The AI Learning Hub Journal

What Happens to What You Type

Where your words actually go1 · YOUYou type a messagequestion, homework, a photo, a secret2 · THE TRIPIt crosses the internetencrypted in transit, not invisible3 · THE PROVIDERTheir servers process itthe model runs, the answer comes back4 · THE COPYA copy may remainon their systemsRETENTIONKept for a period set by policyDays, months, or until you deleteDeleting a chat may not clear backupsHUMAN REVIEWStaff may read sampled conversationsFor safety, abuse and quality checksA small sample — but it could be yoursTRAINING USEMay help train a future modelFree consumer tiers often default to yesSchool and work tiers usually default to noSETTINGS YOU CAN CHANGETurn off chat history and training useDelete conversations and account dataCheck what a school or work account allowsTHE CONTROL THAT ALWAYS WORKSWhat you decide not to type in the first placeNo passwords, addresses or ID numbersNothing about a friend they would not postAssume a human could read it one day — then decide whether to send it
Everything you type leaves your device — settings shrink the footprint, but only what you never type is guaranteed private

Where the Text Goes

When you send a message to a chatbot, it usually leaves your device and is processed on the company's servers. What happens next depends entirely on the product. It may be stored against your account so you can see your history. It may be reviewed by humans for safety or quality. It may be used to train future models, depending on the settings and the plan you are on. Consumer free tiers are generally the most permissive about training on your data; paid and school-provided accounts are often more restrictive. None of this is hidden, but almost nobody reads it, which is how people end up surprised.

  • Your input typically leaves your device and is processed remotely
  • Storage, human review and training use vary by product, plan and settings
  • Free consumer tiers are usually the most permissive; check before assuming
  • Look for a setting like "improve the model for everyone" and decide deliberately

The Practical Rule

You do not need to memorise privacy policies. Use one heuristic: do not type anything you would not be comfortable existing on a company server indefinitely. That covers most real decisions. Specifically, keep out passwords, ID numbers, bank details, your exact address, and anything about your health or your family's. Also keep out other people's information — a classmate's messages, a friend's situation, a private photo — because they never agreed to anything. It is fine to describe a problem in general terms; you rarely need real names and specifics to get a useful answer anyway.

  • Never: passwords, ID or bank numbers, exact address, login codes
  • Careful: health details, family circumstances, anything you would not say publicly
  • Other people's private information is not yours to upload
  • Anonymise — describe the situation without the identifying specifics

Free Products and What You Pay With

Running these models costs real money in hardware and electricity, so a free product is being funded somehow. Sometimes it is a paid tier subsidising you. Sometimes it is data, attention or advertising. Sometimes it is investment money buying growth now for revenue later. None of these are automatically bad, but they tell you where the incentives point. A product funded by your attention will be designed to hold it. A product funded by subscriptions has some reason to actually serve you. Ask this about any tool you are about to hand a lot of your life to, and the surprises get much smaller.

  • Ask how the product makes money and what that incentivises it to do
  • Attention-funded products optimise for time spent, which is not the same as usefulness
  • Check whether the company sells or shares data with third parties
  • A free tool from an unknown company with no clear business model deserves real caution

Prefer slides, quizzes, and saved progress? Read this lesson in the library — free, no sign-up.