Does AI Keep or Learn From What I Enter?
AI services may keep what you enter, remember it for future conversations or use it to improve their systems. These are three different things, and the answer depends on the product, your settings and the type of account or licence you use.
A sensible rule is:
Unless you have checked the current terms and settings and they clearly say otherwise, assume that anything you enter could be stored, reviewed or used to improve the service.
That does not mean you should avoid AI. It means you should think before sharing personal, confidential or sensitive information.
The product information on this page was last reviewed on 18 September 2026. AI services and their terms change frequently, so always check the provider's current privacy information.
Keeping, remembering and learning are different
When people ask whether an AI "learns" from what they enter, they can mean several different things.
Keeping your conversation
An AI service may save your conversations so that you can return to them later.
The conversation might appear in your chat history, remain on the provider's systems for a certain period or be retained for security and legal reasons.
Deleting a conversation from your history does not necessarily mean that every copy disappears immediately. Providers may retain some information for a limited period, and information already separated from your account for safety review or model improvement may be treated differently.
Remembering information about you
Some AI assistants have a memory or personalisation feature.
This allows the assistant to use information from earlier conversations in later ones. For example, it might remember:
- Your name
- Where you live
- Your preferred writing style
- The work you do
- A project you are working on
- Your interests or preferences
This is not the same as training the underlying AI model. It is more like adding notes to your account which the assistant can refer to later.
Depending on the service, you may be able to view, change, delete or disable these memories. Turning memory off may not automatically delete information already saved, so check the available controls carefully.
Using your information to improve AI
A provider may use conversations to improve its products or train future versions of its AI models.
This does not usually mean that the AI instantly rewrites itself or permanently learns a new fact as soon as you type it. Your conversation may instead become part of a much larger collection of information used in later testing, evaluation, safety work or training.
Whether this happens can depend on:
- The AI product you use
- Whether you have a personal or business account
- Your subscription or licence
- Your privacy and data-control settings
- Whether you submit feedback
- Whether a conversation is selected for a safety review
- The country or region in which you use the service
A critical risk: your information could appear in answers to other users
When AI services use your information to improve their models, there is an important consequence that people often do not realise.
An improved model that has learned from your information could eventually include patterns, knowledge or details from what you entered when it answers questions from other users.
This is not the same as publishing your conversation or showing your words to others. The information becomes embedded in the AI's patterns rather than appearing as a direct quote.
However, the end result is similar: information you entered could be returned to other users as part of the AI's response.
For this reason, you should particularly avoid entering to any AI service:
- Personal information about yourself or others
- Unpublished work or confidential documents
- Information protected by confidentiality agreements
- Sensitive business or financial details
- Passwords, security codes or access information
The details that felt safe because they were in a private conversation could eventually be reflected back to other users if that service uses your conversations to improve its AI.
What do the leading AI services currently do?
The following is a simplified overview. It is not a substitute for reading the current terms for your particular account.
ChatGPT
For personal ChatGPT accounts, OpenAI says conversations may be used to improve its models. You can turn this off in ChatGPT's Data Controls by disabling Improve the model for everyone. Once disabled, new conversations should not be used for model training.
ChatGPT also offers Temporary Chat. Temporary conversations do not appear in your history, do not create memories and are not used for training. OpenAI may still retain them for up to 30 days for safety purposes.
One important exception involves feedback. If you give an answer a thumbs-up or thumbs-down, OpenAI says the conversation connected to that feedback may be used to improve its models—even if you previously opted out of training.
OpenAI says information submitted through business products and its API is not used for model training by default.
You can read OpenAI's explanation of how it uses data, its Data Controls FAQ and its information about ChatGPT memory.
Google Gemini
When Gemini's Keep activity setting is enabled, Google may save your conversations, uploaded files, photographs and other information. Google says this activity can be used to develop and improve its services, including training its generative AI models. Some conversations may be reviewed by people.
If you turn Keep activity off, future conversations should not be used to train Google's AI models unless you choose to submit feedback. Google may still retain conversations for a short period so that Gemini can respond and to protect the service.
Gemini also offers temporary chats. These do not appear in your recent chats or Gemini Apps Activity and are not used to train Google's AI models.
If a conversation has already been reviewed by a person and separated from your account, deleting it from your activity may not delete the reviewed copy. Google says reviewed conversations can be retained for longer than ordinary account activity.
See the Gemini Apps Privacy Hub for the current details and settings.
Claude
For personal Claude accounts, Anthropic says conversations may be used to improve Claude when:
- You allow model improvement in your settings
- You deliberately submit feedback
- A conversation is selected for safety review
- You participate in a testing or research programme
Claude's Incognito chats are not used for model improvement, even if the model-improvement setting is enabled. Some information may still be retained temporarily for safety and service operation.
Anthropic says customer data from its commercial products—such as Claude for Work and its API—is not used to train its models unless the customer explicitly opts into a programme or provides feedback.
Read Anthropic's policy for consumer Claude accounts and its information about commercial products.
Microsoft Copilot
Microsoft's approach varies according to the version of Copilot you use and how you sign in.
Microsoft says users of its consumer Copilot can control whether conversation activity is used for model training. Copilot can also use saved information to personalise future conversations when personalisation is enabled.
Different rules apply to workplace accounts, Microsoft 365 Copilot and some Copilot features built into other Microsoft products. Microsoft says several of these business and organisational uses are excluded from consumer model training.
Because there are several versions of Copilot, check the policy for the exact product and account you are using. Start with Microsoft's Copilot privacy information.
Does paying for an AI service protect my information?
Not necessarily.
A paid personal subscription may give you more features or higher usage limits without changing how your conversations are used.
What usually matters is the type of account, rather than whether money changes hands.
There can be an important difference between:
- A free personal account
- A paid personal account
- A business or workplace account
- An enterprise account
- An education account
- Access provided through an API
Business and enterprise products often provide stronger contractual data protections and do not use customer content for model training by default. However, this is not universal, and your organisation may have its own settings and policies.
A paid consumer subscription should not be assumed to provide the same protection as a commercial or enterprise licence.
Your licence and settings can change the answer
The same AI assistant can handle information differently for different users.
For example, one person might use a personal version of an AI with model improvement enabled. Another might use the same underlying AI through a workplace licence which prevents their content from being used for training.
Your organisation may also control some settings on your behalf.
Before entering work-related or confidential information, check:
- The exact name of the AI product
- Whether you are using a personal or organisational account
- The subscription or licence attached to that account
- Whether model training or improvement is enabled
- Whether memory or personalisation is enabled
- How long conversations are retained
- Whether your employer permits the service to be used for that information
Do not rely on what a policy said when you first created the account. Providers add features, rename plans and update their terms. Review the relevant policy regularly and whenever your account, licence or AI product changes.
What about thumbs-up and thumbs-down feedback?
Feedback buttons deserve particular attention.
When you rate a response, report a problem or send comments, the provider may receive more than the individual answer you selected. It may collect the surrounding conversation so that a person can understand what happened.
In some services, submitting feedback can allow the associated conversation to be used for improvement even when you have disabled general model training.
Before submitting feedback, check whether the conversation contains anything private or confidential.
Are temporary or incognito chats completely private?
No. They are usually more private, but that does not mean no information is retained at all.
Temporary modes commonly prevent a conversation from:
- Appearing in your normal history
- Creating a personal memory
- Being used for general model training
The provider may still retain a copy temporarily for security, abuse prevention, troubleshooting or legal purposes.
Temporary chat is a useful precaution, but it is not a secure vault.
Can another user see what I entered?
Your conversation is not normally shown directly to another random user.
However, you should not treat an AI chat as a private conversation with a trusted person. Depending on the service and circumstances, information might be accessible to:
- Authorised staff or contractors
- Safety and abuse-review teams
- Your organisation's administrators
- Connected apps or services
- Law-enforcement authorities following a valid legal request
- Anyone who gains access to your account or device
There is also a small but important risk of software errors, account compromise or accidental sharing.
What should I avoid entering?
Unless you are certain that the service and your licence are approved for the information, avoid entering:
- Passwords or security codes
- Bank or card details
- Passport, driving-licence or National Insurance numbers
- Private medical information
- Confidential work documents
- Customer or employee information
- Personal information about another person
- Legally privileged material
- Unpublished financial or commercial information
- Private photographs or documents you would not want reviewed
- Information protected by a confidentiality agreement
Remember that removing somebody's name may not be enough to make information anonymous. A combination of details—such as their job, location, age and circumstances—may still identify them.
A safer way to use AI
Before pasting information into an AI, ask yourself:
Would it cause a problem if this information were stored, reviewed by a person or used to improve the service?
If the answer is yes, do not enter it until you have confirmed that the particular product, account and licence are suitable.
Where possible:
- Replace real names with invented ones
- Remove account numbers and identifying details
- Summarise a document instead of uploading the original
- Use temporary or incognito chat
- Turn off model improvement if the service allows it
- Review and delete saved memories
- Delete conversations you no longer need
- Use an approved business service for confidential work
- Check your settings again after major product updates
The simple answer
AI services can keep what you enter. Some can remember details for later conversations, and some may use your content to improve future versions of their models.
Those are separate processes, and each can have its own controls.
The exact answer depends on the product, your settings and your type of account or licence. It can also change when the provider updates its service or terms.
The safest baseline is to assume that anything you enter may be retained, reviewed or used to improve the service unless the current terms for your specific product, plan and settings clearly say otherwise.
Use AI for information you are comfortable sharing. For anything personal, sensitive or confidential, check first.
Related: Back to Safety | How to Use AI Safely
