Uptime monitor says down, but the site works in your browser

Your monitor alerts while the page loads fine for you. Why a check from the internet sees something different from your browser, and how to fix each cause: firewalls, bot protection, timeouts, keywords, redirects and certificates.

· 7 min read · By the Spot Downtime team

The alert says your site is down. You open it and it loads straight away. It's tempting to call it a false alarm and move on, but a monitor that disagrees with your browser is almost always telling you something true: the site behaves differently for a visitor who isn't you.

Your browser has cookies, a logged-in session, cached files, an IP address your firewall trusts and a human behind it that passed a bot check last week. An uptime check has none of that. It arrives fresh from a data center, as an automated client, and gives up after a few seconds. Each difference is a possible cause, and the error on the alert tells you which one.

Start with the error message

Open the monitor and look at the last failed check. The error narrows the cause down to one or two rows:

ErrorMost likely cause
HTTP 403A firewall, WAF or bot protection blocks automated or data-center traffic
HTTP 429Rate limiting counts our checks as too many requests
timeout / context deadline exceededThe page took longer than the check allows, or a firewall silently drops our packets
connection refused / resetA firewall or proxy rejects connections from outside your network
keyword "…" not foundThe page loaded, but without the text you told the monitor to look for
certificate / x509 errorA certificate problem your browser works around but a strict client doesn't
too many redirectsA redirect loop, or a redirect to a login page
Failed in one region onlyGeo-blocking, or a network problem between that region and your host

1. A firewall or bot protection blocks the checker

This is the most common cause by far. Cloudflare's Bot Fight Mode, AWS WAF rules, Sucuri, Wordfence and fail2ban all treat requests from data centers with suspicion, and uptime checks come from data centers. You don't see it because your home IP looks human.

  1. Look for a 403, a 429 or a challenge page in the failed check.
  2. Check your WAF or security plugin's log for blocked requests at the alert time.
  3. Allow the checker: add our IP addresses or user agent to the allowlist. The allowlist guide has the current addresses and step-by-step settings for Cloudflare, AWS WAF and Wordfence.

Don't switch protection off

You only need an exception for the checker, not a weaker rule for everyone. Allowlisting by IP is the safest: user agents can be faked.

2. The page is slower than the check allows

A website check waits about 10 seconds. A page that takes 12 seconds is "working" in your browser, especially when parts of it are cached, and down for a monitor. Timeouts that come and go usually mean a slow database query, a cold start on serverless hosting, or a server running out of workers at busy times.

  • Compare the response times on the monitor's chart with the times of the failures.
  • Time the page yourself without the cache: curl -o /dev/null -s -w "%{time_total}\n" https://example.com
  • If a proxy sits in front, its logs show upstream timeouts. Our guide to upstream timeouts covers nginx; connection timeouts covers the network side.

A timeout with no response time at all, every time, points to a firewall dropping packets rather than a slow page.

3. The keyword isn't in the page the server sends

A keyword check passes only when the text appears in the HTML the server returns. Monitors don't run JavaScript, so text that a React, Vue or Angular app renders in the browser isn't there. Neither is text behind a cookie banner that replaces the page, a consent wall or a login.

To see what the monitor sees, fetch the raw HTML and search it:

bash
curl -sL https://example.com | grep -i "your keyword"

No match means the text isn't in the server's HTML: choose text that is, like the page title, a footer line or a heading that's rendered on the server. Also watch for small changes: a keyword is matched exactly, so a renamed button breaks it.

4. You're logged in and the monitor isn't

Dashboards, admin pages and members' areas redirect anonymous visitors to a login page. The monitor follows the redirect, finds a login form instead of your keyword, or hits a redirect loop. Monitor a public page instead, or add a small health endpoint that checks the app and its database and returns 200 without a session. For APIs, an API monitor can send an authorization header.

Loops have their own guide: too many redirects.

5. A certificate problem your browser hides

Browsers cache intermediate certificates and fill in a missing one quietly. An incomplete chain then works on your laptop and fails everywhere else, including for monitors, API clients and some phones. Expired certificates, and certificates for the wrong name, fail for everyone, but you may have clicked through the warning once and forgotten.

6. It fails in one region, or on one network

Checks come from several places. When only one region fails, the cause is between that region and your server: geo-blocking (some hosts block whole countries by default), a routing problem at one network, or a DNS record that differs by location. Your browser takes a different route, so it never sees it.

  • The failed check lists each region's answer: see which agree.
  • Try the page from several places at once with the multi-location checker.
  • Check country blocking in your firewall or CDN settings.

7. It really was down, briefly

A deploy, a restart or a crash that the process manager fixed in 30 seconds is a real outage for anyone who visited then. If the incident lasted a minute or two and lines up with a deploy, the monitor was right. Our guide to checking your site after every deploy covers zero-downtime releases.

Still unsure? Ask the assistant

In Spot Downtime, open Ask AI and ask "why is my monitor down?". It reads the monitor's last checks, errors and each region's answer, runs the check again, and tells you which of the causes above it is and how to fix it. For alerts that are real but too noisy, see how to cut false alarms.

Get new guides by email

Practical posts on monitoring, outages, error fixes and on-call, when they're published. No spam.

Unsubscribe from any email with one click. See our privacy policy.

Keep reading