Free tools Windows power users keep installed
One-click scans. No signup required.
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
Duck.ai is the clearest documented choice if you want a hosted chatbot that does not train on your conversations by default. DuckDuckGo says chats are not recorded or stored by default, requests are anonymized before being sent to model providers, and providers must delete prompts and responses immediately. Claude is a strong standalone alternative when you use Incognito chats. ChatGPT requires you to disable model improvement or use Temporary Chat on consumer accounts; Business, Enterprise, Edu, and API accounts have stronger default controls.
The short answer
Choose according to the privacy control you are willing to manage:
- Duck.ai: best documented default for no training and no chat storage. DuckDuckGo says providers must delete prompts and responses immediately and never use Duck.ai conversations to train their models.
- Claude: use Incognito chats. Anthropic says Incognito chats are not used to improve Claude, even when Model Improvement is enabled in Privacy Settings.
- ChatGPT: on Free, Plus, and Pro accounts, turn off Improve the model for everyone before entering sensitive material, or use Temporary Chat. Business, Enterprise, Edu, and API data is not used to improve models by default.
“Not used for training” does not mean “never retained.” Safety investigations, feedback, legal obligations, optional backups, and abuse prevention can create separate copies or longer retention periods.
Why no training is not the same as no retention
Model training is only one possible use of a conversation. A service may keep text temporarily to produce a response, retain a flagged exchange for abuse or safety review, preserve feedback you submit, or store a backup that you explicitly enabled. Deleting a chat from the visible history may also be different from deleting backend copies.
#1 Best Overall
Controls can be forward-looking. Turning off model improvement generally prevents new conversations from being added to future training workflows; it does not necessarily remove data that was already included in a training run or information already incorporated into a trained model. Treat every privacy setting as a rule for future activity unless the provider expressly promises broader deletion.
Privacy comparison
| Service | No-training default | Chat storage | Deletion and retention | Safety and feedback paths | Model and account scope |
|---|---|---|---|---|---|
| Duck.ai | Yes. DuckDuckGo says conversations are not used to train models by DuckDuckGo or its providers. | No recording or storage by default. Optional Sync & Backup can store chats. | Providers agree to delete prompts and responses immediately under zero-data-retention terms. Backup retention duration is not stated by DuckDuckGo. | Feedback sharing, legal requirements, abuse prevention, and uploaded-file moderation can create exceptions. | Hosted access to multiple providers, including examples such as OpenAI and Anthropic. Applies to Duck.ai use. |
| Claude | Incognito chats are not used to improve Claude. Ordinary chats can be used when Model Improvement is allowed, when flagged for safety review, or after another explicit opt-in. | Incognito status controls improvement use, not every possible operational copy. | Deleted conversations disappear from chat history immediately and are deleted from backend storage within 30 days. De-identified improvement data may remain up to five years; flagged policy-violation chats up to two years; feedback data up to five years. | Safety flags and feedback have separate retention rules. | Standalone assistant. Incognito protection is a consumer-chat control. |
| ChatGPT | Free, Plus, and Pro content may be used for training until you opt out. Temporary Chat is not used to improve models while temporary. | Temporary Chats do not appear in history or create or update memories. | OpenAI does not state in the cited policy that opting out removes data already used in training. Temporary Chat has its own temporary handling rules. | Feedback submitted after opting out can cause the entire associated conversation to be used for training. | Consumer controls differ from Business, Enterprise, Edu, and API products, whose inputs and outputs are not used to improve models by default. |
Duck.ai: the simplest documented default
What happens to a Duck.ai request
DuckDuckGo says Duck.ai does not record or store chats by default. Before a request is sent to an underlying provider, Duck.ai anonymizes it and removes metadata such as your IP address. The providers agree to delete prompts and responses immediately after generating a reply, a policy DuckDuckGo describes as zero data retention (ZDR). They also agree never to use Duck.ai conversations to train their models.
This gives Duck.ai a different starting point from consumer chat services that require an opt-out. You do not have to remember to change a training switch for the default behavior described by DuckDuckGo.
Exceptions you must understand
The no-storage, no-training posture has documented exceptions:
- Sync & Backup: enabling this optional feature creates a stored copy so chats can be synchronized or restored. DuckDuckGo does not state a universal retention period for those backups.
- Feedback: sharing feedback can send conversation content into a separate feedback process.
- Legal requirements: a disclosure or retention action may be required by law.
- Abuse prevention: activity can be analyzed or retained to protect the service and its users.
- Uploaded files: files may undergo moderation, creating processing outside the ordinary prompt-and-response path.
How to use Duck.ai more privately
- Open Duck.ai without enabling Sync & Backup unless you specifically need cross-device recovery.
- Read the current privacy controls before starting a sensitive session; providers can change terms over time.
- Do not submit optional thumbs-up or thumbs-down feedback on a conversation that contains confidential details.
- Avoid uploading files that contain personal, legal, medical, financial, or employer-confidential information unless the moderation and retention implications are acceptable.
- Remember that anonymizing an IP address does not make the text itself anonymous. Remove names, account numbers, addresses, and unique identifiers before sending a prompt.
Claude: choose Incognito instead of relying on a global setting
What Incognito changes
Anthropic’s consumer privacy article dated March 16, 2026 states: “Your Incognito chats are not used to improve Claude, even if you have enabled Model Improvement in your Privacy Settings.” Incognito is therefore the clearest way to keep a specific conversation out of model-improvement use without depending on a global preference.
Outside Incognito, Anthropic says chats can be used to improve Claude when you allow Model Improvement, when a conversation is flagged for safety review, or when you provide another explicit opt-in. Turning off Model Improvement does not eliminate safety-review or other legally permitted processing.
Deletion is not immediate backend erasure
When you delete a Claude conversation, it leaves chat history immediately. Anthropic says backend storage deletion occurs within 30 days. Separate retention periods can apply:
Windows 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 reinstallOutdated Drivers Are Slowing You Down
One free scan finds every outdated or missing driver and matches the right update for your exact hardware.Free scan · exact hardware match- De-identified data used for model improvement may be retained for up to five years when you have allowed that use.
- Chats flagged for policy violations may be retained for up to two years.
- Feedback data may be retained for up to five years.
These periods explain why deleting a visible chat should not be interpreted as instant removal from every system.
A practical Claude workflow
- Start a new conversation in Incognito mode before entering sensitive information.
- Confirm that the conversation is visibly marked Incognito; do not assume a normal chat inherited the setting.
- Keep names, identifiers, credentials, and confidential attachments out of the prompt whenever a redacted example will work.
- Do not submit feedback on the sensitive exchange unless you accept the separate feedback-retention policy.
- Delete the conversation when finished if you no longer need it, while allowing for the stated backend deletion window.
ChatGPT: change the consumer default or use Temporary Chat
Turn off model improvement
OpenAI says individual Free, Plus, and Pro content may be used to train models. To opt out, use the exact path Settings > Data controls > Improve the model for everyone and turn the setting off. OpenAI also provides a Privacy Portal option labeled Do not train on my content.
This opt-out applies to new conversations. It should not be read as a promise that data already used in a training run, or knowledge already embedded in a trained model, will be removed.
Use Temporary Chat for a one-off sensitive session
Temporary Chat is the simpler choice when you want a conversation that is not used to improve models while it is temporary. OpenAI says Temporary Chats do not appear in history and do not create or update memories. Start a new chat, select the Temporary Chat option in the composer, and verify that the temporary indicator is present before typing sensitive content.
Do these 3 things before closing this tab:
1Fix the driver behind crashes, sound loss and screen glitches2Repair Windows errors before they cause bigger problems3Scan for outdated or missing drivers - takes under a minuteThe feedback exception
There is a material trap: submitting feedback after opting out can cause the entire associated conversation to be used for training. If a chat contains confidential information, do not use thumbs-up, thumbs-down, or a feedback form on that exchange.
Rank #4
Business and API accounts
OpenAI states that inputs and outputs from ChatGPT Business, Enterprise, Edu, and API products are not used to improve models by default. That is a stronger starting position than consumer Free, Plus, and Pro accounts, but administrators should still review workspace retention, access, logging, and legal-hold settings. “Not used for training” does not automatically mean that no employee, administrator, safety system, or infrastructure provider can access or retain the data.
Which option fits your situation?
| Your priority | Best fit | What to do |
|---|---|---|
| Least configuration for a hosted chatbot | Duck.ai | Leave Sync & Backup off, avoid feedback on sensitive chats, and redact identifiers. |
| Mainstream standalone assistant with a per-chat privacy mode | Claude | Use Incognito for every sensitive conversation and avoid feedback on it. |
| You already use ChatGPT personally | ChatGPT with controls enabled | Turn off Improve the model for everyone before sending sensitive prompts, or select Temporary Chat. |
| Organizational or developer workloads | ChatGPT Business, Enterprise, Edu, or API | Use the applicable product and verify workspace retention and administrator policies; training exclusion is the default. |
Privacy checklist for any AI chatbot
- Check the default: determine whether training is excluded automatically or only after an opt-out.
- Separate storage from training: ask where prompts, responses, backups, logs, and safety-review copies are retained.
- Read deletion language: distinguish removal from chat history from backend deletion and from removal of already-trained information.
- Handle feedback carefully: feedback can route an otherwise protected conversation into a separate retention or training process.
- Use temporary or incognito modes deliberately: verify the mode before typing, not after.
- Minimize the prompt: replace real names, account numbers, access tokens, addresses, and proprietary text with realistic placeholders.
- Protect uploads: files may be scanned for moderation even when ordinary chats are not stored.
- Recheck settings: privacy policies and interface labels can change, so confirm the current control before a sensitive session.
Common mistakes and fixes
Assuming an opt-out erases old training data
Fix: treat the setting as forward-looking unless the provider explicitly promises deletion from training datasets or models.
Confusing a hidden chat with zero retention
Fix: read the provider’s backend-deletion period and its exceptions for safety, legal compliance, abuse prevention, and feedback.
Submitting feedback on a protected conversation
Fix: give feedback only from a sanitized test chat when confidentiality matters.
Best Value
Turning on synchronization for convenience
Fix: enable backup only when recovery is worth creating a stored copy; otherwise leave it disabled.
Sending secrets because the service “does not train”
Fix: never paste passwords, API keys, authentication codes, or unnecessary personal identifiers. Training exclusion does not remove the risk of exposure during processing, safety review, or account compromise.
Bottom line
For the simplest documented privacy posture, start with Duck.ai and keep optional backup and feedback features off. For Claude, use Incognito chats rather than ordinary conversations. For consumer ChatGPT, disable Improve the model for everyone or use Temporary Chat, and never submit feedback on a sensitive exchange. Business, Enterprise, Edu, and API users receive a no-training default, but should still review retention and administrator controls. Recheck each provider’s current settings before relying on any privacy promise.
Recommended Free Tools
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.
