Wpfavs - Your favourite WordPress news
All Episodes
AI Site Builders vs. WordPress: What the Benchmarks Miss

AI Site Builders vs. WordPress: What the Benchmarks Miss

0:00|0:00

AI can build a polished website in minutes, but new benchmarks and audits reveal the hidden work behind production-ready sites: accessibility, security, performance, and reliable commerce. We weigh AI’s risks against WordPress’s mature foundations—and explore how MCP, Markdown feeds, and builder agents are changing websites for both people and AI assistants.

This show was created with Jellypod, the AI Podcast Studio. Create your own podcast with Jellypod today.


Chapter 1

x

Nick d

Welcome to Wpfavs. Today's question is whether AI site building makes WordPress obsolete. October 2026 gave us some fresh data on it. WebDevStudios ran a benchmark they call It Looks Done. Two websites that look identical in a screenshot. One was vibe coded with AI in four minutes. The other was built on WordPress. Their finding is that AI nails the top 20 percent, the layout and styling you can see. It misses the other 80 percent, which they break into fourteen production requirements.

Ido

Fourteen things. Give us a few, because that's where this gets real.

Nick d

Keyboard accessibility, semantic markup, role based access control, caching layers, secure API routes, database structure, backward compatibility. None of it shows up in a screenshot. And their line stuck with me: the deeper you go, the more expensive these are to miss. WordPress has spent two decades building all of them. A fresh prompt starts at zero.

Ben cohen

I agree with the warning, but I want to challenge the framing. Is the contest really AI versus WordPress? If a prompt can produce a polished frontend in four minutes, then polish is cheap now. Nobody should be selling polish as the scarce thing anymore.

Nick d

Sure, but cheap polish is exactly the trap. Look at the 20i hosting survey of 600 people using AI coding tools. 51.1 percent had very limited or no prior coding knowledge. Only 12.5 percent regularly wrote code before AI tools. And fewer than a third, 31.5 percent, have a professional developer review what the AI generated.

Ido

And 9.3 percent deploy without verifying or testing it at all. That number made me stop.

Nick d

Right. Nearly one in ten shipping blind. When it comes to protecting a business, that's not a speed story, it's a liability story. 47.7 percent said they adopt vibe coding mostly for speed and productivity. I get that. But speed without review is just a faster way to create technical debt.

Ben cohen

Okay, but hold on. Review is the missing piece in that survey, not the AI. Those people could also install twelve random plugins and never test a single one. Is the problem the tool or the absence of a process?

Nick d

Fair, and here's where I'd push back on myself. WordPress isn't magic either. NorthDuty audited 100 live WooCommerce stores pulled from agency portfolios, using read only checks. 5 percent had reproducible public Store API or TLS configuration defects. Another 10 percent had broken browser interactions, like dead navigation menus, footers that didn't work, or broken cart and product links.

Ido

So 15 percent of professionally showcased stores had something broken. That's the part I'd hammer. If agencies on a mature platform miss that, imagine an unreviewed generated site handling checkout.

Nick d

That's my point exactly. A mature platform gives you a floor, but you still have to test. Remove both the floor and the testing, and the risk compounds.

Ben cohen

I'll concede that. A platform with twenty years of plumbing beats a blank prompt on anything that takes money or personal data. My disagreement is about where the next fight happens. Because the market data doesn't look like obsolescence at all.

Chapter 2

Machines Are Readers Now

Ben cohen

Barn2 pulled W3Techs numbers for the top 10 million sites. WooCommerce powers 7.9 percent of all websites, Shopify 5.4 percent. Among eCommerce platforms, WooCommerce holds 47.7 percent and Shopify 32.1 percent. And Nettingale audited 4,096,194 homepages across 32 European countries and found 47.6 percent, about 1.95 million sites, run WordPress. Estonia is highest at 59.4 percent, Czechia lowest at 34.5.

Ido

So nobody is leaving. Even the lowest country is over a third.

Ben cohen

Exactly. WordPress isn't going obsolete. The risk is different. The reader of your site is changing. AI assistants and agents are retrieving your content, and if your site doesn't expose clean structure, it becomes invisible to them. That's the shift from optimizing for page views to optimizing for agent retrieval.

Nick d

That's where the tooling is catching up, and I like that it's practical. The WordPress Core AI team released the canonical MCP Adapter plugin on the directory, so a site can expose a standard Model Context Protocol server. And AI Plugin 1.4.0 added Markdown Feeds, which serve clean Markdown to language models alongside the normal HTML.

Ben cohen

Which is huge, because an agent doesn't care about your hero animation. It wants structure. Headings, schema, clean text. Strip away the styling and the JavaScript, and ask what's left. For a lot of sites, the answer is not much.

Ido

And this is where I think the middle ground is real. Builders are shipping native AI agents, like the Divi 5 AI Agent and Beaver Builder AI. You get prompt driven speed, but the output lands inside a CMS where the other 80 percent already exists. Permissions, caching, database schema, accessibility. You're not asking the model to reinvent them.

Nick d

I like that, with one caution. A builder agent is a game changer only if it respects the plumbing. Streamline your process, yes, but don't add plugin bloat to do it. Every new tool is another thing to maintain.

Ben cohen

Can I push on that? Some plugins now are the plumbing for agents. CrawlBrain, for example, launched as an AI site search that replaces the traditional search bar. Shoppers ask natural language questions and get answers straight from store content. That's a plugin earning its place.

Nick d

If it's tested and it answers from your own content, I'm fine with it. My rule is simple. Does it solve a real friction, and can you verify it? If not, skip it.

Ido

So where do we land? I'd say vibe coding is a great way to sketch, and WordPress is where sketches become businesses.

Ben cohen

I'd add one thing. Being production ready now includes being machine readable. A site that passes the human screenshot test and fails the agent test is only half done.

Nick d

And my takeaway for anyone building this week: review everything, test your cart and your menus, and treat generated code like a draft. The 80 percent nobody sees is still the job. Thanks for listening to Wpfavs.