Claude changed, and saving your chats stopped working. Here's what still does.
If you tried to save a long Claude conversation recently and found that nothing works anymore — copy-paste drops half the message, Save as HTML produces a broken file, the extension you relied on suddenly exports an empty page — you are not doing anything wrong. Anthropic changed how claude.ai renders and serves conversations, and that one change broke almost every method people used to keep a copy of their own chats.
This is a guide to what actually survived, and why. It is honest about the options that are not ours, because the point is not to sell you something — it is that losing months of your own work to a website update is a genuinely bad thing, and it is avoidable.
Why the old methods broke
Almost every "save my Claude chat" method fell into one of two buckets, and the change hit both.
Reading the page (DOM scraping). Copy-paste, Save as HTML, and the popular "save the whole page as one HTML file" extensions all work by reading what is currently rendered on screen. Modern claude.ai only keeps a window of the conversation in the page at any time — as you scroll, older turns are unloaded and newer ones swapped in (virtualized rendering). So when you select-all and copy, or save the page, you capture whatever fragment happened to be mounted at that moment. Long conversations come out truncated, scrambled, or empty. This is not a bug you can work around by scrolling slowly; it is how the page is now built.
The layout itself changed. Class names, container structure, and the markup that scrapers keyed off of all shifted. Any tool that hard-coded "find the message bubbles here" simply stopped finding them.
The net effect was sudden and quiet: the exact methods that worked last month return nothing this month, usually with no error — just an empty or half-empty file. One long-time user described spending weeks copying every turn of a conversation by hand into a text file, because every automatic method had failed on them.
The one approach that didn't break — and why
There is a third bucket the change did not touch: reading the conversation the way the app itself does.
The claude.ai web app does not scrape its own screen. It asks Claude's backend for the full conversation — every turn, in order, with attachments and generated files intact — and then renders it. A tool that asks for the same data, using your existing logged-in session, gets the complete, correct conversation regardless of what is currently painted on screen or how the layout is styled. Virtualized rendering and CSS reshuffles do not affect it, because it never looked at the rendered page in the first place.
That is the structural reason DOM scrapers are fragile and this approach is durable. It is also why "save the whole page as HTML" was never really archiving your conversation — it was photographing a window and hoping.
What still works today, ranked
Start at the top and stop when one fits you.
- Anthropic's official export. Settings → Privacy → Export data. Free, complete, no third party touches it. The catch: it arrives as a ZIP of JSON that is not human-readable, covers everything at once rather than a single chat, and can take a while to be emailed to you. Good for a bulk safety copy; wrong tool if you want this one conversation as a clean readable file right now. (We broke down exactly what is in that file here.)
- Open-source console scripts. If you would rather not install anything, pasteable DevTools snippets can pull a single conversation using the same session-based approach. Auditable and zero-install, but fiddly — you are pasting code into a developer console.
- A session-based archiver extension. This is what we build — AI Chat Archive. One click on any Claude (or ChatGPT) conversation, pick a format (Markdown, PDF, HTML, JSON or TXT), done. Because it reads the conversation through your session rather than scraping the page, it keeps working across the UI changes that break the other tools. Single-conversation export in every format is free; whole-project bundles, batch history, and attachment files are the paid tier. Here is how it compares to other exporters.
If method 1 covers you, use it and don't pay anyone — including us.
This is really about archiving, not exporting
Here is the reframe worth sitting with. "Export" sounds like a one-time convenience — convert this chat, move on. But the reason people are upset when it breaks is not convenience. It is that a conversation they invested real thinking into — a research thread, a project's decision history, months of a working relationship with an assistant — was only ever stored on someone else's server, and that server changed the rules without warning.
An archive is different from an export. An archive is yours, it is complete, and it is durable — it does not stop existing because a website shipped a redesign. That is the thing actually worth having: not the act of exporting, but the peace of mind that your record can't be taken away by a change you didn't make.
"Anthropic changed something about Claude.ai and every previous means of saving long conversations was destroyed — even manual copy-pasting is broken, save-as-HTML broken, extensions that save the entire page broken. But this works. One click, select the file type, done. Such a timesaver and real peace of mind."
How to save a Claude conversation right now
- Open the conversation at
claude.ai/chat/… - Install AI Chat Archive (Chrome, Edge, Brave, Arc, Firefox, Safari), or use an open-source console script if you prefer zero install.
- Click the Export button in the top-right of the chat, choose Markdown (best for keeping and re-reading) or PDF (best for a fixed, shareable record), and it saves straight to your Downloads.
For a whole project — instructions, knowledge files, and every conversation inside — export the project page instead and you get one ZIP with everything. Using the new agent surface? Here's how to export Claude Cowork tasks.
When you don't need any of this
If you only ever have short, throwaway chats and never look back at them, the official export as an occasional bulk backup is plenty. This whole category matters when your conversations accumulate value over time — research, writing, code, an ongoing project — and losing them, or losing access to them at the platform's discretion, would actually cost you something. If that is you, keep your own copy. It should never again depend on a webpage not changing.