A Website From Scratch, in 2026

I have had websites built for more than twenty years, across a portfolio that at one point ran to about 250 platforms. When I rebuilt this site in 2026, I started from a blank page rather than from the last version. This is the list I worked from, with what each choice looks like in practice here.

1. Build pages, not an application

Most company websites are documents: who we are, what we do, proof, how to reach us. They do not need a JavaScript application framework running in the visitor’s browser to show text. This site is built with a static site generator. Every page is produced as finished HTML at build time and served as a file.

The result is that the whole site ships about 1.6 KB of JavaScript in total, and nothing on it needs that JavaScript to be read. Pages are fast because there is almost nothing to do, and every crawler, including the many that do not run scripts, receives the complete page. The detail on crawlers is in What a Crawler Actually Receives.

If your product is software, build the software as an application. Build the website that sells it as pages.

2. Write the facts down once, in one place

The facts about a business change slowly and get repeated everywhere: names, roles, dates, education, locations, ventures, numbers. On this site they live in data files, and every page that mentions them reads from the same source. When a fact was wrong, as one headcount figure was, it was corrected in one file, and the correction reached every page that used it.

The same facts feed the structured data. Each page carries a graph describing the person, organisations, places and the page itself, linked by identifiers, and it matches what the page visibly says. That is for search engines. The single source is for honesty.

3. Put evidence on the page

Answer engines and buyers want the same thing: claims they can check. The articles on this site quote the emails they are based on, with dates. The ventures list what each business was, including the ones that are dormant or no longer mine. Where something was sold or paid for, the page says so.

Evidence is the most expensive content to produce because nobody can fake it for you. That is also why it is the most valuable.

4. Never break an address

Every URL that has ever been published is a promise. When this site’s old blog moved, each of the seventeen old article addresses got a permanent redirect to its new home, in one hop, with no chains. Redesigns that drop old URLs throw away links that took years to earn and send crawlers into dead ends.

5. Publish only what is finished

Drafts exist in the same system as published articles, marked as drafts, and are excluded from the production build, the sitemap and the index entirely. Nothing half-written leaks out because someone forgot to hide it.

6. Measure enquiries, not visits

The enquiry form works without JavaScript and posts to the host directly, with a hidden field to catch spam bots. When an enquiry is sent, one event records it, with the topic chosen. With no analytics provider configured, the site makes no third-party requests at all.

The number I care about is not sessions. It is who wrote in, about what, and what came of it.

7. Own all of it

The site is plain files in a repository I own. No page builder, no proprietary theme, no plugin that stops being maintained. The one host-specific piece is the form handler, which is a few lines to replace. If the host disappeared tomorrow, the site would move.

None of this is new. Several of these rules were already good practice when I was building for Google’s list of ten links. What changed is that there are more systems reading the result now, and they are less forgiving than a person in a browser.