Web Hosting Uptime: What You Need to Know

Uptime is not a personality trait of a brand. It is architecture, maintenance windows, and whether you notice before customers do. Shared hosts print 99.9% because the number is familiar. Independent monitors and status pages are how you stop arguing with marketing. This is the skeptical reading of SLAs in 2026.

HostGator, Bluehost, and most of the category advertise 99.9% uptime. Some publish 99.99% on higher tiers. Those figures are contractual claims with definitions: what counts as downtime, whether DNS is included, whether your plugin crash counts. They are not a measurement you ran from three continents. Put UptimeRobot, Checkly, or similar on the homepage and checkout. Alert on something a human will answer. If your monitor and the host status page disagree, believe the user-facing URL. Kinsta and WP Engine are not exempt from physics; they are better at talking about incidents. A claim without a probe is a brochure.

Typical remedies are days of service credit after you file a ticket. That does not refund a missed launch or a weekend of phone support for your customers. Read the exclusions: upstream, DDoS, scheduled maintenance, “beyond our control.” SiteGround, A2, and Cloudways each write this differently. For a business site, the useful number is your cost of downtime per hour, not the credit. That number tells you whether Kinsta is cheaper than shared. If the SLA requires you to notice and open a ticket within a tight window, calendar that. Credits are not insurance. Architecture and backups are. Buy the host for restore and support; treat the SLA as a footnote.

A public status page is a courtesy. Silence during an incident is data. Subscribe to the host’s status if they have one (Kinsta, WP Engine, Cloudflare, AWS, Google Cloud all do in some form). Your probe should still hit the site, not only the provider’s happy dashboard. DNS failures at the registrar look like host downtime. Cloudflare failures look like origin downtime. Split the checks: origin IP (careful with security) versus the public hostname. When SiteGround chat says “all systems go” and your URL does not load in LTE, you have a local, DNS, or account issue—or they are behind. Document timestamps. Good incident hygiene beats arguing in social media replies.

Planned reboots, live migrations, and network work often do not count against the advertised 99.9%. That can be reasonable. It is not “always on.” Read whether they notify, and how. Cloud and Google Cloud under Kinsta still have maintenance. Shared hosts may move you between machines with a brief blip. If you cannot tolerate a short window, you need redundancy (more than one origin, or a queue) rather than a stern email to support. Schedule your own deploys away from known windows when you can. Maintenance is not an insult. Unannounced maintenance is. Ask pre-sales how they notify. The answer quality is part of uptime.

A single shared account is a single failure domain. Cloudflare can serve cached HTML if you configured it and the incident is origin-only; it cannot invent a cart. Multi-region AWS is a design, not a checkbox on Namecheap shared. Database replicas, object storage for media, and DNS failover are projects. Most small sites should buy a better single origin (Cloudways, Kinsta) plus Cloudflare plus backups rather than pretend they run a failover pair. Assumption is “the host is clustered so I am fine.” Configuration is “I know where the single disk lives.” Hosts advertise 99.9% uptime; your architecture is what happens in the other minutes. Draw the diagram. If it is one box, own that fact.

Probe the URL yourself. Treat 99.9% as a claim with credit-sized remedies. Status pages and Cloudflare help; they are not a second origin unless you designed them that way. Buy restore and support. Uptime is a system, not a slogan.