← Back to blog

The web that people talk about is not the web that actually exists

Developer Twitter is React, Next.js, and agents. The live web is still WordPress, jQuery, and PHP. Like a UK high street, most of it is legacy stock with room to upgrade.

Web development AI Strategy
The web that people talk about is not the web that actually exists

If you spend much time on X, you’d be forgiven for thinking the entire web is React and Next.js (or run completely by agents), and founders who prompt new companies into existence before breakfast. I spend a lot of time looking at various actual websites and they are often still jQuery and WordPress and have been for years.

Now, this might not be the most optimal stack for a website these days, but it says a lot about the state of the web and how a tool like Wordpress has remained accessible to ordinary people for years by being free and open source.

Two webs

To reiterate the point, there are two internets: the one that people talk about, and the one that actually exists.

The talked-about web lives in videos, blogs, and timelines. It is React, Next.js, TypeScript, Cloudflare, agents in the editor, and a landing page that looks like every SaaS startup. If you only listen there, you would think the old stack was already dead.

The web that actually exists is what you find when you have to search for small local businesses. For example if you want to join a local club or find a local plumber. It is what a crawler finds when it trawls the multitude of websites beyond the first page of Google. It is WordPress, jQuery, PHP, Apache or Nginx, a theme from 2017, and a contact form that still somehow works. That web is slower to change and it is where most businesses still live.

To confuse the internet depicted in many social media timelines with the one that actually exists for small business is short-sighted. A founder who shipped a SaaS dashboard in Next.js is not proof that your local florist, solicitor, or parish council needs the same stack. That said there is definitely room for cost effective and appealing upgrades to much of the existing web.

The talked-about web vs the crawl

State of JS 2025 asked 13,002 self-selected developers what they have ever used. React came back at 85%. Stack Overflow 2025 asked who used it extensively last year: 44.7%. W3Techs, late August 2026, asked what is on the site: React is on about 6% of all websites.

React, depending on who you ask

Hover or tap a bar for the exact figure.

State of JS 2025 · Stack Overflow 2025 · W3Techs JS libraries. W3Techs figure is share of all websites.

On the CMS side, W3Techs in September 2026 still puts WordPress on roughly 40% of all websites (about 59% of the CMS market). About 32% of sites use none of the CMSs W3Techs monitors — custom apps, undetected builders, and some static sites mixed together, not a pure-HTML count. The closer static proxy is static files: about 2% of sites with a known server-side language, meaning no PHP or similar on the origin. jQuery is on about 66% of all sites. Among WordPress sites whose JavaScript library is known, jQuery is effectively universal at 99.4%. PHP is still the server language for about 70% of sites where a language is known. Bootstrap sits on about 14% of all websites; Tailwind is on about 0.3%. WooCommerce alone is on about 8% of all sites — a WordPress plugin, so it already sits inside that WordPress bar, and it still outranks Shopify on the crawl. After WordPress, the next biggest standalone CMS is Shopify at about 5%. The no-code builders people talk about are real but still small next to that: Wix about 4%, Squarespace about 2.5%, Webflow under 1%, and Astro about 0.2%. Fast for the sites that use them. Barely visible in the census.

What the public web actually runs

Hover or tap a bar for the exact figure.

Legacy / mass marketModern / talked-aboutMixed / unclear

W3Techs PHP · W3Techs CMS overview · W3Techs static files · W3Techs jQuery · W3Techs Bootstrap · W3Techs JS libraries. Approximate share of all websites, late summer 2026. PHP and static files are share of sites with a known server-side language. No CMS means none of the systems W3Techs monitors. WooCommerce runs on WordPress, so it overlaps that bar. Red = legacy / mass market; black = modern / talked-about; grey = mixed.

HTTP Archive tells the same story at origin scale. In July 2026, jQuery showed up on about 5.77 million origins in an 8.88 million sample. React was on about 1.31 million. Next.js was rising fast year on year, and still looked small next to the legacy pile. Astro was on about 41,000 origins.

Legacy is the default, not the exception

The web that exists is old software stacked on old habits.

On W3Techs’ PHP breakdown in September 2026, PHP is still the server language for about 70% of sites where a language is known. Of those PHP sites, version 8 is 64%, but version 7 is still 28%, and version 5 is still about 8%.

jQuery itself is ancient with the first release in 2006. Of sites using jQuery, W3Techs still finds version 1 on about 17% of all sites using jQuery and version 2 on about 4%. Version 3 takes most of the rest. The library people pretend is gone is still very much present.

Servers look similar. Apache is still on about 22% of sites. Nginx and Cloudflare each sit near 31%. Modern edge hosting provides a real benefit, so it’s pretty telling that so many sites still use Apache which has been quietly serving HTML since before React even existed.

The difference shows up before the page even paints. Time to First Byte is how long the browser waits for the server to start responding. In field data, sites still hitting an origin with no CDN sit around 1.2 seconds at the mobile 75th percentile. Put the same kind of site on an edge network and the first byte usually arrives sooner: Vercel around 600ms, Cloudflare around 940ms, Netlify around 1.1 seconds. That is not the whole story of page speed, but it is the hosting layer the legacy web still underinvests in.

Time to first byte, mobile p75

Hover or tap a bar for the hosting context.

State of Web Vitals (CDNs) · Hosting provider speed study. CrUX phone field data, Q2 2026. Origin means no CDN detected. Edge hosts can still fail overall Core Web Vitals when the site ships a heavy JavaScript app.

Then there is weight. The 2025 Web Almanac puts the median mobile homepage at about 2,362 KB, up from 845 KB in 2015: roughly a 200% increase in a decade. Median JavaScript alone is hundreds of kilobytes. A lot of that bloat is not “the future of the web.” It is plugins, builders, trackers, and themes piled onto foundations that were never designed for them.

That said, bad code hygeine and maximalism of plugin use are as big of a problem in modern web development as they are in older stacks - which skews much of the data to be less favourable than it could be.

I have already argued that WordPress is a weak default for a new brochure site and that static HTML is usually the right call. The crawl does not disprove that. It shows why the argument still matters. Inertia is most of the market.

Like a UK high street

If you walk a British high street, you do not see only flagship stores and concept cafes. You see a mix: busy shops, broken shops, empty units, and shops still trading behind a facade that has not changed since 2009. Knight Frank / Green Street put national retail vacancy around 13.5% in late 2025, still well above the tightest prime pitches. The point is not that every high street is dead. The point is juxtaposition: empty or underused stock sitting next to and competing with the fancy moneyed shops.

The public web feels the same.

Some sites are modern and well kept. Many are open for business on software that should have been refurbished years ago. Some are effectively vacant: neglected domains, abandoned builders, pages that load but no longer represent the company. The talked-about web is Oxford Street. The existing web is every other high street in the country.

A slow website is the damaged shop unit. The stock might still be fine. The business might still have demand. But the entrance sticks, the lights flicker, and the front looks tired next to the refurbished place two doors down. Online, that friction shows up as wait time. People do not leave a review about your jQuery version. They just leave.

Google and SOASTA found that as mobile load time rises from 1 second to 3 seconds, the probability of a bounce jumps about 32%. From 1 to 5 seconds, it jumps about 90%. Separately, Google has reported that 53% of mobile visits are abandoned when a page takes longer than 3 seconds to load. That is the online version of customers walking past a shop whose door will not open cleanly.

Bounce risk vs a 1-second mobile load

Hover or tap a bar for the relative bounce jump.

Google / SOASTA mobile speed benchmarks · The need for mobile speed. Relative increase in bounce probability versus a 1-second load (Google/SOASTA). +123% means about 2.2× the 1s bounce probability, not a 123% bounce rate. Separate finding: 53% of mobile visits abandoned after 3+ seconds.

Speed is not only about bounce. A Google-commissioned Deloitte study, Milliseconds Make Millions, watched tens of millions of sessions across dozens of brands and found that even a 0.1 second mobile speed improvement correlated with 8.4% higher retail conversions and 9.2% higher average order value, plus about 10% more travel conversions. Tiny refurbishment. Measurable till.

That is not an insult to WordPress or jQuery. Those tools have had great success because they are accessible, cheap, simple, and good enough. High streets filled with independent shops for the same reasons. The opportunity is the upgrade path: take a real business that already has demand, and give it a storefront that is faster, safer, easier to change, and owned properly.

AI makes that upgrade cheaper. It does not invent the customers. The customers were always on the existing web.

So what

The web that people talk about will keep moving. The web that exists will keep running legacy stacks until someone pays to replace them. That gap is not a tragedy, it’s work.

Speed is one of the clearest reasons to do that work. Field data does not give a single “average page load” for every stack, but Largest Contentful Paint is a useful stand-in: how long until the main content appears for real users. Astro sites typically land around 1.2–1.8 seconds. Next.js sits near 2.2 seconds at the 75th percentile. Sites detected with jQuery are close behind at about 2.3 seconds. Untuned WordPress is often still in the 3.5–4.5 second band until hosting, caching, and images catch up.

Typical mobile LCP (seconds)

Hover or tap a bar for what the number means.

State of Web Vitals (frameworks) · Core Web Vitals WordPress guide · Astro vs Next.js CWV study. LCP at roughly the 75th percentile on mobile. Astro and WordPress are typical ranges from field monitoring and CrUX joins, not a single HTTP Archive median for all four.

That is the high-street upgrade in numbers. A shop that makes a bad impression is still a shop. It is also losing the people who would have walked in if it wasn’t as unappealing. Most of the live web is not starting from a green static build. It is starting from a stack that is still good enough to trade, and still slow enough that a rebuild pays for itself in clarity, security, and how the site feels on a phone.

Put stack and hosting together and the gap gets clearer. There is no single public study that joins “Astro on Netlify or Vercel” against “WordPress plus jQuery on shared hosting” in one query, but the field proxies line up the same way. Astro sites, which usually ship as static HTML on edge hosts, typically land around 1.5 seconds mobile LCP. Next.js on the same class of edge platforms sits nearer 2.2 seconds. WordPress on a mass-market shared host like Bluehost sits around 2.6 seconds median LCP. Leave WordPress untuned on traditional hosting and the monitoring numbers drift toward about 4 seconds. Nearly all of those WordPress sites are also running jQuery.

Stack plus hosting: typical mobile LCP

Hover or tap a bar for how the estimate was built.

Astro vs Next.js CWV study · State of Web Vitals (frameworks) · Hosting provider speed study · Core Web Vitals WordPress guide. Composite from separate CrUX joins: framework LCP for Astro and Next.js, Bluehost median LCP as a shared-host proxy, and monitored untuned WordPress ranges. Not one crossed A/B of identical sites.

If you have a site that needs to be modernised, upgraded, or replaced but you are not sure where to start, get in touch.

Sources