600 Issues. Still Written by a Human.

Well, right off the bat, I’ve gotta give a nod to the fact that… this is issue #600 of The WP Edge. I’ve written 600 issues. Good lord. 🤯

Of course, it wasn’t always known as The WP Edge. It actually started off as a daily email (creatively called… The Daily) and it evolved from there. But, that was way back in 2017. I literally have memories of writing these emails sitting in campsites while traveling the country in 2018.

So, the fact that I’m still here doing it? Brain bender. 😜

I truly hope you enjoy reading these issues each week as much as I enjoy producing them. Every work week begins, first thing on Mondays, with me producing this newsletter. Coffee in hand, before I jump into any client work or any of the projects of the week.

Anyway, let’s dive right in, shall we?

Half the Internet Is Fighting Over 5%

Remember a few issues back, when I told you about Neil Patel’s stat on blogs losing 85% of their Google traffic? Well, Neil’s back with a sequel. And this one’s actually useful.

His agency ran a test. 744 articles across 68 real websites. Half went live exactly the way the AI wrote them. The other half got a human pass before publishing.

The human-touched half pulled in 444% more traffic. Yes, you read that right. The articles where a human did an editing pass on the AI draft got more than 5X the traffic. 🤯

Google says it doesn’t penalize AI content, and that’s probably true. The thing is, AI content doesn’t earn TRUST long-term. It’s good… for a quick hit. But the traffic tapers off a lot quicker.

And the bigger picture he shares is even more telling. AI now writes about 52% of all new content on the web… but that content earns less than 5% of the organic traffic. Half the internet is fighting over a twentieth of the reward.

That’s their data, not an independent study, so take the exact numbers with a grain of salt. But it lines up with everything I’ve seen. Raw AI content is cheap to make, and the market is pricing it that way. It is absolutely content inflation… over-supply means the “price” drops. People don’t value it nearly the same.

Neil’s fix is what he calls the “20-minute human layer.” Five things you do to an AI draft before you hit publish:

1. Inject firsthand experience
2. Add something the AI couldn’t know
3. Cut the filler
4. Fact-check and freshen it up
5. Put a real author on it

Good list. But the way he frames it, the human shows up at the end. AI writes the thing, then you spend 20 minutes cleaning up after it. And if the AI wrote it from scratch, 20 minutes of editing is a band-aid. You’re trying to inject yourself into something that was never yours to begin with.

Now, Neil didn’t really address the beginning part of this creation process. But I will. This is how I do it. Think of it like a sandwich. I’m the bread. AI is the filling.

I recorded myself producing an issue of this newsletter a few weeks back, start to finish, so you can see the whole thing. You can literally watch how I use Claude to create my newsletter and get it out the door. It breaks down like this…

The front: I talk it out. I never tell AI to “write me an article.” Ever. That’d be a big mistake. Instead I open up Wispr Flow and just talk. The big idea, the story, the client situation, my opinion… all of it, messy as it is. It is basically a “brain dump” of my thoughts that I want in that article. That covers Neil’s first two steps before a single sentence gets written. The experience is mine. And the stuff AI couldn’t know? AI gets it directly from me.

The middle: AI does the heavy lifting. It takes my rambling and turns it into a readable article. Structure, flow, order, tightening. This is the part that used to eat more of my time. Two things make the middle better than it would be otherwise:

• A voice profile that tells AI how I write, the words I use, and the ones I’d never be caught dead using. Neil’s example of AI filler is “in today’s fast-paced digital landscape.” That exact phrase is on my banned list. So the filler mostly never shows up.
• Context nobody else has. Every back issue of this newsletter lives in my Obsidian vault as plain text, along with daily notes of what I actually worked on. AI can pull from all of it, find the callbacks, and check my links.

The back: I edit by hand. Keyboard, coffee, the whole thing. I go through what AI put together and change whatever doesn’t sound like me or doesn’t read the way I want. Sometimes it’s a few words. Sometimes it’s a whole paragraph. But, it is me pounding on the keyboard just like I did before AI ever existed.

And my name goes on it. Along with a public AI disclosure policy, and my face in the Founder’s Memo videos a few times a week. An actual human who stands behind what’s published.

Neil made one more point I really liked. At his agency, if a piece used to take a human two hours, they still spend two hours on it with AI. The goal isn’t to cut the time in half. It’s to make something better than you could’ve made alone.

I mostly agree. This newsletter used to take me three hours plus by hand. Now it’s about two. But the time didn’t just shrink… it moved. Less of it goes to the mechanics in the middle, and more of it goes to the two ends that actually matter. The thinking, and the final read.

So if you’re using AI to write, try this. Don’t hand it a topic. Hand it you… your stories, your opinions, your messy first take. Let it build the middle. Then take it back and make it yours.

Human on both ends. AI in the middle. That’s how you use AI to speed things up without it stripping out everything that makes it you.

WordPress News & Updates

WordPress 7.1.2 is a security release you don’t want to sit on. The bug let an attacker who isn’t even logged in trick WordPress into loading PHP files it shouldn’t, which under the right conditions means running their own code on your server. It reaches all the way back to 4.7, so pretty much every site out there. Auto-updates should have handled it, but go look anyway… I verified it across every client site last week and found a few that hadn’t budged. Release notes.

Running Tutor LMS? Make sure you’re on 4.0.8. Wordfence found a hole in versions up to 4.0.7 that lets any logged-in subscriber run code on your server, on a plugin with 100,000+ active installs. And since most course sites let anybody register as a student, “logged-in subscriber” is a pretty low bar. It’s been patched since September 10. The details.

Wordfence 9 gives you passkeys for free. Log in with your fingerprint, your face, or your password manager instead of typing a password. Other security plugins have charged for this, and Wordfence’s reasoning for not charging is refreshingly simple… it doesn’t cost them anything to provide. You can register passkeys on multiple devices so you don’t lock yourself out. How to turn it on.

Find out if your site’s emails are landing in spam. Email Spam Tester is free, with no signup. Send a real email from your site (a password reset, an order confirmation, a form notification) to their test address, and it tells you whether it hit the inbox at Gmail, Outlook, Yahoo and others, plus checks your SPF, DKIM, DMARC and blocklists. I’ve seen plenty of sites where the contact form “works” and the owner has no idea the notifications are going straight to junk. (If yours are, here’s my take on which SMTP services actually work.) Test yours.

Automattic launched something called Spacefast. They’re calling it “the publishing layer for your agents.” You ask Claude or ChatGPT to make a landing page, a slide deck or even a small app, and it publishes it to a live link in seconds, with a database, logins, file storage and email built in underneath. It’s in research preview with a waitlist, free for personal use and $5 a month beyond that. Interesting to see where Automattic’s head is at… this is a whole lot closer to where I’m heading with the article below than to a traditional WordPress site. Take a look.

A caching plugin you tune by talking to it. WPDeveloper launched xSpeed Cache, and every copy (including the free one) comes with an MCP server built in. Connect Claude or ChatGPT, say “make these sites fast,” and the AI benchmarks the site, flips the right settings, warms the cache and retests. Funny timing, given what I’m about to say below about caching plugins becoming irrelevant. But WPDeveloper says more than half the sites run through their scanner had no caching at all… so if AI is what finally gets people to turn it on, fine by me. xSpeed Cache.

Not one WooCommerce store came back clean. Merchlint scanned 72 stores pulled from agency portfolios… the stores the agencies themselves were showing off. Every single one had problems anyone could see. 83% had images with no alt text, half still listed sold-out products, 47% reused the same description across products, and 38% had items with no price or that couldn’t be bought. If you run a store, go click around your own catalog like a customer would. The full report.

Novamira now has a free desktop app. A few weeks back I recommended Novamira for letting AI work on your WordPress site. Novamira HQ (Mac, Windows and Linux, free and open source) puts all your sites in one place, so you connect your AI app once and it can reach every site. It deliberately can’t delete sites, backups or domains, which I appreciate. Novamira HQ.

People are “vibe coding” and not checking the results. 20i surveyed 600 non-developers who use AI to write code. Only 31.5% have a developer review it, 9.3% don’t check it at all… and 20.5% say AI-written code has already contributed to a security breach. AI writing code for your website is genuinely useful. Pushing it live without anybody looking at it is how you end up on the wrong end of a news item like the ones above. Survey results.

FluentBooking 2.5 adds private notes to bookings. Pro users can jot down notes on any booking that only your team sees, and send them over to the contact’s record in FluentCRM. You can also map Fluent Forms fields straight into your booking questions now, so answers fill themselves in, and reschedule from the admin side instead of bouncing through the public booking page. What’s new.

FluentCart can finally match your site’s colors. 1.6.5 adds a store colors setting with three choices: keep FluentCart’s default look, inherit your theme’s colors (block themes, Astra, Kadence, GeneratePress), or pick your own. It applies everywhere, from product cards to checkout to the customer dashboard, and picks button text that’s actually readable against whatever color you chose. Customers also have to verify their email now before seeing their account history. Release notes.

WordPress Isn’t Going Anywhere. It’s Just Moving Backstage.

Last week I told you I was considering moving the front of Blog Marketing Academy over to Astro. I even said, and I quote, “this is a ‘considering,’ not a ‘doing.'”

Yeah… about that. 😅

Four days later, I was doing it.

Not the whole site. Not even close. But the WP Edge archives (every single one of the 600 issues, including this one) now have an Astro version built and running on my staging site. And the way we set it up turned out to be more interesting than I expected, so let me walk you through it.

Here’s the thing I was most worried about going in. I didn’t want an either/or. I didn’t want to “leave WordPress” and then have to rebuild logins, checkout, courses, the client dashboard, FluentCRM… all the dynamic stuff WordPress is genuinely great at. That’s the rented-land trap I talked about last week.

So we didn’t pick a side. They share the house.

WordPress and Astro both live on the same server, on the same website. When somebody visits a page, a small rule on the server asks one question: “Does an Astro version of this page exist?”

If yes, it serves the Astro page. Just a plain, pre-built file. No PHP, no database, no plugins firing up.

If no, it hands the visitor to WordPress, exactly like always.

That’s it. That’s the whole trick. (For the nerds: it’s a few lines in the .htaccess file. For everybody else, it doesn’t matter.)

So your login page? WordPress. Checkout? WordPress. The Concierge client dashboard? WordPress. Any page I haven’t converted yet? WordPress. Nobody has to know or care which one they’re looking at.

And here’s my favorite part… I still write in WordPress. Same block editor I’ve always used. When I hit publish, Astro grabs the content from WordPress and rebuilds that page automatically. It’s live in a minute or two. My workflow didn’t change at all.

The email sign-up forms still go straight into FluentCRM. My SEO settings still come from SEOPress. Independent Analytics still counts every visit. WordPress is still the brain. It just isn’t the one answering the door anymore.

So why bother?

Speed, for one. And it’s not a little faster… it’s a different kind of fast. Like night and day. I tested one of my back issues. The WordPress version of the page makes 52 requests and weighs about 450 KB over the wire… over a megabyte once your browser unpacks it. That includes 10 JavaScript files and 26 separate stylesheets. The exact same issue on the Astro side? 12 requests, about 160 KB, and zero JavaScript files. Roughly a third of the weight and a quarter of the requests. And no database queries whatsoever.

Think about what that means. For years, the WordPress answer to a slow site has been more plugins. A caching plugin. Perfmatters to switch off scripts you don’t need. Something to minify, something to lazy-load, something to delay the JavaScript. I’ve run all of them. I set them up for clients every week.

On the Astro side, none of that matters. There’s nothing to cache because the page is already a finished file. There’s no plugin bloat to switch off because it never gets loaded in the first place. You’re not optimizing the bloat away… you just never had it.

(To be clear, that’s the public-facing side. The logged-in, dynamic parts of the site are still WordPress, and they still get the usual performance treatment.)

But the bigger reason is how I get to work on it. The design of the Astro side is just code in a folder. So when I want to change something, I describe it to AI and it makes the change. No clicking through Kadence settings. No hunting for which of three places a style is coming from. It’s the same “direct it instead of click it” approach I used on DavidRisley.com last week… except now WordPress is still sitting right there behind it, doing everything a content site can’t.

Which leads to a funny side effect…

I probably won’t need the Kadence theme anymore.

And that’s not a knock on Kadence. I’ve used it for years and I still like it a lot. But think about what a theme actually does. It controls how your site looks. The header, the footer, the layout, the colors, the fonts. Kadence gives you a customizer, elements, and a pile of settings to make all that happen without writing code. If the public design lives in Astro, and I can change it just by telling AI what I want… the theme doesn’t have a job anymore. I’d rather design the site my own way, directly, than work through somebody else’s settings panels. The dynamic parts of the site that stay full-WordPress would likely get a custom theme specific to that goal (and likely built with AI, too).

I’ll likely still use Kadence Blocks when building pages, because the system I’m building can display them on the Astro side. But the Kadence theme, with the customizer and elements and all that? Long term, I think it goes away.

Some folks call this “headless WordPress.” That’s where WordPress only manages content and something else displays it. What I’m doing is more like half-headless. The public front gets converted one section at a time. Everything dynamic stays put. And if an Astro page ever breaks, there’s a WordPress version right underneath it.

The WP Edge archives were the guinea pig, on purpose. They’re a clean, self-contained section. If it works there, the blog is next. Then maybe the main public pages. Slowly. One piece at a time, measuring as I go.

I’ll keep you posted as it goes live.