ChatGPT says the conversation is too long — how to continue without starting over

Three weeks into a client thread, the PDF pasted in twice, every decision argued out — and then the message: start a new chat. Here's what the limit actually is, four ways to carry the work forward, and the part none of them fix.

The thread that hits this wall is never a throwaway. It's the one you've been running for weeks: the engagement where you pasted the client's pricing deck, worked through the objections, drafted the reply, revised it twice, and came back the next morning to pick up where you left off. Then ChatGPT tells you the conversation has reached its maximum length and the only way forward is a new chat. "Start a new chat" reads as "start over," and for a thread like that, starting over is the expensive part.

Below: what the limit is, the four real ways to continue, and then the honest gap — why each of them ends with you re-pasting the same material.

What "You've reached the maximum length for this conversation" means

Every time you send a message, ChatGPT re-reads as much of the conversation as fits its window — your messages, its replies, and the text of whatever you pasted or uploaded — and answers from that. Two different limits are in play here, and it helps to keep them apart.

The first is the context window: how much of the thread the model can read at once. As of August 2026, OpenAI's pricing page lists it at roughly 27K tokens on Free, 54K on Go and Plus, and 128K on Pro for the default model, with larger windows for the reasoning models. A thread can run well past that size; the model simply stops seeing the earliest turns. That's the quieter failure — a long chat forgetting the brief you pasted at the top, with no error message at all — and it has its own guide: why ChatGPT forgets things in the same chat.

The second is the one behind this message: a ceiling on the total size of a single conversation. OpenAI doesn't publish the figure. It isn't a count of messages — it's measured in tokens, so everything in the thread weighs against it, and a pasted contract or a long PDF uses far more room than fifty short questions. That's why the wall tends to arrive right after the big upload. It also sits far above the context window: reports on OpenAI's community forum put it in the hundreds of thousands of tokens, well beyond what the model reads in any single turn.

The wording has changed over the years. In 2023 it read "The conversation is too long, please start a new one." Today it reads "You've reached the maximum length for this conversation, but you can keep talking by starting a new chat." Same limit, friendlier sentence. There's no setting that raises it, and no "compact this thread" button — what you do next is up to you.

One look-alike to rule out: if the message you pasted was "This conversation is too long to continue. Start a new chat, or remove some tools to free up space," that one isn't ChatGPT's. Users report it in Claude's desktop app (a bug report filed in April 2026 quotes it word for word), and it has its own guide: what to do when Claude says the conversation is too long.

First: nothing has been deleted

The full conversation is still in your sidebar. You can open it, scroll it, search it, and copy from it; you just can't add to it. ChatGPT's memory — the memory summary it keeps about you, plus reference to your past chats (since April 2025) — can also draw on that thread in a new conversation when those settings are on. But that's a judgment call the model makes about what's relevant, not a transfer of the transcript. OpenAI's own help page says the summary "will not include everything that ChatGPT remembers based on your chats." So don't rely on a new chat knowing what the old one decided. Carry it over deliberately.

Option 1: the hand-off note (the fix that usually works)

If the thread will still accept one more message, spend it on a hand-off rather than one last question. Ask for a structured note, not a summary — a summary keeps the narrative; a hand-off keeps the decisions:

Before I start a new chat, write a hand-off note I can paste into it. Include, in this order: (1) what we're working on and for whom; (2) every decision we've made, with the reasoning in one line each; (3) the names, numbers, and dates a fresh chat would get wrong by guessing; (4) what's still open; (5) the exact next step. Don't summarize the documents I pasted — list them by name so I can re-attach them. Keep it under 400 words.

Open a new chat, paste the note as the first message, re-attach the files it lists, and state the task in one line. If the old thread won't take another message, write the note yourself from the transcript — the five headings above are the checklist. Either way, the last line of the prompt is the tell: the documents don't travel. You re-attach them.

Option 2: Branch in new chat — and why it doesn't fix this

ChatGPT added Branch in new chat in September 2025 for logged-in users on the web, and brought it to the iOS and Android apps in December 2025. On the web, hover over any message, open More actions, and choose it; in the app, long-press the message and tap Branch in new chat. Either way, ChatGPT creates a new thread containing everything up to that message and leaves the original untouched. It's excellent for what it was built for — trying a different approach from a midpoint without overwriting the path you already took.

It is not an escape from the length limit. A branch from the last message of a full thread inherits the full thread, so you arrive in the new chat already at the ceiling; users have reported branches carrying over nearly the whole working context of the original. The useful move is to branch from an earlier message — the one just before the big paste or the long detour — and then add a short note covering what happened after. You get room back and keep the reasoning that mattered. The document you pasted after that point, though, is gone from the branch. You re-attach it.

Option 3: export or copy, so the record is yours

If the thread is the only place a decision trail exists, get it out of ChatGPT. Two ways:

• Copy the thread into a document. Select the conversation, copy, paste into your notes or a doc. Immediate, and it keeps one thread rather than everything.
• Export your data. Settings → Data controls → Export data → Confirm export. ChatGPT emails or texts you a link to a ZIP of your chat history; OpenAI says the export can take up to 7 days to arrive, and the download link expires 24 hours after it does. It's a full-account archive, so it's the right tool for keeping records, not for continuing tomorrow's work.

Either way you now own the transcript. Neither puts it back into the next chat — for that you're back to the hand-off note.

Option 4: move the documents into a Project

If the real pain is re-uploading the same files every time, Projects are the best in-ChatGPT answer. Files added to a project are available to every chat you start inside it, and project instructions apply to all of them — so the pricing deck goes in once, and each new thread in that project can use it without a re-paste. Projects are available on every plan as of August 2026; the file limit per project is 5 on Free, 25 on Go and Plus, and 40 on Pro. A project can also be set to project-only memory when you create it, or changed later in the project's settings, which keeps one client's context from bleeding into another's.

Two caveats. A project doesn't lift the per-conversation ceiling — each chat inside it still fills up, so you'll still write hand-off notes; the files just stop being part of them. And the project only holds what you carried into ChatGPT by hand. The client's reply from Thursday, the revised deck in your inbox, the notes from this morning's call: none of that is in the project unless you put it there, again, every time it changes.

Which fix, when

ChatGPT controls and limits above were checked against OpenAI's help center and pricing page in August 2026. OpenAI renames and moves these often — treat the labels and numbers as a guide and check your own plan.

Where every workaround runs out

Look at the pattern. The hand-off note ends with "re-attach the documents." The branch drops whatever you pasted after the branch point. The export gives you a transcript you then re-paste. The project holds files — but only the ones you uploaded, only as of the version you uploaded. Every fix is a variation of the same move: you carry the context back into the chat by hand, because the chat is the only place the context lived.

That's the real shape of the problem. In ChatGPT, the conversation is the memory. Outgrow the conversation and you're rebuilding the memory from scratch. And for the work that matters — a client engagement, a negotiation, a review that runs for months — the source material was never in the chat to begin with. It's in the email thread with the client, the PDF they sent on Tuesday, the voice memo you dictated after the call. ChatGPT only ever saw the slice you pasted, and the paste is already stale by the time the thread fills up.

With one workstream, that's a chore. With seven clients, each with its own thread, its own deck, and its own "start a new chat" moment, the hand-off notes become a second job — and the detail you forgot to carry over is always the one the client asks about.

If the thread that filled up was client work — a matter, an engagement, a client review — the fix is a different shape of record, not a longer chat. We've set it out by role: for lawyers, for management consultants, and for financial advisors, including what to keep out of a chat window in the first place.

A different answer: context that never lived in the chat

This is the problem Equerry is built for. It's a personal assistant, not a chat thread: you share the material once, it remembers, and each question starts from everything you've already sent rather than from a blank box.

• Forward or CC the thread once. Add your Equerry address (something like mark@mail.getequerry.app) to the client thread, or forward the PDF the client sent Tuesday. Every reply that lands after that is remembered too — the context stays current without a re-paste.
• Share the rest the way you already capture it. Drop in a screenshot or a PDF from the share sheet, dictate a quick voice memo after the call, or save a note. It all lands in one place.
• Ask, and check the source. "What did we agree on pricing with Acme?" comes back with the answer and a source chip pointing at the exact email or note it came from. There's no thread you have to keep alive — the memory lives in the archive, not the conversation, so a new question is just a new question.
• Get briefed before the call. A morning brief can reflect Thursday's late reply you hadn't opened yet, so the next conversation starts from the current state of the engagement, not from a hand-off note you wrote last week.
• See and delete everything it remembers. Every remembered fact is inspectable and removable, with the note or email it came from. You decide what it keeps and what it forgets.

None of this replaces ChatGPT for drafting and thinking out loud — keep it, and use the hand-off note above when a thread fills up. The shift is for the working context that was never in ChatGPT anyway: the client threads, documents, and call notes, held by an assistant that remembers what you sent it and hands it back with the source attached.

For what ChatGPT actually carries between chats, see does ChatGPT remember past conversations. If the opposite is your problem — old threads bleeding into new ones — here's how to start fresh. And for the fuller picture of an assistant that holds your emails, PDFs, and notes together, see when your apps finally talk to each other.

Last updated 2026-09-25.

All product names, logos, and brands are property of their respective owners. Equerry is not affiliated with, endorsed by, or sponsored by any other product mentioned; comparisons are provided for informational purposes.