It starts innocently. You paste your CV into ChatGPT to polish the wording. You drop a bank statement into Gemini to make sense of the charges. You ask Copilot to summarize a long medical letter you did not fully understand. It feels private — just you and the chat window.
But the chat window is not a private room. On the free personal plans most people use, what you type can be kept, read by human reviewers, and used to improve the company’s AI models — unless you change the settings. That is not a scandal; the companies disclose it. Most people simply never look.
So before your next paste, here is what actually happens to your conversations on the three most-used chatbots, and the settings worth changing.
Your chats travel further than the chat window
Every provider gives you the same basic deal: they store your conversations to give you a working service (chat history, improvements, abuse prevention), and — on consumer plans — they may also use your content to train their models unless you opt out. The details differ, and the differences matter.
ChatGPT. OpenAI says it may use your conversations to train its models on personal plans, and gives you one account-wide switch to stop it: Settings > Data Controls > Improve the model for everyone. Turn it off and new conversations are excluded from training, while your chat history keeps working normally. For one-off sensitive questions there is Temporary Chat: these chats are not saved to your history, do not create memories, are not used for training, and are deleted from OpenAI’s systems after 30 days. Note the distinction — opting out of training is a separate control from Memory, which is ChatGPT’s feature for remembering facts about you across chats. Our guide to how AI chatbot memory works explains what gets stored there and how to see or delete it.
Gemini. Google’s master switch is Gemini Apps Activity, found in your Google Account. When it is on — the default — your conversations are saved and can be used to improve Google’s AI. Google also states that trained human reviewers read some conversations to check quality, and it is unusually direct about this: the privacy notice tells users not to enter confidential information they would not want a reviewer to see. Two retention facts are worth knowing. First, the default auto-delete period for saved activity is 18 months, though you can change it to 3 months, 36 months, or indefinite. Second, chats that were reviewed by humans are kept for up to three years even if you delete your activity — they follow a separate retention path.
Copilot. Microsoft says files you upload to consumer Copilot are stored for up to 30 days and then automatically deleted, and — reassuringly — they are never used to train its generative models regardless of your privacy settings. Images you share with Copilot Vision are processed only to answer your request and are not stored after the session ends. For the conversation text itself, Microsoft offers an opt-out from training in markets where conversation data is used for it, and you can delete conversations from your history at any time.
One honest caveat: defaults and policies shift. The three services above have all changed these settings at least once in the past year or two, which is exactly why this is worth checking rather than assuming.
The details worth keeping out of the chat box
You do not need to treat a chatbot like a hostile stranger. But some information carries outsized risk if it leaks, is misused, or simply sits in a dataset for years. Before pasting, pause on:
- Identity and money documents — passport or ID numbers, full bank statements, credit card numbers, tax returns. A chatbot does not need these to explain how to read a bank statement; a redacted version works fine.
- Health information — full lab reports, diagnoses, medication lists. It is the most sensitive category for most people, and human reviewers may see it.
- Work secrets and internal data — client lists, unreleased plans, credentials. Your personal account is not your employer’s vault (more on this below).
- Other people’s details — your child’s school records, a friend’s legal trouble. You can consent for yourself; you cannot consent for them.
- Photos of documents — Gemini explicitly processes uploaded images, and an image of your passport is still your passport.
A practical habit: replace the real values with placeholders. Ask “help me draft a complaint letter about a £240 overcharge” instead of pasting the full bill. Nine times out of ten, the chatbot answers just as well — and there is nothing to leak.
Three minutes of settings on the services you use
You only need to do this once. Start with whichever chatbot you use most.
On ChatGPT: open Settings > Data Controls and switch off Improve the model for everyone. This stops future conversations from being used for training, and the setting follows your account across web and mobile. For anything especially sensitive, use a Temporary Chat instead of a normal one.
On Gemini: go to myaccount.google.com > Data & privacy > Gemini Apps Activity and turn it off — this stops future chats from being saved, reviewed, or used for training. Then, under the same page, set Auto-delete to 3 months (or delete your existing activity outright) so older conversations do not sit around for the default 18 months.
On Copilot: check your privacy controls for the option to opt out of your interactions being used to train the generative AI models, and delete past conversations you no longer need. Since uploaded files are auto-deleted after 30 days and never used for training, documents are the least of your worries here — it is the chat text that deserves the settings check.
For ChatGPT specifically, one more habit is worth building: a standing rule in Custom Instructions, such as “always warn me if I am about to share sensitive personal information.” The chatbot then reminds you before you paste — a small safety net for the moments you forget.
Work information plays by different rules
The paid business tiers — ChatGPT Team and Enterprise, Gemini under Google Workspace terms, Copilot under a work or school account — generally exclude your content from model training by contract. That sounds safer, and in some ways it is. But it comes with its own trade-off: your employer may be able to audit what you type. So the rule is simple in both directions. Do not paste work secrets into your personal chatbot account, and do not paste personal secrets into your work account. Keep the two worlds in separate windows, on separate accounts, always.
A sane way to think about it
AI chatbots are genuinely useful for drafting, explaining, and summarizing — including with your documents. The privacy risk is not that they are out to get you; it is that convenience makes you forget the chat window is a service with logs, reviewers, and defaults you never chose. Spend three minutes on the settings above, keep the truly sensitive details out (or redacted), and give work data its own separate lane. Then you can enjoy the useful parts without donating your personal life to a training dataset.
Comments 3