Website Uptime: What Actually Counts as Downtime?

Almost every hosting provider advertises an uptime figure, usually 99.9% or something close to it, and the number sounds precise enough to compare directly across providers. It isn't, quite. What actually gets counted as downtime, and what gets quietly excluded, varies between providers, and that fine print matters more than the headline percentage.

What an uptime percentage is actually measuring

Uptime is the share of time, usually over a month or a year, that a server successfully responds to requests. 99.9% uptime allows for roughly 43 minutes of downtime a month; 99.99% allows for about 4 minutes. These numbers are typically measured by the provider's own monitoring, checking the server at regular intervals, which means the measurement method itself is worth understanding before treating the figure as an independent guarantee.

What usually counts against the uptime figure

• The server failing to respond to requests at all
• Application-level errors caused by the hosting infrastructure, not your code
• Network outages on the provider's side that prevent visitors from reaching the server
• Database or storage failures that make the site unreachable

What's usually excluded, even though it feels like downtime to you

This is the part most people don't check until it matters. Scheduled maintenance windows, announced in advance, are typically excluded from the uptime calculation entirely, even though the site is genuinely unreachable during that window. DNS propagation delays after you change a domain's records aren't counted either, since they aren't the hosting server's fault. Problems caused by your own site, like a broken plugin, a bad code deployment, or a misconfigured redirect, also fall outside an uptime guarantee, because the server itself was up and answering requests the whole time.

Why measurement frequency changes what gets caught

If a monitoring system checks a server every five minutes, a two-minute outage that resolves on its own between checks may never register at all. This isn't necessarily dishonest, it's just a limitation of interval-based monitoring, but it does mean two providers with identically worded 99.9% guarantees can have meaningfully different real-world reliability depending on how closely, and how independently, that number is actually verified.

Distributed monitoring versus a single checkpoint

Some providers measure their own uptime from a single monitoring location, checking the server from one point on the internet. Others use distributed monitoring from multiple geographic locations, which catches network-level problems that only affect visitors from certain regions, something a single checkpoint would miss entirely. A regional network issue could leave part of the world unable to reach your site while a single-location monitor reports the server as fully up the whole time. Ask which model a provider uses if uptime for a geographically spread audience actually matters to you.

What redundancy actually buys, beyond the percentage

A published percentage doesn't tell you whether an outage, when one does happen, lasts two minutes or two hours. Redundant infrastructure, backup power, failover servers, and replicated storage, exists specifically to shrink that window, and it's part of what a higher-tier hosting plan is paying for that a bare-minimum setup often skips entirely. Two providers can both report 99.9% over the past year while having very different worst-case outage durations, because the percentage only tells you total accumulated time, not how it was distributed or how quickly a serious failure actually got resolved.

How to read an uptime guarantee before you rely on it

• Check whether scheduled maintenance is excluded from the calculation, and how it's announced
• Ask how uptime is measured, and by whom, rather than accepting the percentage alone
• Look for a status page with historical incident records, not just a live indicator
• Understand what remedy, if any, applies when the guarantee isn't met
• Confirm the guarantee covers the server responding, not full application-level behavior

Third-party monitoring can supplement a provider's own numbers

Independent uptime monitoring exists specifically because relying solely on a provider's self-reported figure means trusting the party being measured to grade its own homework. Running a free or low-cost third-party monitor against your own site, checking every few minutes from an outside vantage point, gives you an independent record you control, one that doesn't depend on a provider's own dashboard or reporting choices. For a site where downtime has a real cost attached to it, this independent record is worth having regardless of how good the provider's own guarantee reads on paper.

What a service-level credit usually is, and isn't

Most hosting uptime guarantees come with a service credit if the provider misses its own target: a partial refund or account credit for the affected period, requested by you rather than issued automatically in most cases. It's a reasonable gesture of accountability, but it's not compensation for lost sales or lost visitors during the outage itself. Treat it as a sign the provider takes reliability seriously, not as insurance against the actual cost of downtime to your business.

What actually matters more than the published number

A provider's track record, visible through a public status page and how it communicates during real incidents, tells you more than a percentage on a marketing page ever will. A company that posts clear updates during an outage and explains what happened afterward is giving you more useful information than one that only ever shows a clean green checkmark. When comparing hosts, ask to see how past incidents were actually handled, not just what the guarantee promises on paper.