574 sites didn’t make the study base
If you own or market a manufacturing shop, I’d probably skip to the end. The middle is me showing my work, including the bit I got wrong.
4,402 manufacturer websites passed eligibility and went through the read. I read every page I could capture and ended up with a study base of 3,828. So what happened to the other 574?
I figured it was housekeeping. Turns out one of the four ways a site leaves the base is a judgment call, and one of my rules was just wrong.
I can’t score a cookie banner
11 errored outright. Another 431 captured something that was not the company’s own content. Here’s what came back:
- 278 captured navigation only.
- 115 captured a cookie banner.
- 19 came back empty.
- 14 returned a 403 (four of them with “Invalid SSL certificate Error code 526”).
- 5 sat behind a JavaScript wall.
All I got from Monroe Engineering was its own navigation. I left all of them out. I kept asking whether a page had nothing on it, or whether I’d failed to find what was there. Count them as companies that said nothing and I’m making a claim about their writing, when the failure was in my capture. The headline number would look tidy and mean the wrong thing.
That leaves 3,960 sites with real content. Now I have to make some calls.
65 are a subdomain of somebody else’s site
catalog.reesinc.com, products.kuriyama.com, us.dmgmori.com, hydroforming.schulergroup.com, southwest.swagelok.com. Six start with catalog, five with products, and four are regional Swagelok distributor pages.
Real companies, some of them big. I left them out because I’m looking at what a shop puts on its homepage. A product catalog on a subdomain is a different page. Asking a parts catalog whether it names a problem its customer has is a stupid test.
25 have a headline that reads “All Categories”
diecastingmanufacturers.com, aluminum-extrusions.net, contract-manufacturers.org, broachingjobshops.com, aluminumanodizing.com. Every one is a directory dressed as a manufacturer. The biggest words on every page give it away: All Categories or Related Categories.
I tried the domain name first. A domain with “companies” or “manufacturers” in it looked like an easy way to spot a directory.
It caught four real firms: jecompanies.com, eaglegroupmanufacturers.com, hopewellcompanies.com, isccompanies.com.
So much for that.
So I threw the rule away and looked at what the page does. A directory says what it is in its own headline.
14 sell what somebody else made
sterlingseal.com (“Large Selection of O Rings, Gaskets, Filters”), circlevalve.com (“Stocking Distributor of Industrial Valves”), bearingsandindustrialsupply.com. Distributors or manufacturers’ reps, with nothing anywhere on the site to say they make anything: no equipment or tolerance, no process they perform, no finishing, no assembly.
A distributor that also machines parts stays in, as long as anything on the site says it made something.
28 I took out by hand, one at a time
The other three rules are mechanical. These I vetted by reading the site and deciding it didn’t belong. I wrote down a reason for each one I took out. Without a reason, this is just a list of sites I didn’t like.
Here’s what that looks like, a few to show the range:
- thebluebook.com is a commercial construction directory. jobshopsohio.com says “700+ Ohio Manufacturers Listed In This Directory”. top.parts says “Find America’s Top Custom Manufacturers Near You”.
- digitalleadership.com is a strategy consultancy. It turned up because its homepage opens “Strategy is the last handmade process in your company”. I like that line more than almost any opening a real shop in the study wrote.
- mfgskillsct.com is a Connecticut state workforce program. metalforamerica.org is a defense metals advocacy group. fuzehub.com is a New York State nonprofit.
- brmcginty.com has a callout in its hero that reads “AC OUT IN THIS HEAT?”. That’s excellent writing for a homeowner with a dead air conditioner. bendheating.com guarantees your Carrier installation. Both are contractors, not parts makers.
- cnc-pros.com sells and services machine tools. It slipped past the distributor rule because it names processes it sells rather than performs.
Six of the 28 turned up in a single afternoon, all for the same reason: I was checking every site in the study that opens by naming a problem its customer has, and the non-manufacturers were disproportionately good at it.
A consultancy, a workforce program and an HVAC contractor all lead with a problem. Of 3,828 manufacturers, seven do.
The eleven I nearly threw away
I’d rather not have this section.
My subdomain rule counted dots. Two or more dots meant a subdomain, because catalog.reesinc.com has two and reesinc.com has one. That’s true in the United States and false for addresses like these, it turns out. lestercast.co.uk carries two dots by construction. So do llewellinsgears.co.uk, westinengineering.co.uk, wrightsplastics.co.uk, get-it-made.co.uk, ansini.co.uk, 3dprint-uk.co.uk, brunitec.com.br, and three more.
Eleven real manufacturers were out of the study, most of them British, because of how their web address was built.
I’d made the same mistake twice in one study. The directory rule caught four real firms because I thought their names looked wrong. This rule did the same thing with dots. If I’m going to judge a name by its shape, I need to nail down which shapes are legitimate. My rule only recognized the American ones.
The fix is a proper test of the domain ending, one that accounts for country-code second-level domains, with eight test cases. That includes a.b.co.uk, which really is a subdomain and would break a naive fix.
Two of the thirteen the fix rescued still came out: the British Plastics Federation is a trade association, and mig-welding.co.uk is a DIY guide.
Ask who came out
Every number in this study is a count, not a sample. So who I let into the count matters.
The base moved four times in one day, from 3,825 to 3,819 to 3,830 to 3,828. I kept reading the sites. They kept giving me reasons to change what the filter had done.
I keep asking shops to count things in their own writing. A page that says “we serve a wide range of industries” hasn’t counted anything. A page that says “we have run 17,000 jobs for 40 customers in three industries, and here are two of them” has. So I owe the reader that same care when I put a number on a finding.
When I see a big percentage, I dig in and confirm it before I trust it. When somebody hands me a number about a market, I want to know what got left out of the count and why. If they can’t tell me, I don’t know what they’ve counted.
Every quote from a site here I matched word for word against the page as I captured it in September 2026. The exclusion list with a reason per site is in scripts/population_check.py; the subdomain test and its cases are in scripts/headline_figure.py.