"Four to six weeks" sounds like a marketing number until you have to hit it on a real project with a real client who has opinions. I hit it because the process is boring and repeatable, not because I work miracles. Here is exactly how a site goes from a first conversation to a live URL, with AI in the loop at every step and none of the steps skipped.
Week zero: decide before you build
The fastest way to blow a six-week timeline is to start designing in week one. Before anything, I run a short discovery: who the site is for, what one action it needs to drive, what the client sells and to whom, what they hate about their current site. I come out of it with a sitemap, a single primary conversion goal per page, and a reference site whose structure I am going to adapt rather than invent from scratch. This portfolio site is a close structural adaptation of a reference I admired. Starting from a proven skeleton is not cheating, it is how you skip a week of second-guessing layout.
AI helps here as a thinking partner, not a decider. I feed it the discovery notes and have it draft the information architecture and a first pass of page-level copy. Then I cut it in half. The generic version it produces is a useful map of every cliche to avoid.
Weeks one to two: copy and design together
I write copy and design at the same time, because a design without real words is a lie. Every build I ship includes SEO-optimized copy, drafted by AI against a keyword I have actually chosen, then edited line by line into something with a point of view. The AI gets me to a full draft in a day instead of a week. The editing is where fifteen years of judgement earns its keep: killing the filler, tightening the headline, making sure each page targets one query a human would really type.
Design is static HTML and CSS from the start. No page builder in the critical path, no theme to fight. I build against a small design system: two or three fonts, a tight colour set, real spacing rules. AI writes a lot of the markup, I own the taste. Because it is just files, I can preview the whole site the second it exists.
Weeks two to four: build it static, make it editable
The sites I ship are static: plain HTML, CSS and a single small JavaScript file, deployed to Cloudflare's edge through CF Pages. No database to hack, no server to patch, load times measured in milliseconds. Upwind Solutions, a yacht management company in Bali, was the first client I put on this stack. It is a static site on CF Pages, and the owner still edits every word himself, which brings me to the part people assume you lose with static.
Wire it to Orbit so the client can edit
"Static" usually scares clients because they think it means calling the developer for every comma. It does not, because I wire every site to Orbit, my own CMS. Orbit reads a set of templates with tokens where the editable content goes, renders valid HTML, and writes it back to the repository, which redeploys automatically. The client gets a form-based editor and a visual click-to-edit mode. They change a headline, hit save, and about ninety seconds later it is live on a staging link. Publish pushes it to production in a minute or two. No FTP, no plugin updates, no "the site is down" at midnight.
Every editable string becomes a token. Contact forms post to Orbit Leads, so submissions land in a proper inbox with AI-drafted replies instead of an email that gets lost. Turnstile handles spam without a captcha most visitors ever see. This is the step that makes static sites viable for normal clients, and it is why I built Orbit in the first place.
Weeks four to six: QA, performance, search, ship
Now the unglamorous finishing. I run Lighthouse and hold the build to a real performance budget, aiming for scores in the high nineties, because on a static edge-served site there is no excuse to miss. Every site ships with the AI-search files done properly: a robots.txt that welcomes GPTBot, ClaudeBot and PerplexityBot, an llms.txt describing the business, and structured data so search engines and chat assistants can quote the site cleanly. Analytics goes on from day one through TrackTrendy, so we are measuring from the first visitor.
Then a staging link protected behind a one-time PIN, a round of client review, fixes, and the domain cutover last. The whole thing, discovery to live, fits in four to six weeks because every stage has a default I trust and AI removes the slow, manual middle of each one.
The timeline is not fast because I rush. It is fast because every step has a proven default and AI deletes the busywork inside it.
Why it holds up
People assume speed means corners cut. The opposite is true here. Tested code, monitored uptime, real performance scores, SEO baked in, a client who can actually run their own site. Four to six weeks is not a discount on quality. It is what happens when you remove the parts of the old process that were slow for no good reason. The parts that need a human, discovery, taste, editing, judgement, still get a human. Everything in between got faster.
Have a project with a deadline that used to sound impossible? Tell me about it, or see what four to six weeks looks like.