What goes wrong in long AI chats, and what to do about it
Plain explanations of the problems that show up once a conversation gets long — written to be useful whether or not you ever install Tvara.
Long chat toolkit
Long AI chats fail in five distinct ways: speed, navigation, search, lost context and invisible limits. What fixes each one, with and without extra tools.
Why long chats get slow
Long ChatGPT and Claude chats get slow because the page holds tens of thousands of elements, not because the model slowed down. What helps, in order.
Find an old chat
How to search old ChatGPT, Claude and Gemini conversations when Ctrl+F fails on long chats. Four methods that work, and why a local archive solves it.
Export your history
How to export ChatGPT, Claude and Gemini chat history: what official exports give you, what a readable export looks like, and why backup is a different job.
See your usage limit
Why assistants hide the counter, what they publish anyway, how to plan around a rolling window, and why an invented number is worse than none.
Recover a deleted chat
Usually not, and not for long. What deletion really does, the three things worth trying in the first hour, and the only fix that works before it happens.
Why the AI forgets
Why does the AI forget what you said earlier? The whole chat is re-sent each message, and what no longer fits the context window drops out. What to do.
Continue in a new chat
To continue a long AI conversation in a new chat, carry the goal, decisions, current state and next question, not the transcript. A handover format and prompt.
Organise your chats
Auto titles describe where a chat began, not what it concluded. Five habits for organising hundreds of AI conversations, and why search beats folders.
Is your history private?
Is your AI chat history private? Where conversations are stored, whether they train models, who can read them, what deletion does, and what extensions add.
Judging an extension
Six tests for choosing a ChatGPT or Claude extension: what it reads, whether chats leave your device, how it fails on a redesign, whether it deletes messages.
Read a long answer
A chat answer has no headings, no contents and a scrollbar that measures the wrong thing. Five ways to read a long AI answer, and the gap underneath.
When the tab freezes
A long AI chat full of code and images can exhaust a tab's memory. What pushes it over, how to recover without losing unsent text, and how to prevent it.
How long can a chat be?
How many messages fit in one AI conversation? No fixed number: two limits apply, what the model can see and what the tab can render. Which comes first.
Keep the good parts
Four approaches in increasing order of effort, and the silent ways copy-paste loses maths, code language, diagrams and images out of a chat answer.
Automatic backups
Account exports are manual snapshots, not a backup. What automatic backup of AI conversations needs (unattended, encrypted, restorable) and how to set it up.
Using two or three
Using ChatGPT, Claude and Gemini together: context does not travel, history splits and limits are separate. Four habits that help, and what never works.
When a chat will not load
AI chat won't load? Three checks that separate almost every cause, the five common culprits, and what to do if the conversation is really gone.
Conversation too long
When ChatGPT or Claude says a conversation is too long, the context window is full. What fills it, and how to continue in a new chat without losing the thread.
Jump to a message
Why Ctrl+F misses old messages in a long AI chat, what works without tools, and a minimap, outline and search that reach any message in one click.
When was it sent?
AI chat sites hide the time on each message. Where the timestamp is stored, how to read it without tools, and how to show real send times on hover.
Glossary
Plain definitions of the words behind long AI chats: context window, tokens, context limit, compaction, memory, handover, usage limits and more.
Claude usage limits
How Claude's five-hour session limit and weekly limit work, where to see what is left, why long chats use it up faster, and how to make it last.
Export one ChatGPT chat
ChatGPT's export covers your whole account and arrives by email. How to save one conversation as Markdown, PDF or JSON, and what each method loses.
Export one Claude chat
How to export one Claude conversation as Markdown, PDF or JSON. Claude's own export covers the whole account and its link expires in 24 hours.
Export one Gemini chat
Gemini exports one response at a time to Docs, Gmail or Sheets, and history through Google Takeout. How to save a whole Gemini conversation as one file.
Gemini usage limits
Gemini meters use by compute, refreshing every five hours until a weekly limit. What spends it, why long chats use it faster, and how to see what is left.
Claude chat too long
When Claude says a conversation is too long, its context window is full. Turn on automatic context management, hand over early, or move files into a Project.