AI Emptied WordPress Boards and Left Owners Editing

WordPress freelance postings on Freelancer.com fell 58% between the second quarter of 2024 and the third quarter of 2026. PHP gigs fell 59% over the same stretch, and HTML fell 55%.

A survey of 201 US small business owners, run in August 2026 by Fractl for Bluehost, found that 53% had used an AI builder instead of hiring a developer. Only 9% of those AI sites went live as the tool first drew them.

WordPress Jobs on Freelancer.com Fell 58%

The board that used to feed a living on five-page brochure sites is thinner, and the drop is concentrated in the skills that made that living. Historical recruitment figures on Freelancer.com put website design ads down 59% from the second quarter of 2024 to the third quarter of 2026, matching PHP. WordPress demand fell 58%. HTML followed at 55%.

Those cuts ran deeper than several other remote categories on the same platform, which is why the headline attached itself to web work rather than to freelancing as a whole.

FREELANCER.COM POSTINGS, Q2 2024 TO Q3 2026

Data entry ads rose 16% in that window, so the slump is not a blanket freeze on cheap remote tasks. Graphic design fell 43%, a hard hit that still trails the web-build categories. The mix on this one board shifted away from “make me a site” and toward work that still looks like typing into a form.

The figures count posted orders on a single marketplace. They do not count how many WordPress developers still have clients, what they earn, or whether the missing ads moved to another platform, into an agency retainer, or onto the owner’s laptop.

Only 9% of the AI Sites Shipped Untouched

Bluehost’s survey is the other half of the same week’s argument, and it is smaller than the headline makes it sound. Fractl questioned 201 US adults who own or co-own a business and help decide on its website. Most of the glowing findings describe the 107 people (53%) who said they had already used an AI tool to build that site. At 95% confidence, the adoption figures sit about 6.9 percentage points either way; the AI-builder subgroup sits about 9.5 points either way.

WHAT THE 107 AI BUILDERS REPORTED

• $850 and 32 hours: Self-reported average savings against hiring a developer or agency, with statistical outliers removed, for AI use in general rather than one product.
• Satisfaction: 72% were very or extremely satisfied with the result, and 97% were at least somewhat satisfied.
• First draft: 9% kept the output exactly as the tool produced it. 61% edited the copy, and 60% adjusted the design.
• After launch: 94% kept using AI to run or grow the business; 34% said traffic rose and 31% said sales or revenue rose, as their own read, not a controlled test.

The savings are real as the owners tell them, and they map onto the two reasons people had delayed a site in the first place: 34% said they had not had the time, and 28% said they had not had the technical skill. Among the 94 owners who had not used an AI builder, 71% said they would consider it.

The part that does not travel with the “developers are finished” line is the edit. The cheap draft still arrives looking like a site. Then someone who used to pay a freelancer rewrites the about page, nudges the layout, and decides whether the thing is fit for customers. That someone is usually the owner, and that labor does not show up as a job ad.

A WordPress CRM plugin drafted with Claude reached official review in mid-October 2026 still missing CSV sanitization, a reminder that the directory still expects a person to read the code. The first draft got cheap. The fix did not.

Login Friction, AI Scraping and a Smaller Bid Pool

Freelancer Limited, which runs that board, gave its own account of a weaker core marketplace on October 20, 2026. In its third-quarter 2026 business update, the Freelancer segment’s gross marketplace value fell 27.6% (21.7% in US dollars) to A$19.1 million from A$26.4 million a year earlier. Segment revenue fell 26.3% to A$6.8 million from A$9.2 million.

The company said three issues identified at the first half of 2026 “continued to weigh on activity: login friction from enhanced security measures, lower organic search traffic after AI scraping disrupted site stability, and reduced payment acceptance at some gateways.” That is not a skill-mix chart. It is a marketplace that was already harder to reach, log into and pay through, in the same quarter the WordPress and PHP ad counts were being compared with 2024.

Read together, the 58% WordPress posting drop and the 27.6% GMV drop are different rulers on related objects. One is how many ads ran in named web categories across nine quarters. The other is how much money moved through the whole Freelancer segment in one quarter. Neither isolates AI site builders as the sole cause, and the company itself pointed at security, search and payments first.

Wix, Squarespace and WordPress page builders already pulled a lot of five-page sites off the freelance menu over the past decade. AI builders compress that same product again: less money, less time, and a first draft that looks finished until you read it.

What Upwork Hired WordPress Designers for in September?

On Upwork, the picture in September 2026 was not a simple vanishing. The Upwork Research Institute’s monthly hiring note, published October 14, 2026, said WordPress Website Design hired contracts ranked first among skills that grew both year over year and month over month. WordPress Customization sat in that same accelerating group, next to video editing software.

That metric is hired contracts whose posts mention the skill, not the raw pile of “build me a WordPress site” ads. It can rise while the cheap end of the board dies, if the remaining buyers are paying for design and customization rather than a first install.

The rise of AI SaaS developers, AI agent developers, and AI video creators shows how a more specialized workforce is emerging around AI implementation.

Jennifer Brett, managing director, Upwork Research Institute

UpHunt, which scrapes public Upwork posts, still shows a large hole on the WordPress line. Comparing April 2026 with January 2025, WordPress was down 2,162 jobs a month. Shopify was up 385, Lovable up 370, and Cursor up 295. The five largest “new” tools in that set added 1,205 jobs a month combined, which did not replace what WordPress alone shed. Lovable had 429 jobs in April 2026 and was still 14 times smaller than WordPress. UpHunt put overall Upwork volume down about 20% in that window and WordPress down about 27%, so the CMS fell faster than the platform.

Some of that work moved to other tools. Some of it left Upwork. Some of it never became a freelance job, because a founder could get a plausible homepage in an afternoon and stop there. The September hiring rebound in WordPress design is what you would expect if the leftover demand is “make this not look like a template,” not “install the CMS.”

Hosting, Domains and the Approval Button

Owners who used AI were glad they did, by their own scores, and 85% said they would not have built the site the same way without it. They still kept a hand on the tiller. Half (50%) said they would be comfortable telling customers that AI built the site, which means the other half would not. Control after launch split the same way.

WHAT THEY WILL NOT GIVE AWAY

• The live edits: 54% of AI builders wanted to approve every change the tool makes to the site; 30% would let it run on its own.
• Hosting and domains: Setting up the domain or hosting was among the tasks owners were least likely to hand to AI, cited by 21%.
• The public story: Only 50% were willing to say out loud that a machine designed the storefront.
• The copy and layout: 61% changed the text and 60% changed the design before launch, so the “finished” AI site was a draft for most of them.

Those refusals are the freelance opening that the empty job board conceals. A buyer who will not let an algorithm touch DNS, who wants a human yes on every later edit, and who will not tell customers about the builder, is not done paying for judgment. They may be done paying $850 for a theme install.

The people still pitching Laravel, PHP and WordPress on the same day the 58% figure circulated are selling that remainder: migrations, sanitization, the last 20% that makes a generated site hold up. Custom code around a platform’s limits, the path some designers already took from Wix and WordPress into Shopify, is the same pattern with a new generator at the front.

A Survey From the Company Selling the Alternative

Bluehost paid for the survey, and Bluehost sells the alternative. Its AI Website Builder turns a prompt to a live website in minutes, with domain and hosting in the same checkout. WonderSuite, the older guided WordPress onboarding kit, was renamed Bluehost WordPress AI Builder, so the replacement product still sits on WordPress rather than retiring it. The CMS is not the thing that left the bid list. The person who used to click through that onboarding for a client is.

Fractl’s write-up is blunt about the limits. Money and time savings are self-reported averages. Findings are not a guarantee of results. Percentages may not total 100% because of rounding. The 34% traffic claim and the 31% sales claim are what owners felt after they started “running on AI,” not a test that isolates the builder from everything else a small firm did that month.

None of that erases the hole in the order book. A 201-person, US-only sample with a 6.9-point margin cannot speak for every small firm, and one freelance platform cannot speak for employment. Two datasets still point the same way: the template site is now a product an owner can attempt without posting a job, and most of those owners then do the junior pass themselves.

The next bid that used to land on a WordPress freelancer has a 71% chance, among the owners who have not tried a builder yet, of being considered as an AI job instead. The site that comes back will still need someone to change the words, fix the layout, hold the domain, and decide what customers are allowed to know. That work used to be an invoice. A lot of it is now an evening.