How to Build a Website That Stays Online

“Always on” is a design, not a host slogan. Single shared boxes go down. Registrars get locked. Plugin updates brick admin. The sites that stay reachable have boring DNS, a CDN in front, backups that restore, and a change process that is not FTP on Friday. Hosts advertise 99.9% uptime. Your job is the other minutes plus your own mistakes.

One WordPress on one SiteGround or Bluehost account is one disk, one account suspension risk, one region. That is acceptable for most sites if backups and Cloudflare cache are real. It is not a high-availability cluster. Do not pretend otherwise. If the business cannot tolerate origin loss, you need a designed failover (hard) or a static fallback page on Cloudflare (easier) plus a host with fast restores (Kinsta, WP Engine). Cloudways snapshots help you return. They do not keep serving checkout during an AWS regional event unless you built that. Honesty about the single origin is how you stop buying magic SKUs. Draw one box. Own it. Then harden it.

Keep the domain at Namecheap or another registrar with 2FA, even if Bluehost gave you a free year. Hosting DNS is convenient until you migrate under duress. Cloudflare as DNS plus proxy is a common grown-up pattern: you can change origin IP without waiting on a host’s nameserver desk. Lower TTL before changes. Document MX so mail does not depend on a web host remaining immortal. DNS is how the site stays findable when you switch from HostGator to Cloudways. If the only copy of the zone is inside a suspended cPanel, you are offline in a more interesting way. Split registrar and compute on purpose.

Cloudflare hides the origin IP, absorbs some attacks, and caches static assets. That keeps many sites “up” for readers during origin blips if HTML cache is configured. Logged-in WordPress and carts still need the origin. A WAF is not a reason to skip PHP and plugin patches on A2 or WP Engine. Flexible SSL is still a trap. Use Full (strict) and Let’s Encrypt at origin. Orange-cloud the website, not the mail. Cloudflare is a layer. SiteGround or Kinsta remains the application host. Together they survive more days. Alone, either can still fail. Configure cache rules you can explain. Then keep patching WordPress like the proxy is not there.

FTP-over-live is how you stay offline during a theme upload. Use staging on SiteGround, WP Engine, Kinsta, or Cloudways. Git plus WP-CLI on hosts with SSH (DreamHost, A2, Cloudways) is even better. Deploy when traffic is low if you must touch production. Avoid plugin bulk-updates on Friday. Maintenance mode is a last resort, not a lifestyle. Database migrations need a backup first—the restore is part of staying online. If your process requires a one-hour window, you do not yet have a process that stays online. The host cannot save a deploy you did not stage. Reliability is a habit with a URL.

Password manager with registrar, Cloudflare, host, and WordPress. 2FA. Off-site backup. A monitor that texts someone who is not only you. Written restore steps. PHP on a current version. That stack on SiteGround or Cloudways beats a clever Kubernetes diagram you cannot run from a phone. Kinsta is boring in the expensive way: fewer knobs, more platform. Shared is boring in the cheap way: fewer knobs, noisier neighbors. Pick boring. Document it. Give a deputy access. Sites stay online when the person who knows the DNS is not the only person on a flight. Vacation is the availability test you actually take.

Control DNS at the registrar or Cloudflare, keep a restorable origin (SiteGround, Cloudways, Kinsta), patch WordPress, and deploy via staging. Hosts advertise 99.9% uptime; your single box and Friday FTP write the rest of the story. Boring wins.