WordPress site keeps breaking? What one rebuild fixed
By Craig Burdett, , 8 minute read
Mary Barneby is a wedding officiant in Delaware. Her business, Vows to Cherish, runs on word of mouth and on her website. That website broke three times, and every time it broke, her web developer sent her a bill to fix it.
If your business runs on WordPress, you may already know how that feels. The site looks fine on Monday. By Tuesday a page has fallen apart, and nobody changed anything. Nobody you know about, anyway. You pay for the repair, you relax, and somewhere in the back of your mind you wonder when it’ll happen again.
Mary asked me to find out why it kept happening. Here’s what I found, what I rebuilt, and what the tests showed.
Six products holding up one small site
Mary’s site has ten pages and nine blog posts. Underneath them sat six products from six different companies.
- Woodmart, a theme designed for online shops, although Mary sells nothing online.
- WPBakery Page Builder, which lays out each page.
- Slider Revolution, which drew the photo slider on the home page.
- Contact Form 7, Yoast SEO and WP-Optimize, a caching plugin.
Each one updates on its own schedule, and each one can change how the others behave. Every page loaded a stylesheet of more than half a megabyte and a 640 KB bundle of scripts, for pages that are mostly words and photographs. The bundles still carried code for a Google+ button. Google+ closed in 2019.
The theme’s demo content had never quite gone away either. Nine leftover tag pages named things like “chair”, “sofa” and “furniture” were still reachable on a wedding officiant’s website.
Why it kept breaking
The most recent break was the page jumping around as it loaded, worst on the Gallery and Connect pages. Look underneath and it isn’t hard to see why.
- 126 of the 303 images across the main pages had no usable width and height, so the browser couldn’t save space for them. As each one arrived, everything below it was pushed down.
- The home page photo wasn’t an image the browser could see in the page. Slider Revolution painted it with JavaScript after the page loaded.
- The logo was set to load lazily, which tells the browser it can wait. Lighthouse found it was the largest thing on screen early on, so the page’s main content was being deliberately held back.
A layout like that depends on everything loading in the right order. When a plugin updates, or the caching plugin rebuilds its files, the order changes and the page falls apart. A repair fixes the symptom, and then the next update brings it back.
I can’t prove which update caused each break, and I don’t think anyone can now. But three breaks with the same kind of fix is the classic pattern. The cause was never in one setting. It was in the way the site was built.
There were smaller problems too. The site stopped visitors from zooming in, which matters for anyone who needs larger text. The reviews section showed 35 entries, but several were the same review twice, one of them under two different names. The footer copyright stopped at 2025, and the contact form had no spam protection at all.
Almost every WordPress site is built this way
It would be easy to blame one developer. But Mary’s site wasn’t unusual. A theme, a page builder, a slider and a stack of plugins is how most small-business WordPress sites are put together, because it’s quick to set up and it looks good in a demo.
The cost comes later. Every plugin is code someone else wrote, updated on someone else’s timetable. Security firms that track WordPress report that the great majority of its vulnerabilities are found in plugins and themes, not in WordPress itself. So the owner is left with two bad choices. Update everything and risk a break, or hold back and risk being hacked. Either way, someone gets paid to clean up.
Ask yourself how many plugins your own site runs, and whether you could name the company behind each one. Most owners can’t, and that isn’t their fault. It’s just that nobody told them they’d be responsible for all of it.
What the rebuild changed
I rebuilt the site in plain HTML, CSS and a small amount of JavaScript. It looks the way Mary’s visitors already know, and every page keeps its old address, so nothing she’s earned in Google is lost. From the first look at the live site to a working copy took two days.
The two pages look almost the same, and that’s deliberate. The difference is underneath, and in the photos that now arrive as real images instead of being painted in by a script.
- Every image has its width and height set, so nothing moves while the page loads.
- The home page photo is a real image that the browser fetches first, not one painted by a script afterwards.
- Visitors can zoom again.
- The reviews were cleaned up with Mary’s help. Repeats came out, the right name went back on the review that had two, and 24 genuine reviews remain.
- The contact form has a spam trap and needs no plugin to work.
- Search engines now get structured data that says this is a local business, where it works and how to reach it.
Just as important is what’s gone. There’s no database, no login page, no plugin and nothing to patch. There’s nothing for an update to break, because nothing updates itself.
Testing it against WCAG
WCAG testing was part of the service. WCAG, the Web Content Accessibility Guidelines, is the standard most accessibility laws and audits point to, and level AA is the usual target. I checked Mary’s site against it, and two things didn’t pass. Both had small fixes, which I recommended to Mary.
The first was the pink. Mary’s brand pink, #f06c9f, appeared everywhere, as white text on pink buttons and menus and as pink headings and links on white. Either way round, the contrast between the two was 2.86 to 1. AA asks for 4.5 to 1 for ordinary text and 3 to 1 for large headings, so the pink failed even for headings, and on the cream bands it fell to 2.49 to 1. People with low vision, or anyone reading a phone in sunlight, simply couldn’t read it well.
The fix was a deeper shade of the same pink, #cc145b. It reaches 5.50 to 1 on white and 4.80 to 1 on cream, so it passes everywhere it’s used, and the heart in the logo was recolored to match. Side by side, the new pink reads as the same brand, just a little richer. The pale grey on the blog’s dates and bylines got the same treatment, going from 2.46 to 1 up to 5.74 to 1.
The second was the button text. Five buttons on the old site said only “Read More”. Someone using a screen reader often jumps from link to link, and hears “Read More, Read More, Read More” with no idea where any of them lead. Now each button says what it opens, such as “Read More about Mary” and “Read More about vow renewal services”. That helps search engines understand the links as well.
These fixes, along with letting visitors zoom again, are a big part of why the site’s Lighthouse accessibility score went from 79 to 100 in the tests below.
Before and after
I tested Mary’s live WordPress home page and the rebuilt home page with Catchpoint’s WebPageTest, under the same conditions. The rebuilt page weighs well under half as much, fills the screen more than a second and a half sooner and recorded no blocking time at all.
| Measure | WordPress site | Rebuilt site |
|---|---|---|
| Speed Index | 2.63 s | 0.99 s |
| Largest Contentful Paint | 0.83 s | 0.92 s |
| Time to first byte | 0.28 s | 0.31 s |
| Total Blocking Time | 172 ms | 0 ms |
| Page weight | 1,000 KB | 401 KB |
| Requests | 30 | 22 |
| Render-blocking requests | 6 | 0 |
| Lighthouse Performance | 83 | 99 |
| Lighthouse Accessibility | 79 | 100 |
| Lighthouse Best Practices | 100 | 100 |
| Lighthouse SEO | 92 | 100 |
Three test runs per page, on November 3, 2026 for the WordPress site and November 4 for the rebuild, using desktop Chrome on a cable connection from Dulles, Virginia. The figures are from each page’s median run by Speed Index. The rebuilt site was tested on its staging copy before launch.
Two numbers went the old site’s way, each by less than a tenth of a second, and you should know why. The rebuilt copy was tested on my staging server before launch, which answered a few hundredths of a second more slowly than Mary’s host. The old site’s Largest Contentful Paint also flatters it, because the largest thing it showed early was the logo, while the real photo was still being assembled by a script. Speed Index, which measures how quickly the whole screen fills in, tells the truer story. I’ll retest everything once the rebuild is live at its own address.
What it costs to keep a site standing
A WordPress site like Mary’s is never finished. It needs updates, backups, a security plugin to watch the other plugins, and someone to call when an update goes wrong. That’s a bill that keeps arriving for as long as the site exists, and it arrives whether or not anything on the site has changed.
A rebuild is paid for once. After that, a plain HTML site costs little more than its hosting, because there’s nothing to patch and nothing to break. It loads faster, it’s easier for Google to read, and it leaves very little for an attacker to work with.
There is a trade-off, and it’s an honest one. Without WordPress, changes to the site usually go through a developer. For a business whose pages change a few times a year, that’s a small price for a site that stays up. If you edit your site every week, the answer may be different, and that’s one of the things the review looks at.
You don’t have to wait for the next break
Maybe your site has broken before. Maybe it hasn’t yet, and you’ve just noticed how many plugins it’s running. Either way, you can find out now what’s underneath it, while it’s working, instead of finding out from a customer.
When you send me your site’s address, I’ll look at what it’s built on, what it loads, what moves as it loads and where it’s exposed. Then I’ll tell you in plain English whether it’s worth patching or time to rebuild. The review is free, and either way you’ll know more about your site than you do today.
Comments
No comments yet. Questions and corrections are welcome.