# 0.2.1 Two choices before your first chat (/how-ai-works/your-first-hour/where-it-goes) --- type: Document title: "0.2.1 Two choices before your first chat" description: "What happens to what you type, the two privacy settings you choose before your first message, and what a temporary chat keeps out." status: stable order: 12.1 ksor: owner: team:panaversity audience: [ public ] approval: by: process:panaversity at: 2026-10-29T06:28:52Z chapter: "0.2" part: "0" expert_status: null # as the Part 0 overview: Part 0 is optional on every route, after the skip quiz concepts: [ "0.2.1" ] sources: - id: anthropic-pro-plan resource: https://support.claude.com/en/articles/8325606-what-is-the-pro-plan title: "What is the Pro plan? (Claude Help Center, updated 2026-10-13, verified 2026-10-16)" - id: openai-pricing resource: https://chatgpt.com/pricing title: "Plans and Pricing (ChatGPT, verified 2026-10-16)" last_verified: 2026-10-17 generated: at: 2026-10-29T06:28:52Z by: process:panaversity trust_tier: unverified build_id: sha256:e1fbcb7d9a8f8b9b7cb96875022325c02c9ba4be94cef13b301cbd39a9edad85 dirty: true ksor_version: 0.0.60 --- **In everyday life.** You call a company's help line, and a recorded voice says, "This call may be recorded for training purposes." The person who answers helps you now. The recording is a copy of the call, and it is kept after the call ends. Later, the company may use it to train new staff. Some companies let you say no to the recording. Saying no changes what happens to the copy. It does not change what you should or should not say on the call. **What you will learn.** How to open a free account, and what the two privacy settings decide. How a temporary chat keeps one chat out of your history. **Why it matters.** Both settings act on a copy the app keeps. Set them once, before your first message, and then use the AI without thinking about them again. ## Your free accounts ### How do you open a free account? Go to claude.ai for Claude, or chatgpt.com for ChatGPT, and sign up. Claude asks that you are 18 or older, in a supported location.[^anthropic-pro-plan] ChatGPT's free version is open to everyone.[^openai-pricing] You can also use both on your computer, in their desktop apps, as [0.2.4](desktop-apps.md) shows. A free plan with each AI vendor is enough for every lab in Part 0. It limits how much you can use, and a few controls need a paid plan, as 0.2.3 shows. ## Where what you type goes ### Where does what you type go? It goes two ways. It goes into the context window for this one reply, where the model reads it. And the app saves a copy in your account, as part of the chat. That copy is useful. It waits in your chat list, the list of your past chats, so you can come back to a chat later. It can also come back in front of the model: when you continue the chat, through a memory note, or as a file kept in a project. 0.2.2 shows each one. The copy may also be used to train the AI vendor's later models, if the training setting allows it. Only the window reaches the model you use. The copy in your account reaches it only when the app puts part of the copy back into a window. ![What you type goes two ways, and only the context window reaches the model you use. At the left, a gate reads "Should this go in? A decision, not a setting (Chapter 10)". An arrow runs from the gate to "what you type and attach". From there, one arrow points to "the context window, for this reply", which points to "the model you use", which points to "you". A second arrow points to "your account", which holds "chat history", "files" and "memory notes". From "your account", an arrow labeled "put back into a later window" points up into the context window, another points to "you, later", and a dashed arrow runs through a "training setting" switch to "the AI vendor's later models", drawn apart from the model you use. A banner reads: "The model you use never changes. The settings act on the copy kept in your account."](img/where-it-goes.png) *Figure 0.2.1. What you type goes into this reply's window and into your account. Only the window reaches the model you use.* ### If the model never learns from your chats, why is there a setting for training on them? Because the setting is about the AI vendor's later models, not the model you use. The model you use is frozen, so no chat changes it, as 0.1 showed. But the AI vendor sometimes trains new models. The setting decides whether your kept chats may be part of that training. Even with the setting off, some chats can be used, such as a chat where you rated a reply with a thumbs up or down. Each AI vendor's details are in 0.2.3. ## Two choices, then get to work ### Why make the two choices before your first message? Because the copy is made as you type. Turning training off later covers new chats, and it cannot take back what training has already used. Memory can save a note from your very first chat, too. So before you send anything, choose whether training is on or off, and whether memory is on or off. If you are not sure, turn training off. You can change either one later. Write both choices down, because no chat shows you the training setting. Then the settings are done, and the rest of this chapter is about using the AI. ![The training setting on each app. On the left, Claude's Settings, on its Privacy tab, with the switch "Help improve our AI models" circled. It reads: allow the use of your chats and coding sessions to train and improve Anthropic AI models. On the right, ChatGPT's Settings, under Data controls, with the row "Improve the model for everyone" circled.](img/training.png) *Figure 0.2.2. The training setting on each app: "Help improve our AI models" in Claude's Privacy settings, and "Improve the model for everyone" in ChatGPT's Data controls.* ### What does a temporary chat keep out? It keeps one chat out of your chat list, adds nothing to memory, and is not used for training. You choose it before the chat's first message. It does not make the chat disappear, because the AI vendor still keeps a copy for a set time. ![What each app's temporary chat promises. Above, Claude's incognito chat, with the line circled: incognito chats stay out of your history and search, and aren't used to train Claude. Below, ChatGPT's temporary chat, with the line circled: this chat can reference memory, plugins, and custom instructions, but it won't appear in your history.](img/temporary-chat.png) *Figure 0.2.3. What each app's temporary chat promises. Claude calls it an incognito chat.* **Try it.** Open a temporary chat, send one message, then look for it in your chat list. ### Does any setting decide whether a file may go in? No. The settings decide what the AI vendor may do with your copy. None of them decides whether a file should go in at all. Turning training off does not make real company data safe to put in. A company's rules and the data's owner decide whether a file may go into an AI chat. Chapter 10 teaches that decision. So practice on your own free account with the book's invented Brightline files, which hold nothing real. At work, ask your manager which AI tools and files your company allows. [^anthropic-pro-plan]: What is the Pro plan?, Claude Help Center. [^openai-pricing]: Plans and Pricing, ChatGPT.