Claude 'Response Could Not Be Generated'? Fix It

That is the reassuring headline, and for most people it is the whole story. You send a prompt, you wait a beat, and instead of an answer you get a short error saying the response could not be generated (or a close cousin, "Claude's response could not be fully generated"). It looks alarming, but it rarely signals a broken account or a lost conversation. The message is not tied to any one model — Claude currently spans models such as Opus 4.8, Sonnet 4.6, and Haiku 4.5, and the same short error can surface on any of them. Below we unpack exactly what the message means, the handful of things that trigger it, and the quickest ways to get a working reply back.

What "response could not be generated" actually means

The phrase is Claude's way of telling you that generation began — or was about to begin — and then stopped before a complete answer reached your screen. Think of it as a dropped call rather than a wrong number. Claude "picked up," but something interrupted the line before it could deliver the full message.

There are two flavors you will see. The plain "response could not be generated" typically appears when the request never really got moving — the server was busy or the connection stumbled at the start. The variant "could not be fully generated" appears when Claude was mid-answer and got cut off partway through, so you may even see a few words or a half-finished paragraph before the error. Both point at the same underlying reality: this is transient, and it is almost never a sign that you typed something wrong.

Because it is transient, the single most effective response is also the simplest. Hit the retry or regenerate control and let Claude try again. A large share of these errors clear on the very first retry, because whatever brief hiccup caused the first attempt to fail has already passed by the time you click. For example, if a retry fails instantly, pausing for 30 seconds before the next attempt gives a passing capacity spike time to clear — and this guide is reviewed regularly so the steps stay accurate.

The common causes, and the fix for each

Most cases trace back to one of a small set of causes. The table below maps each cause to the fix that clears it fastest. Work down the list — the earlier rows are far more common than the later ones.

The reason a plain retry works so often is that the top two causes — passing server load and a momentary connection wobble — are gone within seconds. You are not fighting a persistent fault; you are just catching Claude at a bad instant. When the same message fails two or three times in a row, that is your cue to move down the table to the sturdier fixes: a fresh chat, a shorter prompt, or a connection check.

Server load and busy periods

Claude runs on shared infrastructure, and at peak hours demand can briefly outstrip available capacity. When that happens, an individual request can be turned away with a "response could not be generated" message even though the service as a whole is up. This is closely related to — but not the same as — the dedicated "overloaded" state, which we cover below. For a simple load blip, waiting a few seconds and retrying is enough. If you keep hitting it, trying again during quieter hours makes a noticeable difference. Our guide to the Claude overloaded error walks through the busy-server scenario in more depth.

Connection drops mid-stream

Claude streams its answer to you token by token, so a stable connection matters for the whole duration of the reply, not just the first second. If your Wi-Fi flickers, your VPN reconnects, or your phone hops between towers while Claude is writing, the stream can break and you'll see the "could not be fully generated" variant with a partial answer above it. Confirm you're online, switch to a more stable network if you can, and retry. A stable connection matters right through to the last token, so a quick network check is often all it takes.

Length limits and cut-off replies

Every reply has a maximum output length, and every conversation shares a finite context window. Ask for something enormous — a giant code file, a very long document — and Claude can physically run out of room to finish in a single turn, producing a truncated answer capped by the error. The fix is to work in smaller pieces: request one section at a time, or reply "continue" to get the remainder. If your chat has simply grown huge over many turns, the context window fills and leaves little space for new output. Starting fresh resets that headroom. We go deeper into truncation in why Claude cuts you off.

Distinguishing it from the 529 "overloaded" error

It helps to know that "response could not be generated" is not the same thing as the specific 529 "overloaded" error, even though they can feel similar in the moment. The 529 is an explicit HTTP status that Anthropic's systems return when the service is at capacity — you'll often see the literal word "overloaded" or a numeric 529. The "response could not be generated" message is broader and vaguer: it covers overload, but also connection drops, cut-off replies, and browser glitches, without telling you which one applied.

The practical difference is small, because retrying is the first move for both. But if you specifically see a 529 or the word "overloaded," it's worth reading our dedicated pages on the Claude 529 error and, for the rarer server-fault case, the Claude internal server error. Those explain the status codes precisely so you can tell a capacity issue apart from an outage.

A quick, ordered troubleshooting routine

When the message appears, run through these steps in order and stop as soon as you get a reply:

1. Retry or regenerate the message. This alone fixes the majority of cases because the hiccup has already passed.
2. Refresh the page or restart the app. A stale tab can hold onto a broken connection; a reload gives you a clean one.
3. Check your internet connection. Toggle Wi-Fi, drop the VPN, or move to a stronger signal, then retry.
4. Start a fresh chat. If the current conversation is long or a previous turn failed halfway, a new chat clears the corrupted state. Paste in only what you still need.
5. Shorten or simplify the request. Break a big ask into smaller parts so no single reply hits a length limit.
6. Wait a moment, ideally off-peak. If capacity is tight, a short pause or a quieter hour gets you through.
7. Check for an incident. If nothing works, Claude may be having a wider problem — see whether Claude is down before assuming the issue is on your end.

That routine resolves the overwhelming majority of "response could not be generated" reports. Notice how retrying and refreshing sit at the top: they cost nothing and clear the two most common causes instantly. The heavier steps — new chat, shorter prompt, waiting — are reserved for the minority of cases that survive a couple of retries.

When to suspect something wider

If you retry several times, refresh, switch networks, and start a fresh chat and the response could not be generated message still appears for every prompt, the problem has probably moved beyond your session. At that point it's worth ruling out a broader outage rather than repeating the same steps. Anthropic publishes real-time service health, and checking it takes seconds. Our overview of what to do when Claude is not working collects the wider set of signs — from login failures to blank pages — that point at a platform-level issue rather than a one-off hiccup.

What you don't need to do

It's worth saying plainly: you generally don't need to log out, reinstall the app, change your password, clear every setting, or contact support the instant you see this error. Those are heavy moves for a message that a single retry usually fixes. Reinstalling won't help a passing server-load blip, and it won't un-fill a context window. Save the drastic steps for genuine, persistent failures — and if you reach that point, Anthropic's own help center and its status page are the two authoritative places to check next, in that order.

The mindset that saves the most time is simply this: treat "response could not be generated" as a dropped call. You wouldn't buy a new phone the first time a call drops — you'd just call back. Do the same with Claude. Retry, and in most cases the answer arrives on the second try as if nothing happened.