AI Assistant Not Responding: Common Causes and Step-by-Step Fixes

TL;DR — If the AI assistant sometimes does not answer, the most common causes are an expired API key, usage limits being hit, a broken webhook or backend timeout, or safety/routing rules dropping the request before it reaches the model. Start by checking your provider dashboard for failed requests and quota errors; if you only do one thing, confirm the API key is valid and that you still have available usage. Reading time: ~6 min

The scenario

It is a normal Tuesday afternoon. A customer opens your site chat, types a perfectly ordinary question, and the typing indicator spins for a while before nothing appears. A few conversations work, a few fail, and your team cannot tell whether the problem is the AI provider, your website, or some rule in the middle. The dashboard may still show your app as "up," which makes this extra confusing because users are clearly seeing silence instead of answers.

Symptoms

• The user sees one of these behaviors:
  • The assistant shows a typing/loading state and then nothing appears.
  • The assistant replies in some chats but not others.
  • The send button works, but the message never gets a response.
• Common API or app errors in logs:

  401 Unauthorized
  403 Forbidden
  429 Too Many Requests
  500 Internal Server Error
  502 Bad Gateway
  503 Service Unavailable
  504 Gateway Timeout
  

• Common provider or backend messages:

  Invalid API key
  Rate limit exceeded
  Insufficient quota
  context_length_exceeded
  upstream request timeout
  webhook delivery failed
  

• Browser developer tools (the browser's built-in request inspector) may show:

  Failed to fetch
  net::ERR_BLOCKED_BY_CLIENT
  CORS error
  

• Web server or app logs may show lines like:

  POST /api/chat 504
  Error: socket hang up
  Error: ETIMEDOUT
  Error: request aborted
  

Likely causes

Step-by-step diagnosis

1. Check whether the AI provider is currently degraded.

   • Menu path: open your provider's public status page in a browser.
   • This is your problem if you see an active incident affecting API responses, elevated latency, or errors during the same time window.
   • Jump to: Fixes → Provider outage or degraded service.

2. Check whether your API key is valid.

   • Menu path: in your provider's dashboard, open the API keys page and find the key your app uses.
   • CLI option if you have server access:

   printenv | grep -E 'OPENAI|ANTHROPIC|AI|API_KEY'
   

   • This is your problem if the key is missing, revoked, expired, or different from the one configured in your app.
   • Jump to: Fixes → Invalid, expired, or missing AI provider API key.

3. Check quota and rate limits.

   • Menu path: in your provider's dashboard, open Billing, Usage, or Limits.
   • This is your problem if you see 429 Too Many Requests, Insufficient quota, a hard monthly cap, or no available credit.
   • Jump to: Fixes → Rate limit or quota exhausted.

4. Check your own app logs for timeout errors.

   • Menu path: in your hosting provider's dashboard, open your app/service → Logs.
   • CLI option:

   grep -Ei '504|ETIMEDOUT|timeout|socket hang up|request aborted' /var/log/* 2>/dev/null
   

   • This is your problem if the request starts but your app returns 504, 502, or timeout errors before a model response arrives.
   • Jump to: Fixes → Your backend or webhook timed out before the model replied.

5. Check whether a rule blocked the message.

   • Menu path: in your app admin area, open conversation logs, moderation logs, or workflow/routing rules.
   • This is your problem if the request is marked blocked, filtered, or routed to a path with no response action.
   • Jump to: Fixes → A moderation, guardrail, or routing rule blocked the request.

6. Check the browser request itself.

   • Menu path: in Chrome/Edge/Firefox, press F12 → Network → send a test message → click the failed POST request.
   • This is your problem if the request never leaves the browser, shows CORS, Failed to fetch, or ERR_BLOCKED_BY_CLIENT.
   • Jump to: Fixes → The browser request failed before it reached your backend.

7. If none of the above matched, test one request directly against your backend endpoint.

   • Replace the URL with your actual chat endpoint:

   curl -i -X POST https://your-domain.example/api/chat -H 'Content-Type: application/json' -d '{"message":"Say hello"}'
   

   • This is your problem if the endpoint returns a non-200 status, hangs, or returns an app-specific error message.
   • Jump to the fix section that matches the returned error.

Fixes

Invalid, expired, or missing AI provider API key

• In your provider dashboard, create a new API key if the old one is revoked or expired.
• In your hosting dashboard, update the environment variable used by your app. Common names:

  OPENAI_API_KEY=your_new_key_here
  ANTHROPIC_API_KEY=your_new_key_here
  

• Then redeploy or restart the app from your hosting dashboard. If you have shell access:

  export OPENAI_API_KEY='your_new_key_here'
  systemctl restart your-app
  

• If your app stores secrets in a .env file:

  sed -i 's/^OPENAI_API_KEY=.*/OPENAI_API_KEY=your_new_key_here/' .env
  

• Verify it worked:

  curl -i -X POST https://your-domain.example/api/chat -H 'Content-Type: application/json' -d '{"message":"hello"}'
  

Rate limit or quota exhausted

• In the provider dashboard, add payment details, raise the usage cap, or wait for the limit window to reset.
• If your app sends too many requests at once, lower concurrency (how many requests run in parallel) in your worker or server config. Example Node.js queue setting:

  {"aiRequestConcurrency":2}
  

• Add simple retries with backoff (a short delay before retrying) for 429 responses. Example pseudocode config:

  {"retryOn":[429],"retries":3,"backoffMs":1000}
  

• Verify it worked: send 2-3 test messages in a row and confirm your provider dashboard no longer shows new 429 errors.

Your backend or webhook timed out before the model replied

• Increase the timeout in your reverse proxy or app server. For nginx:

  location /api/chat {
    proxy_connect_timeout 60s;
    proxy_send_timeout 120s;
    proxy_read_timeout 120s;
    send_timeout 120s;
  }
  

• Reload nginx:

  nginx -t && systemctl reload nginx
  

• If your app platform has a request timeout setting, raise it in the service settings page.
• If responses are long, enable streaming in your app so the user sees output sooner and the connection stays active.
• Verify it worked:

  grep -Ei '504|ETIMEDOUT|timeout' /var/log/nginx/error.log /var/log/* 2>/dev/null | tail
  

  You should stop seeing fresh timeout lines during a new test.

A moderation, guardrail, or routing rule blocked the request

• Open your app's moderation or workflow rules and look for conditions that match too broadly, such as blocking all messages with links, code, or certain keywords.
• If you use a routing rule, confirm every route ends in a response action or fallback. A safe fallback pattern is:

  {"route":"default","action":"send_to_ai"}
  

• If you keep rules in config, narrow the blocked patterns. Example:

  {"blockPatterns":["password reset token"],"allowUnknown":true}
  

• Verify it worked: resend one previously blocked message and confirm it now appears in the conversation log as processed, not filtered.

The browser request failed before it reached your backend

• If Developer Tools shows ERR_BLOCKED_BY_CLIENT, disable ad/privacy extensions for your site and test again.
• If it shows a CORS error (the browser blocked a cross-site request), add your website origin to the backend allowlist. Example Express config:

  app.use(cors({ origin: ['https://www.your-domain.example'] }))
  

• If the frontend points to the wrong API URL, fix the environment variable in your hosting dashboard. Example:

  NEXT_PUBLIC_API_URL=https://api.your-domain.example
  

• Verify it worked: in Developer Tools → Network, the POST request should return 200 or your normal success status instead of failing in the browser.

Provider outage or degraded service

• Confirm the incident on the provider status page.
• Temporarily show a friendly fallback message in your app instead of leaving users with silence. Example response text:

  The assistant is temporarily delayed. Please try again in a few minutes.
  

• If your app supports a secondary provider, switch the model/provider in your app settings or config. Example environment variables:

  AI_PROVIDER=backup
  AI_MODEL=backup-model-name
  

⚠️ Switching providers can change output style, cost, and data handling. Review your privacy and compliance requirements before enabling a backup provider.

• Verify it worked: new requests should either succeed through the backup path or return the fallback message immediately instead of hanging.

Prevention

• Add an uptime check for your chat endpoint every minute.

  curl -fsS https://your-domain.example/api/health || echo 'chat endpoint failed'
  

• Alert on 401, 429, and 5xx spikes in your logs. Example grep for a simple cron job:

  grep -E ' 401 | 429 | 5[0-9][0-9] ' /var/log/nginx/access.log | tail -100
  

• Add a startup check that fails fast when the API key is missing.

  if (!process.env.OPENAI_API_KEY) { throw new Error('OPENAI_API_KEY is missing') }
  

• Pin and document your timeout settings so deploys do not reset them.

  proxy_read_timeout 120s;
  

• Add a CI smoke test that posts a tiny prompt to a non-production endpoint after deploy.

  curl -i -X POST https://staging.your-domain.example/api/chat -H 'Content-Type: application/json' -d '{"message":"health check"}'
  

• Log blocked moderation/routing decisions separately from provider failures so "silent" drops are visible.

  {"event":"message_blocked","reason":"policy_rule","conversationId":"12345"}
  

---

This article was written by an AI system and published pending human review. Verify anything you intend to act on.