Kartuli left Site123. The booking page went from 65 to 100 on a phone, and every page kept its address.
Kartuli’s old site was built on Site123, a hosted drag-and-drop builder. It now runs on a custom-built site: the home page reaches its main content in 2.1 seconds on a phone instead of 7.1, the page where guests book scores 100 instead of 65, and all 26 old pages are still at the same address, with no redirects needed.
From a website builder to a built-for-you site
Old site against new, on the live domain, 16 November 2026. Lighthouse on a phone, median of 3 runs.
Lab figures, not real-visitor data. The booking page gained the most, and it is the page that earns revenue.
What changed for the restaurant
- A booking page that works on a phone. Guests book on the site itself. That page went from 65 to 100.
- Menus, events and the story as real pages. Four menus and ten tasting-club events are each marked up so search and AI assistants can read what is on offer, with the address, opening hours and a plain description of the restaurant.
- A site that is easy to keep current. Changes come as a preview and go live on approval, rather than through a page builder.
- Files built for AI assistants. An llms.txt file and a robots.txt that names 11 crawler groups individually. The old site had neither.
And nothing was lost along the way
- Every URL in the old sitemap returns 200 on the new site, including the awkward ones (/menu/lunch-menu-1 and the tasting-club event pages). Both the slash and no-slash forms work.
- No page is set to noindex. Every page has one canonical tag, a meta description and exactly one H1. The old site had one page with a different H1 count.
- Unknown URLs return a real 404, not a soft 200, so the new site does not create duplicate-content noise.
See every check
| URL and on-page SEO parity | Result |
|---|---|
| URLs from the old sitemap that return 200 on the new site | 26 / 26 |
| Same URL, unchanged / needing a redirect | 26 / 0 |
| Redirected to the home page (soft loss) / lost (404 or error) | 0 / 0 |
| Trailing-slash gaps | 0 |
| Pages set to noindex / missing a canonical tag | 0 / 0 |
| Pages with a meta description | 26 → 26 |
| Pages with exactly one H1 | 25 → 26 |
| Page titles rewritten (expected on a rebuild) | 26 |
Lighthouse, page by page
| Page | Device | Performance | LCP | Blocking time |
|---|---|---|---|---|
| Home | Mobile | 66 → 99 | 7.1s → 2.1s | 99 → 0 ms |
| Home | Desktop | 95 → 100 | 1.2s → 0.7s | 0 → 0 ms |
| Book a table | Mobile | 65 → 100 | 11.9s → 1.7s | 64 → 20 ms |
| Book a table | Desktop | 88 → 100 | 2.1s → 0.4s | 0 → 0 ms |
| Our story | Mobile | 79 → 99 | 4.7s → 2.1s | 63 → 34 ms |
| Our story | Desktop | 95 → 100 | 1.1s → 0.7s | 0 → 0 ms |
Readable by search and AI
What machines can read, old site against new.
| Old site (Site123) | New site |
|---|---|
| A search box and WebSite markup on every page (the search box does not exist) | Restaurant, address, geo-coordinates and opening hours on the home page |
| Event and Offer on the tasting-club pages | Event kept on all 10 event pages, with Place and address |
| Article on 4 menu pages | Menu, MenuSection and MenuItem on 4 menu pages |
| No FAQ markup | FAQPage on 4 pages |
| No breadcrumbs | BreadcrumbList on 27 of 28 pages |
- llms.txt. Live, with the address, phone, booking link and a plain description of the restaurant. The old site had none.
- robots.txt. Names the AI search and assistant crawlers (such as OAI-SearchBot and Claude-SearchBot) and the training crawlers (GPTBot, ClaudeBot, Google-Extended) individually: 11 user-agent groups, against one generic rule on the old site.
- Sitemap. 28 URLs against 26 on the old site.
These make the site readable. We do not claim they raise citations or rankings by themselves.
Be straight about
The kit reported “schema lost” on 26 pages, which is mostly a false alarm: the old SearchAction and WebSite markup was Site123 boilerplate, and the search box it described does not exist. WebSite now sits on the home page only. Organization was on 14 old pages and is now part of the home page Restaurant entry. That was a deliberate trade, and it can be reversed if the client wants it. All 26 titles changed, as expected on a rebuild, and the new ones are more specific. Watch organic landing pages for the home and menu pages in weeks 1 to 4.
Rankings, traffic and bookings. This case study shows that nothing was lost and the site is faster. It is not an outcome.
What we check next
The same checks, re-run on the live domain, with outcomes reported separately from parity.
| When | What we run | What it shows |
|---|---|---|
| Day 0 (done) | Compare on kartuli.co.uk, as above | The switch itself was clean |
| Week 1 | URL parity and Lighthouse again; confirm the sitemap is read once Search Console is verified | Nothing broke in production |
| Week 4 | Organic landing pages against the old 90 days; Search Console trend since verification | No hidden ranking loss |
| Week 8 | Organic landing pages and bookings, plus Search Console from its baseline | Outcome, reported separately from parity |
How we measured this
- Chad’s SafeSwitch kit. A snapshot of the 26 sitemap URLs on the old site, taken on 16 November 2026 from www.kartuli.co.uk before the switch, compared the same day against the new site on the live domain after DNS moved to Vercel.
- Lighthouse: median of 3 runs, home page plus two other pages, mobile and desktop.
- The kit reads the HTML the server sends, not the page after JavaScript runs.
- The check covers the 26 sitemap URLs only. The new site has 3 deliberate redirects, none on those 26.
- We make no zero-downtime or ranking promises.
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.