You type a message into Claude, hit send, and nothing happens. Or the spinner just keeps turning. Or your text disappears the moment you submit it. I run four sites and use Claude every day, so I bump into this maybe once or twice a month — enough to have built a personal triage routine that works.
Here is the part that surprised me when I first started keeping notes: almost every "won't send" issue turns out to be on the browser, network, or account-state side, not on Claude itself. With a sensible order, you can usually pin down the real cause in five minutes. The order below is sorted by how often each thing is the culprit, so working top-to-bottom is the fastest route to a fix. Skip ahead if you already know one of the early steps does not apply, but try not to skip more than one — these problems often have multiple contributing causes that masquerade as a single bug.
First, spend 30 seconds checking whether Anthropic is the one having a bad day
The single most wasted activity in any troubleshooting session is clearing your cache when the platform itself is down. Always rule that out first.
Open status.anthropic.com in another tab. If anything in the current time window is marked Investigating, Identified, or Monitoring, there is nothing for you to fix on your end. Wait it out. Only if everything is green do you move to the next step.
Looking at the past six months, the claude.ai surface has roughly one or two minor incidents a month, each lasting between thirty minutes and a couple of hours. It is not unusual for only the Asia/Pacific region to be affected, or for one specific subsystem (web search, file uploads, vision processing) to degrade while the rest of the product still works. So while it might feel like you are the only person whose messages refuse to send, there are often thousands of people in your region staring at the same spinner. Worth knowing too: the status page sometimes lags real-world incidents by 10–15 minutes, so if everything is green but you are seeing widespread reports on social platforms, give it a few minutes before assuming your environment is the problem.
Around 90% of the time, it is a browser extension
Once Anthropic is in the clear, the prime suspect is browser extensions. There are three categories that quietly break claude.ai's send flow:
- Ad blockers (uBlock Origin, AdBlock Plus, Brave Shields, etc.): a few popular filter lists misclassify Anthropic's telemetry endpoints, and as a side effect they sometimes block the POST that submits your message
- Privacy hardeners (Privacy Badger, Ghostery, DuckDuckGo Privacy Essentials, etc.): aggressive tracking heuristics can prevent the cookies that keep your session alive from being written or refreshed
- Page translators (DeepL, ImTranslator, Google Translate's full-page mode, etc.): the kind that rewrite the input element's DOM can also clobber the send button's event handlers, often silently — the button looks normal but does nothing on click
Triage is one minute of work. Open claude.ai in an incognito or private window (Cmd+Shift+N / Ctrl+Shift+N) and try again. If the message sends, it is an extension. To find which one without losing the extensions you actually want, use Chrome's per-site toggle (right-click the extension icon in the toolbar → "This can read and change site data" → "When you click the extension"), so the offender is silent on claude.ai but still active everywhere else. In Firefox the equivalent is in the extension's options page under "Run on these sites."
One subtle variant worth mentioning: I have seen ad blockers fail open on claude.ai itself but break the message send only when you have a specific URL pattern in a connected MCP integration. If you connected an MCP server within the last day or two and only certain conversations fail to send, that is a strong hint to add the relevant domain to your allow list.
When the network is the problem
If incognito mode does not help, the next layer is the network path. Three patterns come up repeatedly:
- A corporate proxy or firewall is closing Server-Sent Events partway through, which Claude relies on for streaming responses
- You are on a VPN whose exit IP collides with Anthropic's regional routing or has been rate-limited as part of a shared egress
- You are on hotel or coffee shop Wi-Fi where the captive portal lets pages load but quietly drops API calls
A quick reachability check can be done from the browser console. Open DevTools (F12), switch to the Console tab, and paste this:
// Check whether you can reach claude.ai's API at all.
// Open DevTools (F12) → Console tab → paste & run.
fetch("https://claude.ai/api/organizations", { credentials: "include" })
.then(res => console.log("Status:", res.status, "OK:", res.ok))
.catch(err => console.error("Network error:", err.message));
// Expected outputs:
// Healthy: Status: 200 OK: true
// Session expired: Status: 401 OK: false → log in again
// Network blocked: Network error: Failed to fetch
// Geo/IP-blocked: Status: 403 OK: false → switch network or VPNIf you get Failed to fetch, switch off the VPN, move to another network, or tether through your phone — whichever is fastest — and confirm whether the symptom moves with the network. If you are on a corporate machine where the issue persists, ask IT whether Anthropic's API domains are on the allow list of your SSL inspection setup. That is the single change that usually unblocks engineering teams behind enterprise proxies. The domains worth flagging by name are claude.ai, api.anthropic.com, and the relevant CloudFront distributions for assets and uploads.
Plan limits are quieter than you'd think
Every Claude plan has a five-hour rolling cap. When you cross it, sends are refused with messaging like "You've reached your usage limit." Strategies for getting the most out of Claude Pro's usage cap covers this in depth — well worth a read if you keep hitting the ceiling.
The detail people miss: attaching a long document — a PDF, a codebase dump, a sprawling chat history — can burn through your remaining budget in a single round trip. If sends started failing right after you uploaded something heavy, that is almost certainly the cause. Wait the cap window out, or use a Project to keep large attachments out of the per-message context. Projects let you index large documents once, after which they're cheaper to query repeatedly. For people who routinely run into the ceiling, that single workflow change recovers the most usable hours per week.
You may have hit the conversation's context ceiling
In long-running threads, sometimes the moment you click send you get something like "Your message is too long." That is not about your message specifically — it is the sum of all prior turns plus your input exceeding Claude's context window.
The fix is mundane: start a new conversation. If you need continuity, ask Claude in the current chat to write a summary you can paste at the top of the next one. That technique preserves about 95% of the working state and almost always lets you keep going. I sometimes ask for two summaries — one a tight 200-word overview, the other a fuller bullet-list version — and pick whichever fits the next task. The brief one is enough for most follow-up questions, and switching to the detailed one is cheap when you actually need it.
A related but distinct symptom — replies that begin and then cut off mid-sentence — is covered in Claude not responding: a complete fix guide. And if your problem only appears when an Artifact is involved, Claude Artifacts not working: troubleshooting walks through that specific failure mode.
When nothing above worked
If you are still stuck, three less-frequent causes remain:
- Account flags: check the inbox for the email tied to your Anthropic account. Policy actions sometimes restrict send capability before any banner appears in the UI, and the only signal is that email. If you have not received anything but suspect this is the case, contact support directly through the help widget on claude.ai
- Desktop app oddities: fully quit and relaunch. If that fails, uninstall and reinstall the latest version — the app's local cache going stale is a real, repeatable failure mode, especially after major version bumps
- Payment status: for Pro/Max users, an expired card silently degrades a subset of features. If you suspect billing, Claude credit card declined: complete troubleshooting walks through every variant of that error
There is a meta-point worth taking from all of this: most "Claude won't send" issues, eight or nine times out of ten, get caught by the first three steps — status check, incognito window, network check. Next time the spinner refuses to stop, just open an incognito window first. That single habit will shave most of the frustration out of your day, and over time it will give you a sharper instinct for which tier of the stack is actually misbehaving — a skill that pays off well beyond Claude itself.