# What can AI see and remember? It depends on the product and what you've turned on. What it usually sees, what memory means, and where to check yours. **Time:** 6 minutes **What you'll have when you're done:** A clear picture of the three things that decide what an AI app can see, and where to look in your own app's settings. ---There isn't one answer, because it depends on which product you use, which plan you're on, and which features are switched on. What there is: a small set of questions that tell you the answer for your app. --- ## Layer 1: the conversation In an ordinary chat, **it works with what's in that conversation**: what you type, and anything you paste, upload, or share into it. Within one conversation, it can refer back to earlier messages. In a very long one, it may lose track of details from the start, which [Why it forgot what you told it](/guides/foundations/why-it-forgot-what-you-told-it/) covers. A new conversation often starts from a blank page, unless one of the next two layers is turned on. --- ## Layer 2: memory and saved instructions Some AI apps offer features that carry information from one conversation to the next: - **Saved instructions** you write yourself, like "I run a small bakery" or "keep answers short" - **Memory**, where the app saves details from your conversations to use later - **Access to past conversations**, where it can look back at earlier chats Whether these exist, whether they're on by default, and what they save varies by product and plan, and changes over time. If an app brings up something you told it last week, one of these is probably on. Where an app offers memory, its settings usually let you see what it saved and turn the feature off. --- ## Layer 3: connections Some AI apps can connect to other places your information lives: email, calendar, cloud files, or a company's internal systems. **Once connected, it may be able to read what's there when it's relevant to your request,** and sometimes take actions. Connections are usually set up on purpose, by you or your organization, often through a screen that asks you to approve access. If you don't remember granting any, check anyway. On a work account, an administrator may have connected things for you. --- ## What it can't see Without a connection or a feature you've turned on, a chat app generally can't see: - Other apps open on your screen - Files on your computer or phone that you haven't uploaded - Your email, messages, or accounts - What you did somewhere else Some computer and phone apps do offer features that can look at your screen or work with another app. Those usually ask for your permission first, so they're worth knowing about rather than worrying about. --- ## Seeing is different from keeping Two separate questions: **What can the AI use while it helps you?** That's the three layers above. **What does the company store, and for how long?** That's set by the product's privacy terms and your settings, not by what the AI appears to remember. A conversation the AI never mentions again, or one you've deleted from your screen, may still be kept for a period under the company's terms. Some products also have a setting about whether your conversations can be used to improve their AI. Personal and work accounts often have different rules. [The plain answers](/plain-answers/) cover where what you type goes. --- ## Something you can try **About five minutes.** Open the AI app you use most and look for three things in its settings. The names vary, so look for words like these: 1. **Memory or personalization.** Is it on? Can you see what's saved? 2. **Connected apps, connectors, or integrations.** Is anything connected? 3. **Data controls or privacy.** Is there a setting about using your conversations to improve the product? Then ask it: ``` What do you know about me from outside this conversation? List anything you've saved or can access, and tell me where I can see or change it. ``` Compare the answer with what the settings show. If they disagree, trust the settings. Where to look in the four apps on the [Compare tools](/tools/) page. These were checked against each company's help pages on September 14, 2026. Menus move, so if yours looks different, use the linked help page.
Go to **Settings**, then **Personalization**, then **Memory**. Memory is limited on the free plan. OpenAI says the controls for deleting memories and turning memory off are on the website now and coming to the phone app. [OpenAI's memory help page](https://help.openai.com/en/articles/8590148)
Go to **Settings**, then **Memory**. You can see and edit what it remembers, pause memory, or reset it. Incognito chats aren't remembered. [Anthropic's memory help page](https://support.claude.com/en/articles/11817273)
Memory of past chats is only for adults with personal Google accounts. To turn it off, open **Settings & help**, then **Personal Intelligence**. To see or delete past chats, open **Gemini Apps Activity**. [Google's help page](https://support.google.com/gemini/answer/16598469)
Memory only works when you're signed in. Go to **Settings**, then **Personalization**. The **Saved memories** switch turns it off, and **Manage** deletes what it saved. Turning it off doesn't delete what's already saved. [Microsoft's privacy controls page](https://support.microsoft.com/en-us/privacy/microsoft-copilot/privacy-controls)
--- ## Something important to know **Assume what you type is sent to and kept by the company, even if the AI never brings it up again.** Not because someone is necessarily reading it. Because that assumption stops you from sharing something you'd regret. The next page is about exactly that. --- ## If it doesn't match what you see **It knew something you didn't tell it in this conversation.** Memory, saved instructions, past-conversation access, or a connection is on. Check the settings above. **It doesn't remember something from last week.** Memory may be off, may not have saved that detail, or may not be available on your plan. Paste the important details into the new conversation. **It says it can see your email or files, and nothing is connected.** It may be wrong about its own abilities. Ask it to quote something specific that you can confirm. If it can't, it can't see it.

Next