Which AI chatbot has the best memory?
Only ChatGPT and Claude let you edit a single memory from a list. We read the memory help pages of seven chatbots on 12 October 2026, and five of them draw on your old chats when they answer. Copilot's new app lists what it saved, but you can only delete a line, not change it. Gemini makes you delete the chat itself to remove what it learned. Here's the count.
Claude wins on control. Anthropic made memory free for every Claude user on 2 March 2026, and since 25 August 2026 it's on by default on the Free, Pro and Max plans. Claude saves what it learns as separate topics, listed under Topics in Settings, and you can edit or delete any one of them. It also leaves sensitive subjects alone. Anthropic's memory page says Claude won't store topics "like your health, race, ethnicity, religious beliefs, politics, gender identity" unless you switch on a setting that lets it.
ChatGPT remembers the most. On Plus, at $20 a month, it can draw on your past chats, saved memories, custom instructions, files in your Library and a connected Gmail account. OpenAI started giving Plus and Pro twice as much memory capacity on 4 June 2026, beginning in the US. The free plan gets what OpenAI's pricing page calls limited memory. Go, at $8, gets longer memory, and Pro, from $100, gets the maximum.
Here's the pick by job, from each vendor's own pages in October 2026:
- Best memory for free: Claude. Every plan gets it, and you can edit each topic.
- Most context: ChatGPT Plus or Pro. It reads old chats, files and Gmail.
- Best inside Google's apps: Gemini. It can draw on the Google apps you connect.
- Fewest surprises: Claude again. Sensitive topics stay out until you let them in.
- One memory across many models: Whizi. The same list reaches GPT, Claude and Gemini.
How do ChatGPT, Claude, Gemini, Copilot, Grok and Meta AI memory compare?
The seven chatbots split on two things: whether they read your old chats, and whether you can see a list of what they kept. We compiled the table below from each vendor's help pages and announcements, read on 12 October 2026. "Not documented" means we couldn't find the answer on a page the vendor publishes, not that the feature is missing.
| Chatbot | Free plan | What it remembers | See and edit | Export | On by default |
|---|---|---|---|---|---|
| ChatGPT | Limited on Free; more on Go ($8), Plus ($20) and Pro (from $100) | Saved memories, past chats, custom instructions; Library files and Gmail on some plans | A memory summary you can correct; edit or delete a saved memory | Not named on OpenAI's export page | Past chats switched on automatically outside Europe; opt in across the EEA, UK and Switzerland |
| Claude | Yes, every plan since 2 March 2026 | Topics it saves as you chat, plus a search of past chats | Topics list in Settings; edit or delete each one | Copy it out as text; no file | Yes on Free, Pro and Max; off for Team and Enterprise |
| Gemini | Past chats: 18 or over, personal account, Keep Activity on. Saved facts: a Google AI plan, US only | Past Gemini chats, connected Google apps, your instructions | Instructions list only; delete chats to remove what it learned | Not documented | Yes, per Google's August 2025 post |
| Copilot | Yes, with a personal Microsoft account | Key details and preferences from your chats; Bing, MSN and Edge activity you share | Saved memories list in Settings; delete one or all, no edit | Not documented | Not stated |
| Grok | Not stated by xAI | Your preferences and past conversations | At launch, forget a chat it referenced under a reply | Not documented | Not stated; launched outside the EU and UK in April 2025 |
| Meta AI | Yes, in Facebook, Messenger and WhatsApp | Things you tell it in 1:1 chats, in the US and Canada, per its 2025 post | Delete memories at any time, per the same post | Not documented | Not stated |
| Whizi | Yes, every plan and guests | About 30 short lines of durable facts, capped at 1,000 tokens | Settings lists every line; Clear memory deletes all | A memory field in the JSON account export | Yes, with no switch to turn it off |
Three rows need a caution. The Copilot row is from Microsoft's privacy controls page for the new Copilot app, which arrived on 18 August 2026; the older app's page still describes a single Delete all Memory button. xAI's help pages don't cover memory, so the Grok row pairs its Grok product page with the April 2025 launch post as TechCrunch reported it. Meta's current memory help page wouldn't open for us, so the Meta AI row rests on Meta's January 2025 announcement, updated in July 2025, and the app may have moved past it.
OpenAI's own help page sets the right expectation for all of them: "Memory does not retain every detail from every conversation." Each app picks what to keep. None of them is a full record.
Is there a free AI that remembers conversations?
Yes. Claude is the best free AI that remembers conversations, because its pricing page lists memory on every plan, Free included. You don't need to pay to get the topics list, the past chat search or the edit controls. Anthropic's release notes date the free rollout to 2 March 2026.
ChatGPT's free plan remembers you too, with less room. It's been reading recent conversations for free users since June 2025, and OpenAI's pricing page labels both memory and memory with past chats as limited on Free. If you keep hitting the cap, ChatGPT memory full covers what to delete first.
Gemini splits the feature in two. Google's help page on past chats asks only that you're 18 or over, use a personal Google account and keep Gemini Apps Activity on. But asking Gemini to remember a specific fact needs a Google AI plan, and Google says that part works only in the US for now. Copilot is free with a personal Microsoft account, and so is Meta AI inside Facebook, Messenger and WhatsApp.
Which AI lets you see and edit what it remembers?
Claude shows you the most. Its Settings page lists every topic it keeps, and you can open one, change it and save it. Fix something once and Anthropic says the change applies to every conversation after that. You can also tell Claude in chat what to remember, change or forget.
ChatGPT gives you two views. The memory summary is a short overview you can correct by typing or by highlighting a line. Sources appear under a reply when memory shaped it, so you can see which past chat or saved memory it used. Deleting a chat doesn't delete a saved memory made from it, a trap how ChatGPT memory works explains in full.
Copilot shows a list but stops at delete. In the new app, Settings, Personalization, Manage opens your saved memories, with a Delete icon beside each one and a Delete all memories button, and Microsoft documents no way to edit a line. Gemini gives you no list of what it learned. To remove something, Google tells you to delete every chat that mentions it in Gemini Apps Activity, or to correct Gemini in chat. Grok, at launch, let you forget a past chat it had used, from a small icon under the reply.
How do you stop an AI chatbot from remembering a conversation?
Start a temporary chat before you type. ChatGPT's temporary chats don't create or update memories, and Claude's incognito chat keeps the conversation out of memory and out of your history. Claude also has a softer option: turn Memory off in the + menu before your first message, and the chat stays in your history without feeding memory. Gemini's Temporary Chat stays out of personalization, and Google keeps it for up to 72 hours.
Switching memory off works differently in each app. When you turn off Reference chat history in ChatGPT, OpenAI says what it learned from past chats is scheduled for deletion within 30 days, and logs of deleted saved memories can stay for up to 30 days too. Copilot is the opposite case. Turn off Saved memories in the new app and Microsoft says the memories it already saved stay put until you delete them under Manage. For how long each vendor keeps a deleted chat, see does ChatGPT delete conversations.
Whizi keeps one memory for every model, but it doesn't draw on your old chats
Whizi, which publishes this page, keeps memory as a short list of facts on your account, and that list goes to whichever model you pick, voice included. In Claude, your memory stays in Claude. In Whizi, the same lines reach GPT, Claude and Gemini. It works on every account, the free tier and guests included, and the export in Settings hands the list back as a memory field in one JSON file. It loses on size and control. The list is capped at 1,000 tokens, about 30 lines, while OpenAI says there's no separate storage limit for what ChatGPT can reference from chat history. You can search your old chats yourself, but no model in Whizi reads them when it answers. You can't edit one line or switch memory off, either. Settings shows the list with a single Clear memory button. The Whizi memory docs explain what earns a line.
- Read what your chatbot remembers once: Settings, Personalization, Memory in ChatGPT, or Settings, Memory in Claude
- Use a temporary or incognito chat for anything you don't want saved
- Delete the memory as well as the chat when you want something gone
- Copy your memory out as text before you switch chatbots
Frequently asked questions
Does ChatGPT remember past conversations?
Yes. With Reference chat history on, ChatGPT uses relevant details from past conversations, and saved memories hold facts you asked it to keep. As of October 2026, deleting a chat does not delete a saved memory made from it. How ChatGPT memory works covers both switches.
Can you move your memory from ChatGPT to Claude?
Yes. Claude has an import flow: it gives you a prompt to paste into ChatGPT, then you paste ChatGPT's answer back and click Add to memory. Anthropic says Claude focuses on work topics and may not keep personal details unrelated to work. Switching from ChatGPT to Claude covers the rest of the move.
Is what an AI chatbot remembers used to train it?
It can be. OpenAI says it may use chats and remembered information to improve its models while Improve the model for everyone is on, and you can switch that off in Data controls. Microsoft keeps training as a separate opt out from Copilot memory. Whizi does not use your prompts, conversations or generated content to train Whizi-owned AI models. The AI chatbot privacy comparison lists each default.
Do you have to pay for an AI chatbot to get memory?
Not for Claude. Anthropic made memory free on every Claude plan on 2 March 2026, with the topics list and edit controls included. ChatGPT's free plan gets limited memory, and Plus at $20 a month adds past chats, files and Gmail. If you're weighing a paid Claude plan, is Claude Pro worth it covers what it gives and where it limits you.
