Why Gemini Keeps Forgetting Context (and 3 Fixes)
You explained the project. Gemini understood it. Twenty messages later it asks what you are working on.
This is not a bug and it is not you doing something wrong. Gemini drops context for three specific reasons, and each one has a different fix. Work through them in order.
Three reasons Gemini drops your context
The thread got too long. This is the most common one. Gemini reads a window of recent text, not the entire conversation. Once the thread outgrows that window, the earliest messages slide out. Users report hitting this far earlier than the advertised context length suggests, especially in long back-and-forth sessions rather than one large document.
You are in a surface that does not carry memory. Temporary Chat, Gems and Live all behave differently from a normal conversation, and this catches people out constantly. Google's help pages state plainly that memory of past chats is not available in Gems, and Gemini's own settings page says memory is "coming soon to Live".
Memory is switched off. Least likely, because it is on by default. Still worth ruling out first, since it takes ten seconds.
Fix 1: Rule out the setting
Go straight to gemini.google.com/personalization-settings, or open the gear icon at the bottom left and choose Personal Intelligence.
Check that Memory is on.
Two things worth knowing here.
Google's own help pages still call this Personal context, while the interface now says Personal Intelligence. Same setting, two names, which is why guides written a few months ago send you looking for something that is not there.
And it is on by default, so this is usually not your problem. If it was off, turn it on and start a new conversation. Existing chats will not retroactively fill memory.
If it was already on, move to Fix 2.
Fix 2: Check which surface you are in
This is the one people miss.
Context carries in a normal conversation with Memory on. Temporary Chat is not saved at all. Gems have no memory of past chats, which Google states directly in its help documentation. And Live does not have memory yet, which Gemini's own settings page describes as "coming soon".
If you have been building context inside a Temporary Chat or a Gem, none of it carries. Move to a normal conversation and start again.
Fix 3: Stop stacking topics into one thread
A single thread holding four unrelated projects will lose all four faster than four separate threads each holding one.
Every message you add pushes older content toward the edge of the window, and unrelated messages take up the same space as relevant ones.
Start a new chat when the subject actually changes. Not for every question, but whenever you move to a genuinely different task. Before you leave a long thread, ask Gemini to summarise what you have established, and use that summary as the opening message of the next one.
When none of these help
If Memory is on, you are in a normal conversation, the thread is short, and Gemini still forgets, then the problem is not Gemini.
It is that your context lives inside one assistant. Open Claude for something it handles better, or ChatGPT for something else, and you are explaining from scratch again. No Gemini setting fixes that.
Both Google and Anthropic have noticed. Gemini now has an Import memory to Gemini option, and Claude has an equivalent import from other providers. Both are useful, and both have the same shape: they run one direction, once, into one product. Whatever you tell ChatGPT next week still stays in ChatGPT.
One AI memory instead of four
Unibase Memory is a free Chrome extension that captures conversations from Gemini, ChatGPT, Claude, Grok, Kimi and DeepSeek into one searchable memory, then sends the specific piece you need into whichever assistant you are in.
The difference from Gemini's own memory is direction. Gemini's memory serves Gemini. This one sits outside all of them, so context you built in ChatGPT last month is available in Gemini today, and it keeps working rather than needing to be re-run.
Setup takes two minutes. It captures from install day forward, so it will not recover conversations you have already lost.
For the full picture across all six platforms, see our guide to moving context between assistants. For the underlying problem, see what cross-AI memory actually means and why you keep repeating yourself to AI.
Which fix applies to you
Forgetting between separate chats means Memory is off. Losing everything after closing the tab means you were in a Temporary Chat. A Gem that never remembers anything is behaving as designed. Forgetting inside one long thread is the context window. And remembering Gemini but nothing from ChatGPT is not a Gemini problem at all.