Oh Dear Uptime Report
The State of Website Uptime
We watch websites for a living. Here's what 1,817,403 real outages tell us: most are over in minutes, but downtime reaches almost every site, and it's the slow tail that does the damage.
These figures come from 1,817,403 outages across tens of thousands of monitored sites, March to December 2025. Everything here is real telemetry, aggregate and anonymized. We refresh it each year at this same address.
- 98.71%
- average uptime, all monitored sites
- 1.9min
- median time to resolve an outage
- 98.6%
- of outages resolved within the hour
- 61.4%
- of sites had at least one outage
Most outages are over in minutes
The typical outage is short. Half are resolved in under two minutes, and 98.6% are back within the hour. Only 1% run between one and six hours, and it's the slowest 0.3% past six hours that do the real damage.
When outages start
Outages start around the clock. About 68% begin outside the 9-to-6 workday, on weeknights and weekends, when fewer people are watching, and the volume barely dips overnight. Things break at every hour; coverage rarely does.
How fast the web responds
A site can be online and still feel slow. Across every HTTP site we monitor, the median full response, DNS lookup, connection, TLS handshake, server processing and download combined, lands at 159 ms. The slow tenth of requests is where your visitors feel it.
Downtime is common. 61.4% of monitored sites went down at least once, and the average site logged 47 outages over the period. Most were brief, half resolved in under two minutes, but every one is a moment a visitor hit a site that wasn't there.
Fast, accessible, findable
Staying online is only the baseline. A page can load slowly, lock out screen-reader users, or sit invisible to search engines while never once going down. We run Lighthouse on every monitored page, on desktop and on mobile, over the last 6 months. Desktop holds up; mobile is where the same sites fall down, performance drops from 78 to 57 and first paint nearly triples.
| Lighthouse audit | Desktop score | Mobile score |
|---|---|---|
| Performance | 78/100 | 57/100 |
| Accessibility | 89/100 | 90/100 |
| SEO | 91/100 | 95/100 |
| First paint | 832 ms | 2,347 ms |
What the nines cost
"Three nines" sounds reassuring until you do the maths: 99.9% uptime still allows nearly nine hours of downtime a year. Here's what each target allows, and what those hours cost, in people and in money.
| Uptime target | Downtime per year | Downtime per month |
|---|---|---|
| 99% | 3.65 days | 7.2 hours |
| 99.9% | 8.8 hours | 43.8 minutes |
| 99.95% | 4.4 hours | 21.9 minutes |
| 99.99% | 52.6 minutes | 4.4 minutes |
| 99.999% | 5.3 minutes | 26 seconds |
The human cost
67.7% of outages start outside business hours, so the average site sees roughly 32 of them a year land on a weeknight or weekend. Each one is a wrecked evening and an engineer pulled away from dinner. Stack them up over a year and that's how good teams burn out.
The business cost
What an hour of downtime costs is relative to your size. For a small SaaS on €10k MRR it might be a few hundred euros and a handful of churned trials; for a large retailer mid-sale it runs into six figures an hour. Whatever your number, it's checkout sessions abandoned, API calls dropped, and trust you don't get back.
Six-figure end of the range: Uptime Institute outage analysis, not Oh Dear telemetry.
More than uptime
A site fails in more ways than going offline. Here's a sample of what else we watch, and why each check earns its place.
Broken links
of the average site's links lead nowhere
Dead links frustrate visitors and bleed away search ranking.
Mixed content
of HTTPS sites served at least one insecure asset
A single http asset breaks the padlock and the trust on an otherwise secure page.
Domain expiry
of domains came within 10 days of expiring
A lapsed domain is the most avoidable catastrophic outage there is.
Scheduled tasks
of cron jobs failed or missed a run in the last 30 days
A backup that silently stops running is only discovered the day you need it.
How we measured this
Every figure here comes from Oh Dear's own monitoring over March to December 2025: 1,817,403 confirmed outages across tens of thousands of sites. An outage is a period where a site failed its uptime check; we count from the first failed check to the first successful one again. Sub-minute blips are excluded.
Timing uses each site's own local timezone, not UTC, so "business hours" means 09:00–18:00 where the site's team works, and weekends are Saturday and Sunday in that same local time. Response times are the full request measured on every HTTP check, reported as percentiles across all sites; Lighthouse scores cover the last six months, and the scheduled-task figure is a rolling 30-day snapshot.
Everything is telemetry from real checks, not a survey. All numbers are aggregate and anonymized; nothing here identifies a customer, a domain, or a site. You're free to cite these figures with attribution to Oh Dear.
Hand these numbers to your AI
Writing about uptime? Copy the whole report as clean Markdown, every figure, source and method, and paste it straight into ChatGPT, Claude or any other model.
