The best chatbot for memory in 2026 is ChatGPT or Claude for most people. Both save facts about you, search your past chats, and let you see and edit what they keep, on free and paid plans. Gemini and Copilot remember too, with fewer controls. Character.AI is the pick for roleplay, but its memory is small and mostly manual.
What “memory” means in a chatbot
Chatbot memory is the ability to use information from earlier conversations in a new one. Without it, every chat starts blank. With it, the app stores facts you shared (your job, diet, a project) or searches old chats, then adds the relevant bits to the model’s input before it answers.
There are two common designs:
- Saved facts. The app writes short notes like “is vegetarian” or “works in Python.” You can usually see and delete them one by one.
- Chat history reference. The app searches past conversations when a new question seems related. This covers more ground but is harder to inspect.
Most big apps now do both. The differences are in what you can see, what’s free, and how much control you get.
Best chatbot for memory: comparison table
This table reflects each vendor’s help pages as of October 2026. Features roll out by plan and country, so your account may differ.
| App | Saves facts | Uses past chats | View and edit memory | Free plan | Off switch for one chat |
|---|---|---|---|---|---|
| ChatGPT | Yes (memory summary or saved memories) | Yes (“Reference chat history”) | Memory summary, Sources under answers, Manage | Yes | Temporary Chat |
| Claude | Yes (topics) | Yes (chat search) | Settings > Memory > Topics | Yes, since March 2026 | Incognito chat, or Memory off in “+” menu |
| Gemini | Yes (paid Google AI plan, US) | Yes (18+, Keep Activity on) | Memory summary (US, paid plan); otherwise ask Gemini, delete chats in Activity | Past chats: yes | Delete the chat afterward |
| Microsoft Copilot | Yes | Recent conversations | Ask “What do you know about me?”, Delete all Memory | Yes | Not documented |
| Character.AI | Pins and a 400-char chat memory | Within the chat’s context only | Per chat | Yes (15 pins) | n/a |
Sources: OpenAI Memory FAQ, Claude memory help, Gemini past chats help, Copilot privacy controls, Character.AI plan comparison.
ChatGPT: the most sources of memory
ChatGPT can draw on past chats, saved memories, custom instructions, files in your Library (now Space on some paid plans) and, on some plans, connected apps such as Gmail, according to OpenAI’s Memory FAQ. In June 2026 OpenAI switched Plus and Pro users in the US to an “improved memory” that keeps a running memory summary instead of a flat list. The rollout note says those plans also get twice the memory capacity, with Free and Go to follow.
What makes it strong:
- Sources appear under some answers, showing which past chat or memory shaped the reply. You can delete or correct them from there.
- Temporary Chat gives you a chat that doesn’t create or update memories. Since August 2026 it can still read your memory if you pick Personalized; locked chats (October 2026) are another twist, covered in what changed in ChatGPT memory this fall.
- Project-only memory keeps a project’s context from leaking into other chats.
Free users have had saved memories plus references to recent chats since June 2025, per the ChatGPT release notes. For the step-by-step on viewing and deleting it, see how to find and manage ChatGPT memory.
Claude: the clearest controls
Claude saves memory as individual topics while you chat, rather than summarizing a chat after it ends. You can open Settings > Your Claude (Settings > Memory in the older layout), read each topic, edit it, or delete it. Each project has its own memory, but Anthropic’s help center (checked 1 November 2026) says Claude also uses your regular-chat memory inside projects and can reference project memory outside them. To keep work and personal context fully apart, turn off Use account memory on the project’s page; the Claude memory guide lists what else changed.
Two details stand out in Anthropic’s help article:
- By default Claude doesn’t store sensitive topics like health, religion or politics. You opt in if you want them kept.
- You can turn memory off for a single chat from the “+” menu, and that chat still stays in your history. Incognito chats skip both memory and history.
Anthropic brought memory to the free plan on March 2, 2026, Engadget reported. Claude also supports importing memory from other assistants, which helps if you’re switching from ChatGPT. More on how it works in Claude’s long-term memory.
Gemini and Copilot: memory with fewer knobs
Gemini can personalize answers from your past chats if you’re 18 or over, signed in with a personal Google Account, and have Keep Activity on. You switch it under Settings & help > Personal Intelligence. Asking Gemini to remember a specific fact (“Remember that I am vegetarian”) needs a paid Google AI plan and is US-only for now, per Google’s help page. That page also says saved items are removed after at most 18 months. To make Gemini forget something, you delete the chats that contain it in Gemini Apps Activity. US users on a paid plan now also get a Memory summary page (Settings > Personal Intelligence) that shows what Gemini has learned, edited by prompting, with a change history; see Gemini’s Memory summary. Everyone else still has no list of saved facts to prune. Switching from ChatGPT or Claude? Gemini can import your memories and chat history; see how Gemini’s memory and import work. Gemini’s new skills (September 2026) are saved task instructions, not memory; the difference is explained in Gemini memory vs skills.
Microsoft Copilot, signed in with a personal Microsoft account, remembers details you share. You see them by asking “What do you know about me?” and remove one by asking it to forget. Profile > Memory > Delete all Memory wipes everything without deleting your chat history, per Microsoft’s privacy controls page. Microsoft keeps conversation history for 18 months unless you delete it.
Character.AI: best for roleplay, small memory
Character.AI is built for long character chats, but its memory tools are manual. Free users get 15 pinned messages per chat and a 400-character Chat Memories box. The paid lite plan raises pins to 30, and c.ai+ adds auto-memories and private Lorebooks, according to the Character.AI help center.
Anything not pinned, saved or recent can drop out of what the character “sees.” The full breakdown is in Character.AI’s memory limit explained. For story games, AI Dungeon does more on its own: it summarizes every six actions and searches up to 100 stored memories on the free tier (800 on Ultimate), as covered in how the AI Dungeon memory system works.
How to pick the right chatbot for memory
- Decide what you need remembered. Personal preferences need saved facts. Long projects need chat search or project memory.
- Check your plan and country. Gemini’s fact memory is US and paid only; ChatGPT’s newest memory reached Plus and Pro first.
- Look for a visible memory list. Claude’s topics and ChatGPT’s summary let you audit what’s stored. Gemini’s Memory summary does the same but only in the US on a paid plan; Copilot mostly relies on asking the bot.
- Check the per-chat off switch. Temporary or incognito chats matter if you sometimes ask things you don’t want stored.
- Use projects for long work. ChatGPT’s project-only memory keeps a project sealed off; in Claude, switch off Use account memory on the project to get the same.
- Put critical facts in instructions. Custom instructions are always applied; memory is applied when the app thinks it’s relevant.
- Test it. Tell it a fact, open a new chat a day later, and ask.
No app remembers everything. OpenAI says ChatGPT’s memory “does not retain every detail from every conversation,” and a 2024 benchmark found commercial chat assistants lost about 30% accuracy when recalling details across long interactions (LongMemEval, arXiv). Why that happens is covered in why AI memory is so bad.
If you’re building your own chatbot with memory
Consumer apps don’t let you change how memory works. If you’re a developer building a bot, you add memory yourself: store facts or past messages, then retrieve the relevant ones before each model call (how AI memory works compares how Mem0, Zep, Letta and Hindsight run each step). Open-source options include Mem0, Letta, Graphiti (the open-source core of Zep) and Hindsight, an MIT-licensed memory server with retain, recall and reflect operations. Hindsight reports 94.6% on LongMemEval, the benchmark above where commercial assistants struggled (vendor-run), though it adds an LLM call to every write; the agent memory leaderboard lists the published scores of each system side by side. The tradeoffs are covered in how to add memory to a chatbot, and how to give AI agents memory has tested Python examples plus a recall test built on the same LongMemEval question types. If you’d rather self-host a ready-made chat app, Open WebUI and LibreChat both have per-user memory; the chatbot memory repos on GitHub page compares them with license and setup notes. For a low-code build, Langflow’s Agent component keeps chat history per session and its memory bases add recall across sessions; see the Langflow memory chatbot guide.
Agent apps are a different case from chat apps. xAI’s Grok Bot, for example, keeps learned memory per Bot rather than per chat, and plugins can add memory shared across Bots; see Grok Bot memory. If you build on LangChain and are weighing a stateful agent runtime instead, Letta vs LangChain memory compares the two.