When you type a question or upload a document into an AI tool, where does it go? For many people the honest answer is “I am not sure”. Some tools keep your conversations, some use them to improve their products, some let you opt out, and some give business customers different terms from everyone else. The wording is often buried in long policies, and it changes from tool to tool.
This guide helps you find out whether an AI tool is training on your data, and what to do about it. You will learn what “training” actually means, which parts of a privacy policy to read, which settings to look for, how consumer and business plans often differ, and how to decide what you should and should not paste into any tool. This is general information, not legal advice.
Key takeaways
- Training means using your content to improve or build an AI model, which is different from storing your chats so you can see them again.
- Policies, settings and plan type decide what happens to your data, and they can differ between free, personal and business accounts.
- Look for plain statements on training use, retention, human review, deletion and opt-out, and ask the vendor in writing if anything is unclear.
- Opting out of training does not mean your data is not stored, and storage carries its own risks.
- The safest habit is to decide in advance what you will never paste in, and to remove names and identifying details wherever you can.
What training on your data means
The word “training” gets used loosely, so it helps to be precise. Several different things can happen to the content you give an AI tool, and they are not the same.
Training and improving models
Training is the process of building or adjusting an AI model by showing it large amounts of text, images or other content. If a provider uses your conversations or files for this purpose, pieces of your content may become part of the material the model learns from. That does not usually mean the model will repeat your exact words to someone else, but it is a use of your content that you may not want, especially for confidential material.
Storage and history
Many tools save your chats so you can come back to them later. That is storage, and it is a separate matter from training. A tool can store your conversations without training on them, and in principle a tool could use data for training without giving you a visible history.
Human review
Some providers allow staff or contractors to read a sample of conversations to check quality, investigate misuse or improve the product. This can happen even where automated training is switched off, so it is worth looking for in the policy.
Other uses
Data may also be used for safety monitoring, abuse prevention, analytics, billing and legal compliance. Not all of these are the same as training, but each one means someone or something is processing your content.
| Term | What it means | Questions to ask |
|---|---|---|
| Training | Your content helps build or adjust a model | Is it on by default? Can I turn it off? |
| Storage or retention | Your content is kept for a period | How long? Can I delete it? |
| Human review | People may read samples | When, who, and can I opt out? |
| Logging and analytics | Usage details are recorded | Does it include content or only metadata? |
| Sharing with third parties | Data passes to other companies | Which ones and for what purpose? |
Why it matters to you
Whether training is a problem depends on what you put in. Asking for a poem about your cat carries little risk. Pasting a draft contract, a customer list or a staff performance review is another matter.
Confidentiality and trust
If you handle information that belongs to clients or colleagues, you may have duties of confidentiality, whether written into contracts or simply expected. Putting that information into a tool whose terms allow wide use of it may breach those duties. Check your agreements and your employer’s rules.
Personal data
If the content includes information about identifiable people, data protection rules in your country may apply, including rules on who may process the data and for what purpose. The details vary by place. This is general information, not legal advice, so take advice if you handle personal data at scale or in sensitive areas.
Your own work
Writers, designers and consultants may also care about how their own drafts and ideas are used. Read the terms about ownership of what you input and of what the tool produces.
Business risk
For organisations, an unintended leak of strategy, pricing or product plans through casual use of a tool is a real risk. Our guide to protecting company data when using AI tools sets out the wider picture of controls and habits.
The journey of your data
It helps to picture what happens after you press enter. The exact path differs by product, but the stages are broadly similar.
- Input. You type text, upload a file or speak. The content leaves your device and travels to the provider’s servers.
- Processing. The model reads the content and produces a reply. Some tools also pass your input to other services, such as a search function or a plug-in.
- Storage. The conversation may be saved in your account history, and copies may sit in logs or backups for a period.
- Review. Automated systems, and sometimes people, may check samples for safety or quality.
- Possible reuse. Depending on terms and settings, content may be used to improve the product.
- Deletion. When you delete a chat or account, the provider’s policy says how and when the content is actually removed.
Where the risk points are
The riskiest moments are the first and last: what you choose to send, and whether deletion is complete. You control the first directly. The last depends on the provider, so read what the policy promises and over what timeframe.
How to read a privacy policy
Privacy policies are long, but you do not have to read every word. Search for the key terms and read the surrounding paragraphs.
Terms to search for
- “train”, “improve”, “develop” and “machine learning”
- “retain”, “retention”, “store” and “delete”
- “review”, “human”, “contractor” and “quality”
- “third party”, “partner”, “affiliate” and “share”
- “opt out”, “settings” and “controls”
- “business”, “enterprise”, “team” and “API” for differences between plans
Read for specifics, not reassurance
Statements like “we value your privacy” tell you nothing. Look for concrete commitments: what is done, by default or not, and for how long. Phrases such as “may” and “from time to time” leave room for wide use, so note them.
Check more than one document
Data rules often sit across several pages: a privacy policy, terms of use, a data processing addendum for businesses, and help pages about settings. If these disagree, ask the vendor which one applies to you.
Note the date
Policies change. Record the date you reviewed the terms, and check again when you are notified of an update or at least once a year. A setting you chose can also be reset or renamed after a product change, so recheck it occasionally.
| What the policy says | What it may mean | What to do |
|---|---|---|
| “We may use your content to improve our services” | Training use may be on by default | Look for an opt-out setting and the exact scope |
| “We do not use your content to train our models” | A clear commitment, but check which plans it covers | Confirm it applies to your plan and keep a copy |
| “We retain data for as long as necessary” | No fixed limit stated | Ask for specific retention periods |
| “Authorised personnel may review conversations” | Human review is possible | Find out whether you can opt out or limit it |
| “We share data with trusted partners” | Data may leave the company | Ask who the partners are and why |
Settings to look for
Many tools give you controls, but the names and locations vary. These are the usual places and options to check, without assuming any one product has them.
Data controls in your account
Look in settings for sections called privacy, data controls or similar. Common options include an on-off switch for using your content to improve the product, a way to stop saving chat history, an export function and a delete-account option.
Temporary or private modes
Some tools offer a temporary mode where chats are not added to your history. Read what it really does. It may stop the chat appearing in your list while still allowing short-term storage for safety checks.
Workspace and admin settings
Business accounts often put controls in the hands of an administrator, who may set defaults for everyone. If you use a work account, ask your administrator what has been configured instead of assuming.
Settings that are easy to miss
- Separate switches for text, voice and images
- Different rules for files you upload and conversations you type
- Settings that apply to the browser or device and not to the account
- Defaults that apply to shared or public links
- Feedback buttons that send a copy of a conversation for review when you press them
The feedback trap
Clicking a thumbs-up or thumbs-down icon, or submitting a report, may send the conversation to the provider for review even if you have switched off training. Read the small print next to those buttons before using them on sensitive chats.
How plans differ
A common pattern is that terms differ by type of account. This is a general pattern, not a rule, and you must verify each vendor.
| Account type | Typical pattern to check | What to verify |
|---|---|---|
| Free personal | Fewer controls; wider data use may apply by default | Whether training is on and whether you can turn it off |
| Paid personal | Some controls, but terms may still allow some use | Whether paying changes the data terms at all |
| Team or business | Often stronger commitments and admin controls | The written business terms and who administers them |
| Enterprise | Negotiated terms, security reviews, custom retention | The signed agreement, not the marketing page |
| API or developer access | Frequently different terms from the chat product | The specific developer data policy |
Paying does not automatically mean privacy
Do not assume that a paid plan has better data terms. Check explicitly. Some paid personal plans may have the same data use as the free one, with only the features differing. The question to put to the vendor is not “is it paid?” but “what exactly happens to my content on this plan?” Our guide to reading an AI tool’s pricing page shows how to spot what a plan really includes.
Tools built into other products
AI features inside email, documents or design software may follow the host product’s terms, or those of a separate AI provider behind the scenes. Find out which, because the privacy position can differ from the standalone tool.
Red flags and green flags
Red flags
- No clear statement on whether content is used for training
- A privacy policy that cannot be found, or that is unreadable or contradictory
- No way to delete your data or your account
- Vague promises with no timeframes
- Pressure to grant broad permissions to your files or contacts
- No information about who the company is or where it operates
- Terms that claim wide rights over your inputs and outputs
Green flags
- Plain-language statements on training, with a visible control for it
- Stated retention periods and a clear deletion process
- Distinct, published terms for business customers
- Admin controls and clear descriptions of who can see what
- A named contact for privacy questions who actually answers
- Independent security assessments described in specific terms you can ask to see
Questions to ask the vendor
If the documents do not answer your questions, email the vendor. Keep the message short and specific, and save the reply.
- Is content from my account used to train or improve your models, and is that on by default?
- If I opt out, does it apply to all content types, including files and voice?
- How long do you keep my content, and what is the process for deleting it?
- Can people at your company or contractors read my conversations, and under what circumstances?
- Which third parties process my data, and for what purpose?
- Are the terms for my plan different from other plans, and where are they written?
- How will you notify me if these terms change?
- Where is data stored, and can I choose the region?
What to do with the answer
If the reply is clear and consistent with the policy, file it with the date. If it is evasive, or contradicts the documents, treat the tool as unsuitable for anything sensitive. Add these questions to your normal trial checklist whenever you test a new tool, using the method in our guide to choosing an AI tool.
How to reduce your exposure
You cannot fully control what a provider does, but you can control what you send.
Sort your information into levels
| Level | Examples | Guidance |
|---|---|---|
| Public | Published articles, public product details | Generally fine to use in any tool |
| Internal | Meeting notes, draft plans, routine emails | Use approved tools only, with training off where possible |
| Confidential | Client data, contracts, unreleased products, financial detail | Only tools with clear written terms, and minimise detail |
| Restricted | Health records, credentials, government identifiers, anything legally protected | Do not paste into general AI tools |
Anonymise before you paste
Replace names with roles or labels, remove account numbers, strip email addresses and generalise figures where the exact value does not matter. “Customer A complained about a late order” works as well for most tasks as the real name and address.
Never paste secrets
Passwords, access keys and recovery codes should not go into any AI tool. If one is pasted by accident, treat it as exposed and change it.
Use the least data that does the job
Paste the paragraph you need help with, not the whole file. Share the structure of a spreadsheet, not the real rows. This is good practice even where the terms are strong.
Setting rules for a team
If you manage others, personal vigilance is not enough. People differ in caution, and one careless paste can undo careful policy.
Approve tools and accounts
Publish a short list of approved tools and require work to be done on work accounts with the agreed settings. Ban the use of personal free accounts for company information.
Write down what may and may not be shared
Use the levels above, adapted to your business, and give examples people recognise. Short, concrete rules are followed; long abstract ones are not. Our guide to writing an AI use policy for your team gives a structure you can adapt.
Review regularly
Tools and terms change. Put a review date in the calendar, check settings and policies for each approved tool, and update the rules. Remember that checking outputs matters as much as inputs; our guide to checking AI output for mistakes covers that side.
Data check checklist
Before you use a new tool
- Found and read the privacy policy and terms, searching for training, retention and review
- Identified which plan and account type applies to you
- Located the data controls and switched off training where possible
- Checked how to delete chats and your account
- Noted the date reviewed
- Emailed the vendor about anything unclear, and saved the reply
Before every sensitive paste
- Is this public, internal, confidential or restricted?
- Is the tool approved for this level?
- Have I removed names and identifying details?
- Am I sending the minimum needed?
- Is there any password, key or code in this text?
Every few months
- Recheck settings and policy dates
- Delete old chats
- Review connected apps and permissions
- Update team guidance if the tools have changed
Frequently asked questions
If I turn off training, is my data private?
Not necessarily. Turning off training limits one use of your content, but it may still be stored, logged and sometimes reviewed by people. Read the retention and review sections as well, and still avoid pasting anything highly sensitive.
Can an AI repeat my private text to another user?
Providers say models are built to learn general patterns and not to hand out individual inputs, but no policy can promise that nothing could ever leak, and mistakes happen. This is a reason to minimise what you share, not a reason to panic about every chat.
Are business versions always safer?
They often come with stronger terms and admin controls, but you must verify the specifics for each vendor. Compare the written business terms to the consumer ones and ask for the differences in writing.
What about images, voice and files?
These may be handled under different rules from text. Check whether your training opt-out covers them, and consider that images and audio can contain faces, voices, locations and other personal details you did not intend to share.
How do I know if the setting I chose is working?
You usually cannot verify it directly. The practical approach is to confirm the setting is still on after updates, read any notices of change, and rely on the vendor’s written commitments, while sharing less where the stakes are high.
Is it illegal for an AI company to train on my data?
That depends on the terms you accepted, the type of data and the laws where you live. This is general information, not legal advice. If you think your rights are affected, speak to a qualified adviser or the relevant data protection authority.
Conclusion
Whether an AI tool trains on your data is a question you can answer with a little effort: search the policy for the right words, find the settings, check how your plan differs and ask the vendor when the documents are unclear. Remember that training is only one of several uses, and that storage, review and sharing deserve the same scrutiny.
The strongest protection is your own habit. Sort information by sensitivity, strip identifying details, share the minimum and keep secrets out entirely. Pair this guide with protecting company data when using AI tools and how to choose an AI tool, and make data terms part of every decision you make about which tools to use.
