{"course":{"slug":"choosing-your-tools","title":"Choosing and Using the Tools","url":"https://addaly.com/learn/choosing-your-tools","level":"beginner","track":"work"},"lesson":{"slug":"who-reads-this","title":"Who reads what you type","url":"https://addaly.com/learn/choosing-your-tools/who-reads-this","position":48,"of":76,"minutes":8,"answer":"Not training on your data and not storing it are different promises, and most providers only make the first.","keyIdea":"Not training on your data and not storing it are different promises, and most providers only make the first."},"body":{"format":"markdown","text":"## Three questions, asked of every tool\n\n1. Is my input used to train the model?\n2. Is it stored, and for how long?\n3. Can a human being read it?\n\nThese are separate questions with separate answers, and companies answer the first one loudly and the other two quietly. \"We do not train on your data\" is not \"we do not keep your data.\" A provider can truthfully say the first while retaining your conversations for thirty days and allowing staff to review flagged ones.\n\n## The pattern, as of 2026\n\nThis is the shape it has held for a while. Verify it for your specific tool, because it changes.\n\n**Consumer free tiers.** Frequently used for training by default. Usually there is a switch in settings, and it is usually not the first thing you see. In some jurisdictions the default is reversed by law.\n\n> Figure: Three separate promises, and only the first is made loudly. Layers, top to bottom: We do not train on your content (Commonly stated and easy to find. It says nothing about the two below it.); We do not store your content (A different promise. Thirty days for abuse monitoring is common, and deletion carve-outs for backups and legal obligations are close to universal.); No person will ever read it (Almost nobody promises this. Sampled and flagged conversations are reviewed by people, which is how abuse gets caught.). A provider can truthfully make the first while doing both of the others. Check which one the sentence you found actually says, and whether it covers your tier: free, paid consumer, business and API are frequently treated differently by the same company.\n\n**Consumer paid tiers.** Varies by company, and this surprises people. Paying does not automatically opt you out. Several products have trained on paying subscribers' conversations, with a setting available to decline.\n\n**Business, team and enterprise tiers.** Generally covered by a contract that says your content is not used for training. If you handle other people's information for a living, this is what you are actually buying.\n\n**API access.** Across the major providers, API inputs are generally not used for training by default. Retention for abuse monitoring is common, often around thirty days, and zero-retention arrangements exist for customers who ask and qualify.\n\n**Local models.** Nothing leaves your machine, provided the app around it is not sending telemetry. Check the settings once.\n\n## The sentence people miss\n\nSomewhere in most privacy policies is a line saying that a sample of conversations, or conversations flagged by automated systems, may be reviewed by human beings for safety and quality.\n\nThis is not sinister: it is how abuse gets caught and how models get fixed. But it means the honest mental model is not \"a machine reads this and forgets.\" It is \"a machine reads this, and occasionally a person might.\"\n\n## How to check, in five minutes\n\n- Open **Settings**. Look for **Data controls**, **Privacy**, or **Improve the model**. Read every toggle, including the ones already on.\n- Open the privacy policy and use your browser's find function. Search for: *train*, *improve our models*, *human review*, *retention*, *delete*.\n- For a paid work tool, find the trust or security page rather than the consumer policy. They often say different things.\n- Take a screenshot with the date visible. Terms change, and a dated screenshot is the only record you will have of what you agreed to.\n\nIf you cannot find a clear answer in five minutes, treat that as the answer.\n\n## Rules that do not depend on trusting anyone\n\n**Do not paste what you could not email.** Passwords, API keys, a customer database, unredacted medical or legal files. Not because a leak is likely, but because the cost of being wrong is far larger than the effort of not doing it.\n\n**Redact by role.** The model does not need the name. \"The patient,\" \"the supplier,\" \"the employee\" produces the same quality of answer as \"Mrs Adeyemi.\" This one habit removes most of the risk from most of the work.\n\n**Remember deletion has limits.** You can usually delete a conversation. If data has already gone into a training run, it cannot be extracted from the trained model afterwards. Deletion going forward is real. Deletion going backwards is not.\n\n**Check whether it is even your decision.** Your employer may have a policy. Your country may have a law: GDPR in Europe, the DPDP Act in India, LGPD in Brazil, POPIA in South Africa. Handling someone else's personal data usually shifts this from a preference to an obligation.\n\nNone of this requires paranoia. It requires knowing which of three promises a company is making, and reading the one sentence where they say it."},"check":{"question":"Miguel upgrades to a paid consumer plan specifically so his conversations will not be used to train the model. Is that a safe assumption?","options":[{"text":"Yes. Payment creates a commercial relationship, and providers do not train on paying customers' content.","correct":false,"tempting":"Intuitive, but several products have trained on paid consumer conversations by default; business tiers and APIs are what reliably change it."},{"text":"No. Some consumer plans train by default whether or not you pay; a settings toggle or a business tier is what changes it.","correct":true},{"text":"Yes, as long as he also turns off chat history, because history and training are the same control.","correct":false,"tempting":"They are usually separate settings, and turning off history may not stop training."},{"text":"No, but it does not matter, because deleting a conversation removes it from the model.","correct":false,"tempting":"Data already absorbed into a training run cannot be pulled back out of a trained model."}]},"prev":{"slug":"the-hybrid-setup","title":"A working setup with two models","url":"https://addaly.com/learn/choosing-your-tools/the-hybrid-setup"},"next":{"slug":"reading-terms-in-ten-minutes","title":"Reading the terms in ten minutes","url":"https://addaly.com/learn/choosing-your-tools/reading-terms-in-ten-minutes"},"authorship":"Founder and series editor: Dr Mustafa Mun. Written by AI models to a written standard. Exam keys checked blind by a second model: 3,665 of 3,666 agreed. Academic review invited.","review":{"status":"Draft, open for academic review"},"cite":"Addaly (2026) 'Who reads what you type', in Choosing and Using the Tools. Edited by M. Mun. Addaly. Available at: https://addaly.com/learn/choosing-your-tools/who-reads-this (Accessed: 31 October 2026).","licence":{"name":"CC BY-NC-SA 4.0","url":"https://creativecommons.org/licenses/by-nc-sa/4.0/"},"reader":{"schema":1,"title":"Who reads what you type","courseTitle":"Choosing and Using the Tools","module":{"index":6,"slug":"data-terms-and-law","title":"Data, terms and the law"},"answer":null,"keyIdea":"Not training on your data and not storing it are different promises, and most providers only make the first.","hero":{"kind":"figure","figure":{"t":"figure","label":"Figure 6.1","figure":{"type":"stack","title":"Three separate promises, and only the first is made loudly","layers":[{"label":"We do not train on your content","detail":"Commonly stated and easy to find. It says nothing about the two below it."},{"label":"We do not store your content","detail":"A different promise. Thirty days for abuse monitoring is common, and deletion carve-outs for backups and legal obligations are close to universal."},{"label":"No person will ever read it","detail":"Almost nobody promises this. Sampled and flagged conversations are reviewed by people, which is how abuse gets caught."}],"caption":"A provider can truthfully make the first while doing both of the others. Check which one the sentence you found actually says, and whether it covers your tier: free, paid consumer, business and API are frequently treated differently by the same company."},"alt":"Figure: Three separate promises, and only the first is made loudly. Layers, top to bottom: We do not train on your content (Commonly stated and easy to find. It says nothing about the two below it.); We do not store your content (A different promise. Thirty days for abuse monitoring is common, and deletion carve-outs for backups and legal obligations are close to universal.); No person will ever read it (Almost nobody promises this. Sampled and flagged conversations are reviewed by people, which is how abuse gets caught.). A provider can truthfully make the first while doing both of the others. Check which one the sentence you found actually says, and whether it covers your tier: free, paid consumer, business and API are frequently treated differently by the same company."}},"blocks":[{"t":"h2","id":"three-questions-asked-of-every-tool","runs":[{"text":"Three questions, asked of every tool"}]},{"t":"ol","start":1,"items":[[{"text":"Is my input used to train the model?"}],[{"text":"Is it stored, and for how long?"}],[{"text":"Can a human being read it?"}]]},{"t":"p","runs":[{"text":"These are separate questions with separate answers, and companies answer the first one loudly and the other two quietly. “We do not train on your data” is not “we do not keep your data.” A provider can truthfully say the first while retaining your conversations for thirty days and allowing staff to review flagged ones."}]},{"t":"h2","id":"the-pattern-as-of-2026","runs":[{"text":"The pattern, as of 2026"}]},{"t":"p","runs":[{"text":"This is the shape it has held for a while. Verify it for your specific tool, because it changes."}]},{"t":"p","runs":[{"text":"Consumer free tiers.","b":true},{"text":" Frequently used for training by default. Usually there is a switch in settings, and it is usually not the first thing you see. In some jurisdictions the default is reversed by law."}]},{"t":"p","runs":[{"text":"Consumer paid tiers.","b":true},{"text":" Varies by company, and this surprises people. Paying does not automatically opt you out. Several products have trained on paying subscribers’ conversations, with a setting available to decline."}]},{"t":"p","runs":[{"text":"Business, team and enterprise tiers.","b":true},{"text":" Generally covered by a contract that says your content is not used for training. If you handle other people’s information for a living, this is what you are actually buying."}]},{"t":"p","runs":[{"text":"API access.","b":true},{"text":" Across the major providers, API inputs are generally not used for training by default. Retention for abuse monitoring is common, often around thirty days, and zero-retention arrangements exist for customers who ask and qualify."}]},{"t":"p","runs":[{"text":"Local models.","b":true},{"text":" Nothing leaves your machine, provided the app around it is not sending telemetry. Check the settings once."}]},{"t":"h2","id":"the-sentence-people-miss","runs":[{"text":"The sentence people miss"}]},{"t":"p","runs":[{"text":"Somewhere in most privacy policies is a line saying that a sample of conversations, or conversations flagged by automated systems, may be reviewed by human beings for safety and quality."}]},{"t":"p","runs":[{"text":"This is not sinister: it is how abuse gets caught and how models get fixed. But it means the honest mental model is not “a machine reads this and forgets.” It is “a machine reads this, and occasionally a person might.”"}]},{"t":"h2","id":"how-to-check-in-five-minutes","runs":[{"text":"How to check, in five minutes"}]},{"t":"ul","items":[[{"text":"Open "},{"text":"Settings","b":true},{"text":". Look for "},{"text":"Data controls","b":true},{"text":", "},{"text":"Privacy","b":true},{"text":", or "},{"text":"Improve the model","b":true},{"text":". Read every toggle, including the ones already on."}],[{"text":"Open the privacy policy and use your browser’s find function. Search for: "},{"text":"train","i":true},{"text":", "},{"text":"improve our models","i":true},{"text":", "},{"text":"human review","i":true},{"text":", "},{"text":"retention","i":true},{"text":", "},{"text":"delete","i":true},{"text":"."}],[{"text":"For a paid work tool, find the trust or security page rather than the consumer policy. They often say different things."}],[{"text":"Take a screenshot with the date visible. Terms change, and a dated screenshot is the only record you will have of what you agreed to."}]]},{"t":"p","runs":[{"text":"If you cannot find a clear answer in five minutes, treat that as the answer."}]},{"t":"h2","id":"rules-that-do-not-depend-on-trusting-anyone","runs":[{"text":"Rules that do not depend on trusting anyone"}]},{"t":"p","runs":[{"text":"Do not paste what you could not email.","b":true},{"text":" Passwords, API keys, a customer database, unredacted medical or legal files. Not because a leak is likely, but because the cost of being wrong is far larger than the effort of not doing it."}]},{"t":"p","runs":[{"text":"Redact by role.","b":true},{"text":" The model does not need the name. “The patient,” “the supplier,” “the employee” produces the same quality of answer as “Mrs Adeyemi.” This one habit removes most of the risk from most of the work."}]},{"t":"p","runs":[{"text":"Remember deletion has limits.","b":true},{"text":" You can usually delete a conversation. If data has already gone into a training run, it cannot be extracted from the trained model afterwards. Deletion going forward is real. Deletion going backwards is not."}]},{"t":"p","runs":[{"text":"Check whether it is even your decision.","b":true},{"text":" Your employer may have a policy. Your country may have a law: GDPR in Europe, the DPDP Act in India, LGPD in Brazil, POPIA in South Africa. Handling someone else’s personal data usually shifts this from a preference to an obligation."}]},{"t":"p","runs":[{"text":"None of this requires paranoia. It requires knowing which of three promises a company is making, and reading the one sentence where they say it."}]}],"check":{"question":[{"text":"Miguel upgrades to a paid consumer plan specifically so his conversations will not be used to train the model. Is that a safe assumption?"}],"options":[{"text":[{"text":"Yes. Payment creates a commercial relationship, and providers do not train on paying customers’ content."}],"correct":false,"tempting":[{"text":"Intuitive, but several products have trained on paid consumer conversations by default; business tiers and APIs are what reliably change it."}]},{"text":[{"text":"No. Some consumer plans train by default whether or not you pay; a settings toggle or a business tier is what changes it."}],"correct":true,"tempting":null},{"text":[{"text":"Yes, as long as he also turns off chat history, because history and training are the same control."}],"correct":false,"tempting":[{"text":"They are usually separate settings, and turning off history may not stop training."}]},{"text":[{"text":"No, but it does not matter, because deleting a conversation removes it from the model."}],"correct":false,"tempting":[{"text":"Data already absorbed into a training run cannot be pulled back out of a trained model."}]}],"order":[3,2,1,0],"why":null},"prev":{"slug":"the-hybrid-setup","title":"A working setup with two models","minutes":8},"next":{"slug":"reading-terms-in-ten-minutes","title":"Reading the terms in ten minutes","minutes":8}}}