AI Memory Is Here: How ChatGPT and Claude Now Remember You (And Why It Changes Your Workflow)

What AI memory actually means (and what it doesn’t)

If you’ve used ChatGPT or Claude for any length of time, you’ll know the frustration. You open a new conversation, and the AI has no idea who you are. You’re a marketing director at a mid-size B2B company who prefers concise bullet points, writes for an audience of CFOs, and always needs to reference your company’s product line. But none of that travels with you. Every session, you’re back to square one.

AI memory is the fix for exactly this. It’s a feature that lets the AI retain information between separate conversations, building a profile of you over time. Think of it less like the AI “learning” in a deep technical sense, and more like the AI keeping notes on you. When you start a new chat, it checks those notes and adjusts its responses accordingly.

What it doesn’t do: it’s not magic. It won’t read your mind. It won’t pull in data from your email or company systems (unless you’ve connected specific integrations). It’s simply carrying forward the things you’ve told it, or that it has observed about your preferences from past conversations.

How Claude’s memory works (March 2026 rollout)

Anthropic launched Claude’s persistent memory feature for paid subscribers in late 2025, then extended it to all users including the free tier on March 2, 2026. [1] That’s significant because it means you don’t need a subscription to use it.

Here’s how it actually works: Claude automatically generates a memory summary from your conversations and stores it. Over time, it builds a picture of how you work, what you care about, and how you communicate. You can also manually tell Claude things to remember using natural language. Just say “remember that I’m a project manager at a logistics company and I need summaries in plain English,” and it stores that explicitly.

Claude’s memory has a three-layer structure, though you don’t need to understand the technical details to use it well:

• Conversation memory: The context within your current chat session.
• Persistent memory: What carries between sessions. This is the new bit.
• Project memory: Specific context you can pin to ongoing projects (most useful for people doing repeated work in the same area).

You can view, edit, or delete everything Claude has stored about you by going to your account settings. Anthropic has made the controls genuinely accessible, not buried in a menu somewhere. [2]

One useful development worth noting: Google Gemini launched a feature in March 2026 that lets you import your conversation history and memory from Claude and ChatGPT. [3] So if you’re switching platforms, your preferences don’t have to start from scratch.

How ChatGPT’s memory works

ChatGPT has had some form of memory since 2024, but it’s matured significantly. The current version builds a persistent profile that grows with use. OpenAI’s approach is slightly more automatic than Claude’s: it observes patterns in your conversations and surfaces things it thinks are worth remembering, rather than waiting for you to instruct it.

This is both a feature and a limitation. On the one hand, it means you don’t have to actively manage it as much. On the other hand, it can store things you didn’t intend, so checking what it’s accumulated is worth doing occasionally.

You can see your stored memories by going to Settings, then Personalization, then Memory. You can delete individual items or clear everything. If you want to make sure something gets stored explicitly, just tell it: “remember that I always want responses in UK English” and it will add that to your profile.

The most practical recent addition is the “memory summary” feature, which lets you see a readable digest of everything the AI has retained about you. It’s surprisingly useful for catching outdated assumptions. If ChatGPT still thinks you’re working on a project you wrapped up six months ago, that’s worth knowing.

How to set it up so it actually helps your workflow

The biggest mistake people make with AI memory is passive setup. They turn the feature on, let the AI figure out what to remember, and then wonder why the output doesn’t feel more personalised. The better approach is intentional. Here’s a practical setup process for any professional.

Step 1: Tell it your professional context in one clear paragraph. Don’t wait for the AI to figure this out over time. Open a new chat and type something like: “I want you to remember this about me for all future conversations. I’m a [job title] at a [company type]. I work primarily on [main responsibilities]. My audience when I’m writing is usually [describe them]. I prefer responses that are [concise/detailed/structured/conversational]. I need to avoid [jargon/technical language/etc.]”

Step 2: Add your preferred formats. If you always want bullet points for summaries, always want a specific section order for reports, or always need a specific reading level, say so explicitly. “Always give me summaries as numbered bullet points, not paragraphs.”

Step 3: Add recurring context it would otherwise need to ask you. Things like: the name of your company, your main product or service, the stage of a project you’re working on, or the names of people you frequently need to write about or to.

Step 4: Check and refine every few weeks. Memory can accumulate outdated information. A quick audit every month keeps it useful and prevents the AI from applying context that no longer applies.

What to actually put in your memory profile

Most people under-use memory by storing generic things. “I’m a marketer. I like short answers.” That’s fine, but you’re leaving most of the value on the table. Here’s what makes memory genuinely time-saving:

Job-specific standing context. The things that are always true about your work and that you’d otherwise paste into every prompt. If you’re in HR: the size of your team, your industry, the types of roles you hire for. If you’re in finance: whether you’re reporting to a board or an ops team, your preferred number formats, the fiscal year calendar. If you’re in marketing: your brand tone of voice, your target customer description, the channels you focus on.

Recurring project details. If you’re three months into a product launch, you can store the product name, the launch date, the key messages, and the core audience. Then every piece of copy or comms you draft with AI has that context baked in without repeating it.

Output preferences that rarely change. “Always use UK English.” “Never use bullet points for executive communications.” “Keep paragraphs to three sentences maximum.” These are the details that make AI output actually usable without a round of editing every time.

For reference, if you’re building a workflow around multiple AI tasks, the guide on how to build an AI workflow in 2026 covers how memory integrates into a broader process.

The privacy side you need to know

This is where I want to be honest with you rather than just telling you to enable everything. There are real privacy considerations with AI memory, and they matter more for some roles than others.

First, the practical reality: the information you store in AI memory can be used to train or improve the AI model, depending on your settings and your plan. Anthropic and OpenAI both publish their data handling policies, but the defaults vary. If you’re on a free plan, the defaults may lean toward data being used for model improvement. Paid enterprise plans typically give you stronger controls. [2]

Second, and this is important for anyone handling client information: don’t put sensitive data in your memory profile. Client names, project details, financial figures, personally identifiable information about employees. None of that should be in your AI memory. The memory is great for storing your preferences and professional context. It’s not the right place for confidential work specifics.

If your organisation has an AI policy (and if it doesn’t, someone should be writing one), check whether it has any guidance on AI memory features before you start using them for work. This is especially relevant for people in HR, legal, and finance.

Honest limitations of AI memory right now

Memory is genuinely useful, but I’d be doing you a disservice if I sold it as a complete solution to the context problem. There are real limitations to be aware of.

Memory is personal, not shared. Every person on your team builds their own memory profile independently. There’s no way to share a team-wide memory layer where, say, everyone’s AI knows the company’s brand guidelines automatically. You can work around this with prompt templates and shared documents, but the memory feature itself is individual. [4]

It doesn’t pull from your other systems. AI memory stores what you tell it in chat. It doesn’t automatically read your emails, Slack messages, or documents. If you want the AI to work with data from those systems, you’ll need specific integrations or you’ll need to paste content manually.

It can store wrong or outdated things. If you told the AI three months ago that you’re working on a restructuring project, it might still apply that context long after the project ended. Regular audits prevent this, but it does require some active management.

Memory doesn’t work across every interface. The memory feature works in the main chat interface. Some third-party apps built on the API may not have access to your stored memory profile. Check before you assume it’s carrying over.

For a practical comparison of how Claude stacks up against ChatGPT more broadly for business use, the guide on Microsoft Copilot vs Google Gemini covers how the different AI tools compare across real work tasks.

Sources

1. [1] The New Stack. Claude’s Free Plan Can Now Remember You. 2026.
2. [2] Anthropic. Use Claude’s chat search and memory to build on previous context. 2026.
3. [3] Business Standard. Gemini now lets you import memory, chats from ChatGPT, Claude: How it works. 2026.
4. [4] XTrace AI. Is Claude’s Memory Worth the Price? What the 2026 Tiers Actually Get You. 2026.
5. [5] SuperClaude. Claude AI Persistent Memory: How It Works and Why It Changes Everything. 2026.