I have spent the past few weeks running the same tasks through Claude, ChatGPT, and Gemini side by side, and one thing stood out more than raw intelligence: memory. Which of these tools actually keeps track of what you told it last week, and which one makes you start over every single time?
Claude’s Memory Feels the Most Deliberate
Anthropic rolled out persistent memory to every Claude user, free and paid, in March 2026, alongside a tool that imports saved context from ChatGPT, Gemini, and Microsoft Copilot so switchers do not have to rebuild their history from scratch. Users on the free tier can now draw on earlier chats when interacting with Claude, with the system tailoring responses based on prior exchanges. What impressed me most is how Claude handles this: it doesn’t save full conversations; it synthesises them, extracting details like role, preferences and formatting style into a running summary you can view and edit.
In practice, this is the difference I noticed. When I go back to Claude after days away, it picks up my writing style, my publication preferences, and the corrections I made to earlier drafts without me repeating myself. That consistency is the single biggest reason it is my first choice.
ChatGPT Is the Easiest to Understand
ChatGPT’s memory setup has become more layered over time. By 2026, it runs on two separate systems: an explicit, user-editable list of “saved memories,” and an implicit “reference chat history” that recalls patterns from past chats, both controlled independently from Settings. In June 2026, OpenAI pushed this further with a system nicknamed “Dreaming,” which consolidates memory in the background between sessions, even revising stored facts as circumstances change, such as updating a future trip to a past one once it’s over.
What I like about ChatGPT is not the sophistication of the memory itself but how the whole experience is packaged. Its responses, settings, and explanations are the easiest to follow of the three. Even when its recall of older chats is imperfect, the interface never leaves you confused about what it does or does not remember.
Gemini Struggles Where It Matters Most
Gemini’s memory setup is the most fragmented of the three. Google built it in layers: a “Personal context” layer that references past chats and is on by default, alongside saved preferences, custom instructions, and, with opt-in permission, activity pulled from Gmail, Photos, Search, and YouTube. On paper, that sounds powerful. In use, it is inconsistent. Gemini combines explicit saved information with automatically learned context, but the deeper automatic personalisation requires a paid plan and isn’t available in some regions.
This matches my own experience exactly. Gemini often “forgets” details I shared just days earlier, or applies them inconsistently across sessions. Compared to Claude’s tidy running summary or ChatGPT’s two-tier system, Gemini’s recall feels the least reliable of the three, which is frustrating for anyone doing repeat work like editing, research, or long-running projects.
My Ranking
For my own workflow, Claude comes first. Its memory is dependable, and it consistently applies my past corrections and preferences without me repeating instructions. ChatGPT comes second: its recall is solid and improving, but its real strength is how effortless it is to use and understand. Gemini comes last, mainly because of how often it fails to carry context between sessions, which undermines the very point of having memory in the first place.
None of these tools is perfect, and Google, OpenAI, and Anthropic are all racing to fix the gaps. But if you are choosing an AI assistant for repeat, high-volume work, memory reliability should weigh as heavily as raw capability. Right now, that puts Claude ahead of the pack, with ChatGPT close behind and Gemini still catching up.
