I recently read a hosting provider's marketing page that claimed "99.99% guaranteed uptime." It was a shared hosting plan. £3.99 a month. I nearly spat out my coffee.
Not because the number is impossible - it's not - but because the word "guaranteed" is doing a lot of heavy lifting in that sentence. When you actually read what it means, it's a lot less comforting than it sounds.
The Maths Nobody Does
Let's start with what those percentages actually look like in practice:
| Guarantee | Allowed downtime per year | Per month |
|---|---|---|
| 99% | 3 days 15 hours | 7 hours 18 min |
| 99.5% | 1 day 19 hours | 3 hours 39 min |
| 99.9% | 8 hours 45 minutes | 43 min |
| 99.99% | 52 minutes | 4 min |
That 99.9% everyone advertises? Nearly 9 hours of downtime a year and they've kept their promise. A full working day where your website is offline, spread across 12 months. If you sell anything online, that's a lot of missed orders.
And 99.9% is the good tier. Most budget hosts guarantee 99% or 99.5%, which is literally days of acceptable downtime per year.
What "Guarantee" Actually Means
Here's the bit nobody reads. When your host guarantees 99.9% uptime, they're not making a promise that your site will be up. They're saying: if it falls below 99.9%, you can claim a credit. Not a refund. Not compensation. A credit on next month's hosting bill.
The credits are usually pathetic. 10% off for missing 99.9%. 25% off for missing 99.5%. So your site is down for 12 hours, you lose orders and search rankings and customer trust, and your reward is £2 off next month's £20 hosting bill.
It's less "guarantee" and more "we'll give you a discount if we do a bad enough job."
The Fine Print Gets Worse
Most SLAs define "downtime" in ways that would make a lawyer proud.
Scheduled maintenance usually doesn't count. Your host can take the server offline every Tuesday night for an hour and it's not considered downtime. Some hosts schedule maintenance windows weekly - that's potentially 52 hours a year of planned outages that don't touch their uptime number.
Network problems between their data centre and your visitors? Often excluded. The server was technically running, so it's "up" as far as they're concerned. The fact that half your users couldn't reach it is apparently a different problem.
DDoS attacks get excluded by a lot of providers too, which is a bit like a security company saying "we guarantee protection except when someone actually tries to break in."
And here's my favourite: most SLAs require you to notice the downtime and file a claim within 24-48 hours. If your site went down at 3 AM on a Saturday and you didn't check until Monday morning, tough luck. The window's closed.
Some hosts don't even measure what you'd expect them to measure. They track server uptime, not application uptime. Your server's operating system could be running perfectly while your website throws 500 errors on every request - and the host considers that "up." Brilliant.
The Obvious Problem
Your host reports their own uptime numbers. That's it. That's the whole system.
It's like letting students grade their own exams. Even assuming they're honest (and they usually are), their definition of "up" and yours might be completely different things. They're checking whether the server process is running. You care whether your customers can load your website.
The only way to know your actual uptime is to measure it yourself, from outside your host's network. An external monitor that hits your site every few minutes and records what happens. When it goes down at 3 AM, you find out at 3:01 AM instead of when you open your laptop on Monday.
Incidentally, this is what WebMon does. I built it partly because I wanted real data on my own hosting, not just a number on a provider's status page.
What I'd Actually Recommend
Forget arguing about uptime percentages with hosting support. Just monitor independently and keep the receipts.
Track response times alongside uptime, because a site that takes 12 seconds to load might as well be down - and slow performance is often the canary in the coal mine before a full outage. Keep an eye on more than just HTTP status codes too. Your site can return a perfectly healthy 200 while displaying "account suspended" or a database error. A keyword monitor that checks for actual page content catches those cases.
And keep your monitoring logs. If you ever do need to make an SLA claim, having timestamped evidence of every outage makes it straightforward. More importantly, it helps you figure out when it's time to switch hosts entirely.
The real value of monitoring isn't getting £2 credits from your hosting provider. It's knowing what's actually happening with your site before your customers tell you.
Curious what your actual uptime looks like? Add a free monitor and find out. You'll have real data within a week.