Claude AI Not Working? Fix These 10 Common Errors (2026)

I use Claude for probably four hours a day. Code reviews, writing drafts, debugging sessions, API work -- it's embedded in my workflow in a way that means when it breaks, I notice immediately.

And it does break. Not often, but the errors it throws are confusing if you haven't seen them before. I've hit every single one of the issues below at least a dozen times. Here's what each one actually means and what to do about it.

1. "Claude is at capacity right now"

The most common one, especially if you're on the free tier. This means Anthropic's servers are handling more traffic than they can comfortably serve -- usually during peak US work hours (roughly 9am-6pm Eastern).

What it means: Not a bug on your end. The servers are just busy.

Fixes:

• Wait 10-30 minutes and try again. Seriously, this usually clears on its own.
• Try off-peak hours. Early morning or late evening typically sees lower wait times.
• Upgrade to Claude Pro -- paid subscribers get priority access and hit this error far less often. At $20/month, it's worth it if you're using Claude for work. This is a genuine fix, not a pitch.
• If you're a developer, switch from the web interface to the API with your own key -- rate limits are handled differently.

When to escalate: If you're a Pro subscriber still hitting this repeatedly, check Anthropic's status page for ongoing incidents.

2. Context window limit reached

You're mid-conversation -- maybe a long debugging session or a multi-part document review -- and Claude suddenly can't continue. Either it stops making sense, or you get an error saying the conversation is too long.

This happens because Claude processes the entire conversation history every time it responds. When that history exceeds the model's context window (200,000 tokens for Claude 3 models -- roughly 150,000 words), something has to give.

What to do:

• Start a new chat and paste only the relevant portion of your work. Don't try to summarize the whole conversation -- just give it the context it needs for the next task.
• Split complex tasks upfront. If you're reviewing a 10,000-line codebase, don't dump it all at once. Break it into modules.
• Use Projects (Claude's long-term memory feature). Projects let you store persistent context so you're not rebuilding it from scratch every session.
• Ask Claude to summarize the conversation so far, copy that summary, start a new chat, paste it in. Quick and effective.

3. "Unable to generate a response"

Vague error, annoying error. Usually one of two things is happening.

First possibility: your prompt is doing something that's tripping a content filter -- even if your intent is completely legitimate. The filters aren't perfect.

Second: the prompt is genuinely too complex or ambiguous. Claude sometimes gets stuck on prompts that require it to hold too many conflicting instructions simultaneously.

Fixes:

• Simplify and resubmit. Break one long prompt into two shorter ones.
• Rephrase. If you asked Claude to write something "dark" or "dangerous" in a fiction context, try providing more framing about the creative purpose.
• If it's a technical prompt, be more specific. "Fix my code" fails more often than "Here's the function. It's supposed to return X but returns Y. What's wrong?"
• Try a completely fresh conversation. Sometimes it's a transient issue.

4. Claude giving incomplete responses or cutting off mid-sentence

You asked Claude to write something long -- a detailed guide, a full code file, a thorough analysis -- and it stops halfway through. Sometimes with an apology, sometimes just... stops.

This happens because Claude has an output token limit per response (around 8,000 tokens in most models, roughly 6,000 words). It's not a bug. It's a hard ceiling.

Workarounds:

• End your prompt with "If your response is cut off, I'll ask you to continue." Then, when it stops, just reply "continue" -- Claude will pick up exactly where it left off. Works almost every time.
• Break large tasks into smaller pieces. "Write section 1 of 5" instead of "Write the whole document."
• For code specifically: ask for one function at a time, or one module at a time.

The "continue" trick is legitimately one of the most useful Claude habits I've developed. Use it constantly.

5. Claude Artifacts not loading

You're in the Claude web interface, an Artifact should pop up (a code block, a document, a chart), and either nothing appears or it loads infinitely.

Most likely causes:

• Browser extension conflict. Ad blockers, privacy extensions, and script blockers are the usual suspects.
• Stale cache. The Artifacts renderer can get stuck.
• Browser version issue.

Fix sequence:

1. Try incognito/private mode first. No extensions, fresh state. If it works in incognito, an extension is your problem -- disable them one by one to find the culprit.
2. Clear browser cache and reload. (Chrome: Ctrl+Shift+Delete → All time → Cached images and files)
3. Try a different browser. Chrome, Firefox, and Safari all behave slightly differently with Artifacts.
4. Disable ad blockers or whitelist claude.ai specifically.

If none of that works, it's likely a temporary issue on Anthropic's end. Check status.anthropic.com.

6. Claude "forgetting" earlier parts of a conversation

You mentioned something important at the start of a long conversation, and 20 messages later, Claude acts like it never happened. This is maddening, especially when you've given it important context about your project.

Here's the thing: Claude doesn't have persistent memory between sessions by default. Within a session, it can theoretically recall everything -- but in very long conversations, it starts to weight recent messages more heavily, which means early context fades.

Solutions:

• Projects is the real answer here. If you're working on anything ongoing -- a software project, a writing series, recurring research -- set up a Project and add your persistent context there. Claude will reference it every session.
• Keep important context near the end of long conversations, not just at the beginning.
• For complex work sessions, paste a "briefing" at the start of each conversation: "You're helping me with [project]. Here's what you need to know: [3-5 bullet points]."
• Say "remember this for the rest of our conversation" for critical constraints. It's not magic, but it helps.

Projects genuinely changed how I use Claude for work. If you haven't set one up, do it.

7. Slow responses / timeouts

Claude's just... slow. Or it starts generating and times out before finishing.

Multiple possible causes here.

Peak traffic hours. Claude is slower from roughly 10am-4pm Eastern when US users are most active. Not much you can do except wait or try a different time.

Model size. Claude Opus is considerably slower than Sonnet or Haiku. If you don't need the heaviest model, switch. Claude.ai's default is Sonnet, which is a good balance. Opus is worth it for genuinely complex reasoning tasks, not for quick questions.

Very long context. The more conversation history Claude has to process, the slower each response gets. Starting a fresh conversation with just the relevant context often speeds things up significantly.

For API users: You might be hitting rate limits that throttle your request throughput. Check your usage dashboard. Consider implementing streaming so responses appear progressively rather than all at once -- this makes the UI feel much faster even if total latency is the same.

8. Claude API errors (429 and 529) -- for developers

If you're building with the Claude API, you'll eventually see these:

429 (Too Many Requests): You've exceeded your rate limit. This isn't a server problem -- it's your account's quota.

529 (API Overload): Anthropic's servers are under load. Similar to the capacity error in the web interface.

How to handle 429:

• Implement exponential backoff. If you get a 429, wait 1 second, retry. If you get another, wait 2 seconds, retry. Then 4, then 8, up to some maximum. Don't just hammer the API.
• Check your usage tier in the Anthropic console. If you're consistently hitting limits, you may need to request a quota increase or move to a higher tier.
• Cache responses where possible. If you're making the same call repeatedly, store the result.

How to handle 529:

• Same exponential backoff approach.
• The 529 is transient -- it almost always clears within minutes. Your retry logic should catch it.

Here's the minimal Python pattern:

import anthropic
import time

client = anthropic.Anthropic()

def call_with_backoff(prompt, max_retries=5):
    for attempt in range(max_retries):
        try:
            response = client.messages.create(
                model="claude-sonnet-4-6",
                max_tokens=1024,
                messages=[{"role": "user", "content": prompt}]
            )
            return response
        except anthropic.RateLimitError:
            if attempt == max_retries - 1:
                raise
            wait = 2 ** attempt
            print(f"Rate limited. Waiting {wait}s...")
            time.sleep(wait)
        except anthropic.APIStatusError as e:
            if e.status_code == 529 and attempt < max_retries - 1:
                wait = 2 ** attempt
                time.sleep(wait)
            else:
                raise

Set up this pattern once and you'll basically never have to think about API errors again.

9. Claude iOS/Android app crashing

The mobile apps are generally solid, but crashes happen. Usually one of three things.

Fix sequence:

1. Clear the app cache. iOS: Settings → General → iPhone Storage → Claude → Offload App (reinstalls without deleting data). Android: Settings → Apps → Claude → Storage → Clear Cache.
2. Update the app. Check the App Store or Play Store. If you're running an old version, update first -- a lot of stability fixes land in point releases.
3. Restart your phone. Boring, but effective. Clears memory pressure that causes crashes.
4. Reinstall the app. Last resort, but it solves most persistent crash issues. Sign back in with your account.

If it's crashing on a specific conversation, try starting a new one -- it might be corrupted conversation data, not an app issue.

10. Claude refusing a reasonable request

This one's the most frustrating, because it often feels arbitrary. You asked something completely legitimate and Claude said no -- or gave you a watered-down version of what you asked for.

First: understand that Claude's content filters are calibrated conservatively. They're designed to catch genuinely harmful requests, but they occasionally flag edge cases. It's not personal, and it's not necessarily wrong about everything -- but it does mean you sometimes need to give it more context.

Approaches that actually work:

• Add context. "I'm a security researcher" or "This is for a historical fiction novel set in WWII" or "I'm a medical professional asking about..." -- context changes what Claude can help with.
• Rephrase. Sometimes specific words trigger filters even when the underlying request is fine. "Explain how hackers break into systems" might work better than "How do I hack into a system."
• Break it down. Instead of one big ask that includes a sensitive element, approach it in steps.
• Be direct about what you're actually trying to accomplish. Claude sometimes refuses a surface request but will help with the underlying goal if you explain it clearly.

What doesn't work: arguing with Claude about whether the refusal was fair. It won't help. Adjust the framing and try again.

If Claude is consistently refusing things that seem completely reasonable, check Anthropic's usage policy -- there may be a category mismatch between what you're expecting and what the policy allows.

---

Quick reference

Error	Quick fix
Capacity error	Wait it out or upgrade to Pro
Context limit	Start new chat, use Projects
Unable to generate	Simplify prompt, add context
Response cuts off	Reply "continue"
Artifacts not loading	Try incognito, clear cache
Forgetting context	Use Projects feature
Slow / timeout	Off-peak hours, shorter context
API 429/529	Exponential backoff
App crashing	Clear cache, update, reinstall
Refusing requests	Add context, rephrase

---

Most of these issues resolve in minutes once you know what's actually happening. The ones that don't -- genuinely stuck Claude Pro capacity issues, persistent API errors that backoff isn't fixing -- are worth contacting Anthropic support about. Their help center is reasonably useful and the support team is responsive.

The context window and "continue" workarounds are the ones I'd prioritize learning. They come up constantly, they're easy to implement, and they make a noticeable difference in daily Claude workflow.

---

Want to dig deeper into specific Claude use cases? Check out our comparison of Claude vs. ChatGPT for writing and ChatGPT vs. Claude head-to-head breakdown for more on when to use each model.