Comparing the Memory Implementations of Claude and ChatGPT
Source: https://simonwillison.net/2025/Oct/20/comparing-memory-implementations/ Author: Simon Willison Date: 2025-10-20
Summary
Simon Willison’s technical comparison of how Claude and ChatGPT implement memory features. As someone who has reverse-engineered both systems through extensive testing, his analysis is uniquely grounded.
Key Claims
- ChatGPT memory: explicit and user-visible. You can see what it remembers, edit it, delete it. Implemented as a list of facts in the system prompt.
- Claude memory: more implicit in the Projects feature. Claude maintains context across sessions but the memory mechanism is less transparent.
- Design philosophy difference: ChatGPT optimizes for user control; Claude optimizes for seamless continuation.
- Both approaches have weaknesses: ChatGPT’s explicit memories are decontextualized (no relationship between facts); Claude’s implicit approach is opaque.
- What’s missing from both: memory provenance (when was this learned, from what source) and confidence levels.
Entities
- Simon Willison
- Anthropic — Claude
- OpenAI — ChatGPT
Concepts
- Agent Memory — direct comparison of two production memory implementations
- AI Character & Personality — memory shapes how AI presents itself over time