Where does training data come from?
Most of it is the public web. Common Crawl, a non-profit that has been archiving the open web since 2008, is the best-known source, and model makers run their own crawlers as well. On top of that sit books and articles (some licensed through deals with publishers), computer code, transcribed audio, and a smaller amount of carefully made material: example answers and ratings written by people the model maker hires, which teach the model to act like an assistant. A growing share is synthetic: text written by one model to train another.
Training happens before release. A model does not learn from your chat while you type it. Whether your chat becomes training data for a later model is a separate question, and the app's settings answer it.
Could my business's website be in it?
If your site is public and has been online for a while, it is almost certainly in some crawl. A website can ask crawlers to stay away from now on through its robots.txt file. OpenAI's crawler is called GPTBot and Common Crawl's is CCBot, and Google documents a separate token, Google-Extended, that controls use of your pages for its Gemini models without affecting Google Search (Google Search Central, its list of Google crawlers). An opt-out works from the day you add it. It does not pull your pages out of models already trained.
For most small businesses, being read by AI is the point: a model that has never seen your site cannot recommend you. That side of it is covered under answer engine optimisation.
Are my chats used to train the next model?
That depends on the app, the plan and a setting you may never have looked at. Consumer chat apps commonly use conversations to improve future models unless you switch it off; in ChatGPT the switch is "Improve the model for everyone" under Data Controls. Business plans and developer access from the major providers generally leave your data out of training by default, and their privacy pages say so. Read the page for the tool you use, because the settings change.
The Office of the Australian Information Commissioner's October 2024 guidance on commercially available AI products is plain about the safe default: as a matter of best practice, keep personal information, and sensitive information above all, out of publicly available generative AI tools.
What should a landscaper paste to write a win-back email? (example figures)
A landscaper in Lismore has 1,400 past customers in a spreadsheet, and 300 of them have not booked in two years. The average job is $2,200. The owner wants an email to win some of them back with an autumn garden tidy-up from $450.
| Give the AI | Keep out of the chat |
|---|---|
| The offer: autumn tidy-up from $450, booked by 30 April | Names, emails and phone numbers |
| The kinds of jobs those customers had: hedging, mulching, new turf | Street addresses |
| The tone: friendly, local, no hard sell | Notes about complaints or late payers |
| A past email that got bookings | The spreadsheet itself |
The AI writes the email from the left column. The names go in afterwards, through the merge field in the email tool, so no customer's details pass through the chat at all. If 1 in 20 of the 300 rebook at the average job, that is 15 jobs and $33,000 of work, and the privacy step cost nothing.
What does CoraVision keep from your sessions?
CoraVision keeps the transcripts of your calls and chats with your account, and the still images Cora sees of your screen are not saved to your account. Cora sees your screen only during a session you start, as still images rather than recordings. Once you connect Google Ads, Meta and Google Analytics 4, CoraVision reads them nightly and changes nothing in them. CoraVision also publishes an AI safety page at coravision.ai/ai-safety.
Questions
Can my data be removed from a model once it is trained?
Not in any practical way for one person. What a model learned is spread through billions of values rather than stored as a copy of your text. What you can do is delete your chat history and account data with the provider, and switch off training use from now on.
Does paying for an AI app keep my chats out of training?
Not automatically. Some paid personal plans still use chats for training unless you turn the setting off, while business plans from the major providers usually exclude your data by default. Check the data controls for your own plan rather than assuming.