Don't stop here
Hand-picked guides our readers explore right after this one.
Master ChatGPT with advanced prompting techniques, mega-prompts, and proven frameworks
Read the guideExpert guide to Claude prompts with XML tags, artifacts, and complex reasoning
Read the guideResearch-grade prompts for Perplexity AI's search-powered responses
Read the guideA ChatGPT thread that was snappy last week now takes seconds to register each keystroke, scrolls in jerks, and sometimes locks the whole browser tab. This is a different problem from ChatGPT being slow generally. Two distinct things degrade with thread length: your browser has to keep every previous message, image, code block, and table rendered in the page, which balloons the DOM until the tab runs out of headroom, and the model has to reprocess a growing conversation history before each new reply. The first is a client-side rendering problem you can fix outright. The second is inherent to how context works and is best solved by handing off to a fresh thread. This guide covers both.
Typing in the message box lags behind your keystrokes by a second or more
Scrolling the conversation stutters or jumps
The browser tab freezes, greys out, or shows a 'page unresponsive' prompt
Fans spin up and the tab uses well over a gigabyte of memory
Replies take noticeably longer in this thread than in a brand new one
The problem clears instantly when you open a new chat
ChatGPT keeps the entire conversation in the page rather than virtualising it. Hundreds of messages, especially ones with images, long code blocks, or tables, produce tens of thousands of DOM nodes, which is what makes typing and scrolling stutter.
Every new reply is generated with the accumulated conversation as input. As that history grows, time to first token grows with it, so the same question answers slower in a long thread than in a fresh one.
Grammar checkers, ad blockers, privacy tools, and ChatGPT-specific extensions run handlers over the page content. On a huge DOM that cost multiplies, and one badly behaved extension can account for most of the lag.
A long-lived ChatGPT tab can end up running outdated cached assets. Users who see a sudden slowdown after weeks of normal use most often clear it with a cache purge and hard reload.
On machines with 8GB of RAM or a phone with many apps open, a heavy ChatGPT tab competes for memory with everything else and the browser starts swapping, which is felt as freezing rather than slowness.
When to try: First, for any thread that has become painful to use.
Ask the thread for a handoff summary: the goal, decisions made, current state, and anything that must carry forward. Copy it, open a new chat, paste it as the first message, and continue there. This is the single most effective fix because it resets both the DOM size and the reprocessed history.
When to try: For immediate relief when you need to finish something in the current thread.
Press Shift and click reload (or Ctrl+Shift+R on Windows, Cmd+Shift+R on Mac). This rebuilds the page and often restores responsiveness for a while, since a freshly loaded long chat renders fewer messages up front than one you have been scrolling through for an hour.
When to try: When lag appeared without the thread getting much longer.
Open a private window, sign in, and open the same conversation. If it is noticeably faster, an extension is the cause. Re-enable extensions one at a time in your normal profile to find the culprit, and disable it on chatgpt.com specifically rather than everywhere.
When to try: When even short new chats feel sluggish.
In your browser settings, clear cached images and files, and clear cookies and site data for chatgpt.com specifically. Sign back in. This resolves the case where performance suddenly degraded across all chats rather than just the long one.
When to try: As a working habit once you have hit this problem once.
Instead of one mega-thread per project, open a chat per task or per day, and use Projects to keep them grouped with shared files and instructions. You keep the continuity benefit without any single thread growing large enough to choke the tab.
When to try: When you cannot start a new thread and need the existing one now.
The desktop app and mobile apps handle long threads differently from a browser tab and are sometimes noticeably better. Likewise, if you are on one browser, try another. This is a workaround rather than a fix, but it unblocks urgent work.
When to try: When the whole browser, not just ChatGPT, becomes unresponsive.
Close other heavy tabs and applications, then reload ChatGPT. On machines with limited RAM, a long ChatGPT thread plus a dozen other tabs is enough to push the browser into swapping, which presents as freezing rather than gradual slowness.
When to try: Preventively, in threads you expect to keep for weeks.
Avoid pasting large files or long logs directly into an already-long conversation. Upload them as attachments, or better, put them in a Project so they can be referenced without being re-rendered inline in the transcript every time you scroll.
Start a new chat per task and use Projects to group related threads
Ask for a handoff summary before a thread gets unwieldy, not after
Keep ChatGPT-specific browser extensions to a minimum
Attach or reference large files rather than pasting them inline
This is a client-side and context-size issue, so support generally cannot fix it. Contact OpenAI at help.openai.com only if brand new, short chats are also slow or freezing on a machine and browser that handle everything else fine, which would suggest an account or delivery problem rather than thread size.
Two reasons stack up. Your browser keeps every previous message rendered in the page, so the DOM grows until typing and scrolling stutter, and the model reprocesses the whole accumulated history before each new reply, so responses take longer to start. A new chat resets both.
Ask the current thread for a handoff summary covering the goal, key decisions, current state, and open questions. Copy it into the first message of a new chat. You keep the substance while dropping the rendering and reprocessing cost of the old transcript.
Yes, usually dramatically. Almost all lag specific to a long thread disappears in a fresh one, because both the page DOM and the conversation history the model must reprocess start from nothing.
Yes. Grammar tools, ad blockers, privacy extensions, and ChatGPT add-ons run handlers over the page, and on a very large conversation that cost multiplies. Testing in an incognito window with extensions disabled is the quickest way to confirm it.
No. Rate limiting shows as an explicit message about hitting a limit, not as sluggish typing or a frozen tab. Interface lag that clears in a new chat is a rendering and context-size problem, not a quota one.