, or use the Privacy Portal. OpenAI’s published wording is precise: “After you opt out, we won’t use your new conversations to improve our models.” The setting applies to new conversations after the change; it does not retroactively remove older saved chats or erase account data. See OpenAI’s data-use guidance.
Feedback is an important exception. OpenAI says a conversation associated with feedback may be used to improve models even after you have opted out. If you do not want a conversation considered through that route, do not attach a rating or written feedback to it.
Temporary Chat
Temporary chats do not appear in history, do not create or update memories and are not used to improve models while they remain temporary. OpenAI says they may nevertheless be retained for up to 30 days for safety. Saving one changes its status: “If you save a temporary chat, it becomes a regular chat and follows your account’s personalization and model-improvement settings.” That means temporary mode is a limited-use option, not a promise that no copy exists on provider systems.
Business, Enterprise, Edu and API contexts
OpenAI says Business, Enterprise, Edu and API inputs and outputs are not used for model improvement by default. “By default” matters: workspace administrators, contractual terms, regional settings and a deliberate opt-in can change the treatment. A consumer account and an organization’s managed workspace should never be assumed to have identical controls.
Claude: consumer settings, Incognito and feedback
Consumer products
Anthropic’s consumer guidance covers Free, Pro, Max and accounts using Claude Code. Anthropic says chats or coding sessions may be used for improvement when the user allows it, when a conversation is flagged for safety review, or when the user otherwise explicitly opts in. Review Privacy Settings and the Model Improvement choice for the account you are using. The policy is not a blanket statement that every consumer conversation is trained on.
Rank #3
Incognito chats
Anthropic states: “Your Incognito chats are not used to improve Claude, even if you have enabled Model Improvement in your Privacy Settings.” Incognito therefore provides a stronger model-improvement exclusion for eligible conversations, but it should not be described as a complete ban on service processing or safety checks. The conversation still has to be processed to answer you, and safety-related rules can apply.
Feedback and connected sources
When you submit feedback, Anthropic says the whole related conversation is stored in its backend for up to five years. That is a retention rule for the feedback-associated conversation, not a claim that every Claude chat is retained for five years.
Anthropic also distinguishes raw connector content from text you copy into the chat. Raw content returned by a connected service is not included in model-improvement data unless you directly copy it into the conversation. Once you paste that material into your prompt, it becomes conversation content and is governed accordingly. Commercial Claude processing has a different controller-and-customer context, described in Anthropic’s commercial and personal-data explanation.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Gemini Apps: Keep Activity, temporary chats and human review
Keep Activity off
Google says that when Keep Activity is off, future Gemini chats do not appear in Gemini Apps Activity and are not used to train AI models unless you send feedback. Google still retains those chats with the account for 72 hours for providing responses, processing feedback and protecting the service. This is an operational retention period, not a promise that the content is immediately destroyed when the response finishes.
Temporary chats
Google gives temporary chats the same stated 72-hour retention window. Its notice says: “Temporary chats and chats you have when Keep Activity is off are retained with your account for 72 hours” for response, feedback and protection purposes. Temporary status therefore limits activity history and training use under the stated conditions, while allowing short-term service and safety handling.
Activity auto-delete and reviewer retention
Gemini Apps Activity uses an 18-month auto-delete default. Google says users can adjust that period to three months, 36 months or indefinite retention. A separate exception applies to conversations selected for human review. Google writes: “Human reviewers (including trained reviewers from our service providers) review some of the data we collect for these purposes.” Reviewer-handled chats may be kept for up to three years, and Google says those records are not removed when you delete Gemini Apps Activity. Before sending chats to service providers for review, Google says they are disconnected from the account.
Feedback, files and connected apps
Gemini feedback can include the recent conversation context, uploaded files and content from connected apps. Sending feedback can therefore create a different handling path from simply asking a question. Review connected-app permissions separately, and remember that audio and Live features can have their own controls and notices.
Recommended Free Tools
Details are in Google’s Gemini Apps Privacy Hub, including the June 29, 2026 notice update.
Best Value
Provider comparison at a glance
| Service and account scope | Model-improvement treatment | Retention, review and exceptions | Useful action |
|---|---|---|---|
| ChatGPT Free, Plus and Pro | May be used for improvement; opt out through Settings > Data controls or the Privacy Portal. New conversations after opt-out are excluded. | Temporary chats may be retained up to 30 days for safety. Feedback can make the associated conversation eligible for improvement. Opting out does not delete saved chats. | Set model improvement, history/deletion and Memory separately. |
| Claude Free, Pro, Max and Claude Code accounts | Used when the user permits improvement, when safety review requires it or when the user explicitly opts in. Incognito chats are excluded from improvement. | Feedback-associated conversations may be stored up to five years. Safety handling can still apply to Incognito chats. Connector content is treated differently unless copied into the chat. | Check Privacy Settings and use Incognito for conversations you do not want used for improvement. |
| Gemini Apps | Keep Activity off prevents future chats from training models unless feedback is sent. Temporary chats are not used to train models. | Keep Activity-off and temporary chats are retained 72 hours. Activity auto-delete defaults to 18 months, with three-month, 36-month and indefinite options. Human-reviewed material may remain up to three years. | Review Keep Activity, Activity auto-delete, connected apps and separate audio or Live controls. |
| OpenAI Business, Enterprise, Edu and API | Inputs and outputs are not used for model improvement by default. | Workspace, contract, administrator and configuration rules determine additional handling. | Confirm the organization’s current data-processing terms rather than relying on consumer settings. |
The durations in this table are provider policy limits or defaults, not measurements of how often a specific user’s data is reviewed or trained on. The documents do not provide a comparable cross-provider review rate.
What happens to an uploaded file?
During processing
The app may extract text, recognize images, transcribe audio, inspect metadata or convert the file into internal representations so it can answer your question. A file can therefore be processed even when the final answer quotes only a small portion of it. Uploading a document to a temporary chat does not mean the provider never receives or stores a copy during the response and safety window.
After the answer
File handling normally follows the conversation, account and product policy, but the three providers do not publish one universal file-lifetime rule in the material cited here. A saved chat, a temporary chat, a feedback submission and a connected-app result can each have different retention paths. If a file contains sensitive material, remove unnecessary pages, names, account numbers, hidden metadata and comments before upload.
Connected files versus pasted text
A connector may retrieve a file or record on demand rather than permanently importing an entire account. Anthropic explicitly says raw connector content is not included for model improvement unless you copy it into the conversation. Once copied, the text is part of the prompt and can be handled under the conversation’s rules. Other products may define connectors differently, so read the connector-specific notice instead of assuming the same behavior.
How to use AI apps with less exposure
- Classify the material first. Separate public information, ordinary personal data, confidential business material, regulated records, credentials and information about other people. Do not upload secrets merely to save typing.
- Choose the right account. Consumer and business/API defaults differ. Use a managed workspace only when its administrator and contract provide the controls your organization requires.
- Minimize the prompt. Replace names with roles, remove unnecessary columns, crop unrelated pages and redact account numbers. Ask whether the model can solve the task with an excerpt rather than the original file.
- Set the product controls before sending. For ChatGPT, check Settings > Data controls and Memory. For Claude, review Privacy Settings and select Incognito when appropriate. For Gemini, check Keep Activity, Activity auto-delete and connected-app permissions.
- Do not send feedback on sensitive conversations. Feedback can attach the whole conversation to a longer review or retention process. If you need to report a problem, remove sensitive details first when the interface permits.
- Use temporary modes for the cases they actually cover. They can limit history, memory or model-improvement use, but they do not eliminate response processing, safety checks or stated retention windows.
- Clean up separately. Delete saved chats, remove memories, revoke connectors and adjust activity retention as separate actions. Turning off training alone is not a deletion command.
- Verify after account changes. Recheck the active account, workspace, browser profile and region. A setting changed on one account or product may not carry over to another.
Common misconceptions and the accurate version
- “Turning off training means nothing is stored.” False. ChatGPT saved chats remain unless deleted; temporary and Keep Activity-off modes have stated safety or service windows; feedback and human-review exceptions can last longer.
- “Temporary means instantly erased.” False. ChatGPT documents up to 30 days for safety, while Gemini documents 72 hours for temporary and Keep Activity-off chats.
- “A connector never exposes the file to the AI provider.” Too broad. The app must process returned content to answer you, and copying connector content into a prompt can change how it is handled.
- “Human review means every chat is read.” Unsupported. Google says some data is reviewed, but the cited policies do not publish a universal share or a probability for an individual conversation.
- “Business plans have exactly the same privacy rules everywhere.” Unsupported. Defaults, contracts, administrator settings, region and API configuration matter.
Bottom line for everyday users
Assume that anything you submit is processed to produce the answer, then identify which additional purposes you are willing to accept. Keep model improvement, history, memory, feedback, temporary mode and connector permissions conceptually separate. Use the narrowest account, prompt, file and setting that will accomplish the task, and read the current policy for the exact product and workspace before uploading information that would cause harm if retained or reviewed.
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.
