Rebuilding a 690 page regulated site, and taking search clicks up 26% in a week

I co-founded Claims Bible, so this isn’t a client reference. It’s something I built myself, including the plan, the build, the content, the numbers and the decisions. It was a great thing to work on because I had full autonomy, which you never really get with a client, and it means I can show you all of it rather than the version a client would let me publish.

Where it started

The site had grown up on WordPress over five years. Around 1,320 files, a plugin stack nobody could account for, and years of accumulated URL debt from sections that had been renamed, merged or quietly abandoned.

Three things had to happen at the same time, which is why it ended up being a rebuild rather than a redesign.

It had to get faster, because most of the traffic arrives on a phone from a Google result and people leave if they’ve got to wait for it. It had to get accurate again, because content written over five years had drifted out of line with the rules, so there were interest rates that had changed, injury tariffs that had been superseded, and partner firms that had stopped buying while their pages were still sending them enquiries. And it had to hold its search visibility through the move, because on this site search is more or less the whole business.

What I built

A build system rather than a CMS. Around 2,800 lines of Node that generate the whole site, so that’s 670 to 690 pages depending on the configuration, plus the redirect map, the sitemap, the robots file and the security headers. Content lives in a source folder, the output goes to a build folder, and nobody edits the build folder by hand. If something’s wrong on the live site it’s wrong in the source, which means you find it and fix it once instead of on 129 pages.

Feature flags for regulatory state. A whole claim category can be switched off at build time. Set CAR_FINANCE_LIVE=false and every car finance page rebuilds without a form, without a call to action and without a partner link, while keeping the guidance that actually helps people. That matters because in a regulated sector “stop soliciting for this claim type by Friday” is a real instruction, and the alternative is editing 150 pages by hand and hoping you got them all.

An audit that blocks bad builds. A script that has to pass with zero errors before anything deploys. It checks title and description lengths, orphaned pages, broken internal links, missing structured data, and a set of specific regulatory conditions. One of those fails the build if a decommissioned phone number turns up anywhere on the site, so it can’t come back by accident.

A worker for the dead URLs. Five years of WordPress left a lot of paths that shouldn’t resolve and shouldn’t redirect either, because redirecting a retired claim type to a live one sends somebody to a page about a claim they weren’t asking about. A Cloudflare Worker returns 410 Gone for 98 of those, alongside a resolver that maps the genuine legacy URLs onto their new homes.

Deploy guards. The deploy script refuses to run against the wrong Cloudflare account, and refuses to push a build that still has staging’s noindex header on it. Both of those are mistakes I could see myself making at half past ten at night.

Consent-first measurement. A cookie banner with analytics denied by default, GA4 set up for consent mode, and the events that matter defined before launch rather than bolted on afterwards. It means measuring less, but doing it lawfully and knowing what the numbers actually mean.

The result

The site cut over on 19 August 2026. Comparing the week before with the week after:

Two things I’d rather tell you than have you find. The week after that, 27 August to 2 September, clicks came back to 443, below the week before cutover, so the 26% is a one-week lift and not a new level. Over the 18 days after cutover against the 18 before, clicks were up 7%. And average position got slightly worse in the same comparison. When I first wrote this up, on the day, Search Console showed 442 to 563 and a better position, and Google has revised the figures since. The numbers here are what it showed on 8 September 2026, and the chart is drawn from its export.

On the tracked keyword set the site passed moneysavingexpert.com on 1 September and was still ahead on 8 September, 40.2% against 39.2%. That’s the result I’d point to over the click count, because it kept going.

Indexed pages fell from 1.1k to 751, against a target of roughly 670, which is what I wanted to happen. A lot of what the old site had sitting in Google was duplication and thin pages competing against each other.

A later pass on the redirects recovered around 470,000 impressions of traffic that had been landing on 404s in the personal injury section.

The whole site was reviewed and signed off by the external FCA compliance consultant on the day it went live.

How it was built

I specified it, designed it, tested it and I own it. The code was written working alongside Claude, which is how I build now.

I mention that because it’s the honest answer, and because it’s the bit people tend to get wrong. Working this way doesn’t remove the need for judgement, it concentrates it. Somebody still has to know that a 410 isn’t the same as a 301, that a fuzzy-matched lender redirect is a regulatory problem rather than a tidy-up, and that a build which can’t be stopped by its own audit is a build waiting to go wrong. That’s the job, and it always was. The typing is just the part that got quicker.

What this looks like for your business

Most of the sites I get asked to look at have the same three problems in a smaller form. They’re slow, they’re cluttered with pages competing against each other, and they’re carrying claims nobody has checked in years.

The approach is the same whatever the size. Work out what the site is actually for, build it so it can be changed safely, and then measure whether it worked.