Agentic website performance
AI Software Company Website · Live 14 November 2026
AI software company website
An Emergent AI website app got the new company website up and running quickly, but pages were assembled in javascript react, making the website slow and hard for Google to read. It now runs on a fast Agentic WordPress. Same design, same words, now in a agent supported CMS
Moved from Emergent AI
2.8× faster content appears−53% page weight
Google PageSpeed Insights, mobile, 16 November 2026 Run it yourself
- 97 Performance
- 96 Accessibility
- 96 Best Practices
- 100 SEO
94
was 62
+32Why they switched
Google couldn’t read it
A React app needs workarounds before search engines see a website at all. Search Console reported 63 pages not indexed against 50 indexed, on a site of around 63 pages.
Credits burned on basics
Simple changes ate through the AI builder’s credits, and fixes didn’t stick.
Measurement unclear
Tracking was installed, probably. Nobody could say whether anyone had signed up for the free course.
Silent breakage
One change broke the automatic blog feed. Ten posts went missing for weeks before anyone noticed.
What was needed next was a foundation for targeted marketing: hubs for each UK nation, sign-ups that can be counted, and remarketing to the care providers most likely to need it.
Home page
Site speed score
was 62 94
Content appears
was 6.6s 2.4s
Page weight
was 1,286 KB 604 KB
Across the site, median of 9 pages: 61 → 95 · 6.4s → 2.2s · 1,114 KB → 584 KB
Page by page
BeforeLive
Speed score on a phone (Lighthouse mobile performance), out of 100.
Why it’s faster
Half the script. The app built each page in the browser from 750–843 KB of JavaScript before anything showed. WordPress sends the page ready-made.
Images sized for the screen. The home page’s images went from 438 KB to 56 KB, as WebP at the size shown.
A page cache from day one. The server answers in about 0.12 seconds.
Also fixed at go-live
A real 404 page; the app answered any made-up address with the home page.
One address per blog post, duplicates redirected.
FAQ markup only where the page shows it.
Search engines and six AI crawlers welcomed by name.
Seven steps to better performance
- Agentic website migrationDone
- Faster optimisationNext
- Site redesign
- Demand mapping
- Topic hub structure
- Converting content
- Campaign building
Next on the list
Security
- Who looks after the servers, and how often they are patched
- Who has access: keys, users, old logins
- Domain and DNS in your own Cloudflare account
- Off-server backups with a tested restore
Measurement
- Six-monthly re-measure, cached and uncached
- Search visibility checked after launch
- Analytics and Search Console baseline
- Re-audit every six months
Faster optimisation
- Image sizes, dimensions and compression
- Caching and CDN
- Redirects and broken links
AI discoverability
- Organisation and author schema
- AI crawler access
- Descriptions on images that carry meaning
How we measured
The same 9 pages and the same settings at both ends: Lighthouse mobile, median of three runs, then the median page. Before was measured on 8 November 2026, after on 14 November 2026. A speed score that moved less than 5 points is called level. Both ends were measured with the same tracking tags on. SEO scores 100 on every live page.
Every measure
The results
Every measure, before and after, with "not yet" where it can't be measured.
| Measure | Before | After | What it means |
|---|---|---|---|
| Site speed score | 62 | 94 | +32 points |
| Content appears | 6.6s | 2.4s | 4.2s sooner |
| Page weight | 1,286 KB | 604 KB | -53% |
| Pages on the old site | 65 | ||
| Pages kept | 65 | ||
| Pages merged | 0 | ||
| Pages culled | 0 | ||
| Time taken | 7 days, from 6 November to go-live on 13 November 2026 | From the clock starting to the site going live. |
Lighthouse on a phone, the median of the sampled pages, three runs each: before on the old site, after on the new one, measured the same way. A score that moves less than 5 points isn't reported. Measured on 14/11/2026.
