Kanvass left WordPress. Its pages are less than half the weight, and every one now explains itself to search and AI.
Kanvass moved from WordPress to a custom-built site on Vercel. Every page now has a search description (10 of 40 before), images are described (40% before, 97% now), the average page carries 41 KB of code instead of 16, and 41 of the 42 old addresses still lead to a working page. The old site is gone, so we compared the 42 pages the Internet Archive captured on it with the live site, page for page.
From a WordPress site to a built-for-you one
The Internet Archive’s copies of the old pages (40 read in full), against the live site on 17 November 2026.
The old site could not be speed-tested, so we show what it carried, not how fast it loaded.
What changed for Kanvass
- Every project is its own page. All 20 are marked up as creative work, so what Kanvass has designed and built is readable by search and AI assistants, with breadcrumbs on 41 pages.
- Services and a journal. All 8 services are marked up as services, and the 6 journal posts as articles. The old WordPress blog became the journal.
- Tidy and lean. One sitemap of 42 real pages instead of two overlapping ones, and almost every image now has a description.
- Files built for AI assistants. A live llms.txt that lists the services, projects and journal, and a robots.txt that names 10 AI crawlers individually. The old file allowed everything and said nothing about AI.
And nothing of substance was lost
- 41 of the 42 archived addresses reach a working page on the live site. 35 redirect to their equivalent, for example /blog is now /journal and /projects/7saints is /projects/7-saints. The 5 that go to the home page are WordPress leftovers: a “hello world” post, an author page, a category and two gallery pages.
- The one address that returns 404, /services/fit-out, had its content removed in WordPress before the move, leaving an empty page. Leaving it unredirected removed a thin page that was hurting site health. Every Internet Archive copy from 2020 to June 2024 shows a site-acquisition page, so it was emptied after that.
- Every page on the new site has a canonical tag, a search description and exactly one main heading. In the archived pages, 10 of 40 had a description.
See every check
| URL and on-page SEO parity | Before (40 archived pages) | After (42 live pages) |
|---|---|---|
| Archived URLs that reach a working page on the live site | 41 / 42 | |
| Same URL / redirected to the equivalent page | 1 / 35 | |
| Redirected to the home page (WordPress leftovers) / not redirected (an empty page) | 5 / 1 | |
| Addresses that need two redirects to arrive | 16 | |
| Pages with a search description | 10 | 42 |
| Pages with exactly one main heading | 39 | 42 |
| Pages with a canonical tag | 38 | 42 |
| Pages with structured data | 38 | 42 |
Speed today
A baseline, not a comparison. The old site could not be measured, so there is no before figure and no claim of improvement. Lighthouse 12, median of 3 runs, 17 November 2026.
| Page | Device | Performance | LCP |
|---|---|---|---|
| Home | Mobile | 70 | 5.2s |
| Home | Desktop | 100 | 0.7s |
| Projects | Mobile | 83 | 3.8s |
| Projects | Desktop | 99 | 0.8s |
Readable by search and AI
| Old site (WordPress) | New site |
|---|---|
| Generic page and website markup, a search box and a person profile | Professional-service markup with the address; all 8 service pages marked up as services |
| No project or article markup on any archived page | All 20 project pages as creative work; Article on all 6 journal posts |
| 1 image in 3 without a description | 1 in 38 without a description |
| Two overlapping sitemaps | One sitemap of 42 real pages; breadcrumbs on 41 |
- llms.txt. Live, with who Kanvass is, contact details, the 8 services, the 20 projects and the journal, each with a link and a one-line description. The Internet Archive has no copy of one for the old site.
- robots.txt. Names 6 AI search and assistant crawlers and 4 AI training crawlers individually. The old file (2022 copy) allowed everything and said nothing about AI.
These make the site readable. We do not claim they raise citations or rankings by themselves.
Be straight about
There was no before snapshot. Old pages come from the Internet Archive (captures from 2022 to 2024). It covers only what it captured, and the old site may have changed before the switch. The one page it holds that the new site does not, /services/fit-out, is the page described above.
16 of the old addresses take two redirects to arrive. They work, but one redirect is faster, so collapsing them is a small clean-up.
The home page on a phone scores 70, with the main image taking 5.2 seconds. The other pages score 83 to 100. The home page image is the place to start.
Rankings, traffic and enquiries. This shows URL continuity and page quality, not an outcome.
What we check next
| When | What we run | What it shows |
|---|---|---|
| Week 1 | URL check and Lighthouse again | Nothing broke; home page speed looked at |
| Week 4 | Search Console trend for the redirected pages | No hidden ranking loss |
| Week 8 | Landing pages and enquiries | Outcome, reported separately |
How we measured this
- Chad’s SafeSwitch approach, adapted for a site that had already moved. Old addresses come from the Internet Archive index (HTML captures with a 200 response); each was requested on the live site.
- The Archive’s raw copy of 40 old pages and all 42 pages in the live sitemap were read for the same checks. Counts come from the HTML the server sends. Images with a description are images with non-empty alt text.
- This shows URL continuity and page quality, not rankings, traffic or bookings.
Moving your own site? See how SafeSwitch works, or more client results in the case studies.
How does your site look to Chad?
Drop your domain. In 11 seconds you get your GEO score, your biggest gap, and the first thing Chad would fix.