Web hosting uptime SLAs: what 99.9% actually means

Ninety-nine point nine percent sounds almost perfect until you translate it into hours of downtime a year. Before you buy on an uptime badge, find out what counts as downtime, who measures it, and what the credit is really worth.

Hosts put big uptime numbers next to the price for a reason. Review sites copy them. Shoppers treat 99.9% as proof the site will stay up. The percentage is only useful if you know the measurement window, the exclusions, and whether a miss pays you anything that matters. If you are already comparing hosts without leaning on top 10 roundups, pair this with how to choose web hosting without relying on top 10 lists and support quality. For why sales copy and review copy so often rhyme, read The Truth About Hosting Review Sites.

Translate the percentage into time

Rough annual downtime if the SLA is met exactly:

99% is about 3.65 days a year
99.9% is about 8.8 hours a year
99.99% is under an hour a year

Those numbers assume a clean calendar year and no exclusions. Real contracts rarely work that way. A host can miss your busy weekend, still land inside the monthly target, and you get nothing. Or they can exclude the outage that hurt you most because it was "scheduled," "upstream," or "outside the network edge they monitor."

Ask for the number in hours for the billing period they actually use, usually a month, not a marketing year.

Ask what counts as downtime

Before you buy, send the same short question to every host on your shortlist:

On this plan, what does the uptime SLA measure? Is it ICMP ping, HTTP from your edge, or something else? What is excluded (maintenance, DDoS, third-party DNS, my code, my plugins)? How do I prove an outage, and what credit do I get if you miss?

Save the reply. "We guarantee 99.9% uptime" without those details is a badge, not a promise.

Push on the boring edges:

Does planned maintenance count against the SLA, or is it carved out if they post a notice?
If their upstream provider fails, is that their problem or yours?
If DNS is at a third party and the site cannot resolve, does the SLA still apply?
If the server answers ping but your site returns 500s because PHP or the database is wedged, is that covered?
Is the target measured per server, per account, or across their whole fleet?

A high percentage that only covers "the network was reachable" can still leave your shop closed.

Credits are usually tiny, and often hard work

Most shared and VPS SLAs do not pay cash. They pay service credit, often a slice of one month, and only if you open a ticket inside a short window with logs they accept.

That means:

A four-hour outage that costs you sales may earn a few dollars of hosting credit
If you never file, you get nothing even when they clearly missed
Some hosts require their own monitoring as the only evidence, so your screenshots do not count

Ask how credits are calculated, whether they stack, whether unused credit expires, and whether you have to request them. If support cannot explain the claim steps before you pay, assume the credit is decorative. Same honesty test we use for money-back guarantees and backup restores: a claim without a workable path is incomplete.

Whose monitoring are we talking about?

Hosts like to show green status pages. Those pages usually watch their view of the world: their data center, their network, sometimes a handful of synthetic checks they control.

Your visitors sit somewhere else. A regional routing mess, a DNS outage at the registrar, a CDN misconfig, or a certificate that expired overnight can take you offline while their status page stays green.

Before checkout, check:

Whether they publish a status page with history, not only a current green light
Whether you can export or view account-level uptime for your own sites
Whether they allow or encourage independent monitors (UptimeRobot, Pingdom, Better Stack, and so on)

Run your own monitor for a week on a trial or staging site when you can. Their number and your number will not always match, and the gap is the story.

Shared hosting makes SLAs softer

On a busy shared box, your neighbors matter. An SLA that looks solid on paper can still feel like a slow site during someone else's traffic spike, even when the host never calls it "down."

If you are shopping shared plans, ask about:

Fair-use CPU and I/O limits, and what happens when you hit them
Whether noisy neighbors get moved, or whether you get throttled first
How often they migrate accounts off overloaded machines

Overselling is part of this picture. A pretty uptime badge does not fix a server that is packed too tight. If you need firmer isolation, you are already looking past the cheapest shared tier, same as when you outgrow free hosting for anything that has to stay up.

Red flags worth pausing on

Be careful when:

The SLA is on the sales page but missing from the terms, or gutted by exclusions that cover every realistic failure
Support will not put measurement method and credit math in writing before you pay
Scheduled maintenance is unlimited as long as they emailed you
Only their monitoring counts as proof
Credits require a phone call with a retention team
Reviews praise "99.9% uptime" but never mention anyone receiving a credit

Useful hosts make uptime boring and measurable. Useful reviews talk about exclusions. If neither does, you are reading advertising.

We care about this for the same reason we care about renewal honesty and refund honesty: the real product is the plan that stays up for your visitors, not the badge next to the price. Our FAQ on how we rank hosts prefers evidence over splash ratings, and an SLA you cannot verify or claim is just another sticker.

The best course of action is still the boring one. Shop around, ask the same uptime questions of every shortlisted host, and treat any percentage that cannot survive a written reply as marketing until proven otherwise. Browse more web hosting guides and keep the homepage ranking list handy while you compare.