We run a 897-page static site, ten languages, deployed on Cloudflare Workers. No CMS, no database, no framework — just HTML files generated from templates and pushed through a build step.
For the last few weeks I've been auditing B2B directories to see which ones actually give a company page a clickable link back to our site. I started with a list of 37. Zero gave us a usable link.
This post is about the six ways that audit went wrong before it went right, because five of those six mistakes are the same mistake wearing different clothes: trusting something that looks like evidence but isn't.
The naive approach, and why it fails
The obvious method: search Google for "yoursite.com", see which pages come up, assume those are backlinks.
That's what I did first. I found six results. I wrote "6 backlinks" in my notes and moved on.
Two weeks later I opened one of those pages in a real browser and ran this:
document.querySelectorAll("a[href*='yoursite.com']").length 0 Zero. The company profile had a field labelled Website showing https://yoursite.com — as a plain text node. No <a>, no href, nothing. It was a <td> with a string in it.
The page ranked for our domain because the string yoursite.com appears in the body text. That's it. Google found the characters, not a link.
Mistake #1: "It ranks for my domain" is not "it links to my domain." These look identical in a search result. They are completely different in the DOM.
Check 1: is there an actual <a href>?
This has to be a DOM query, not a text search.
// Bad: matches plain text, JSON-LD, meta tags, everything document.body.innerHTML.includes("yoursite.com") // Good: only real anchors document.querySelectorAll("a[href*='yoursite.com']").length One caveat: includes() on innerHTML will also match a comment like <!-- see yoursite.com -->. If you're scripting this at scale, parse the DOM, don't regex the HTML.
Check 2: does the link survive a static fetch?
This is the one that cost me the most time.
I wrote a Python script using urllib to check a supplier-mapping platform. It fetched 13,196 bytes and reported zero links to the sample company's website. I marked the platform dead.
Then I opened the same URL in a browser and ran the same query. One link. dofollow.
The page is a React SPA. The server sends a shell; the profile data — including that Website row — is fetched and rendered client-side. urllib sees the shell. The browser sees the page.
Mistake #2: static HTML fetching cannot audit client-rendered pages. If a directory is a modern SPA, curl, urllib, requests, and most "page speed" style fetchers will all lie to you in the same direction: they'll under-report links.
There's a nasty asymmetry here. A static fetch will never report a link that isn't there, but it will frequently miss one that is. So a negative result from a static fetch is not evidence. A positive result is. Plan your verification accordingly.
Check 3: rel attributes, and where nofollow hides
Once you've confirmed the anchor exists, you need the attributes:
const links = [...document.querySelectorAll("a[href*='yoursite.com']")]; links.map(a => ({ href: a.href, rel: a.rel, // "" when absent target: a.target, })); A real dofollow link from a claimed profile on a supply-chain mapping platform came back as:
<a href="http://example-supplier.com/" target="_blank" rel="noopener noreferrer"> rel is present — but it's noopener noreferrer, not nofollow. Those are security and privacy directives, not link-equity directives. A lot of audit scripts flag any non-empty rel as suspicious. Don't.
The check you actually want:
document.querySelectorAll("a[href*='yoursite.com'][rel*='nofollow']").length === 0 Mistake #3: treating "has a rel attribute" as "is nofollow." noopener, noreferrer, external, ugc, and sponsored are not the same thing. Only nofollow, ugc, and sponsored affect link equity, and only the last two do so unconditionally.
Check 4: can you see the page at all without an account?
Several directories show a public listing index and a gated profile. The index page looks fine — company names, countries, product categories. Click any of them and you land on a login screen.
That's not a link you can get by registering. That's a link nobody can see except logged-in buyers, which means it does almost nothing for crawl budget.
Same category, different flavour: directories where the free tier simply doesn't have a website field. You can register, you can fill in your company details, and the field where a URL would go doesn't exist in the free template. No error message. The field just isn't rendered.
Mistake #4: "can I register?" is a different question from "will it link to me?" Registration is the easy part. It's the part every directory wants you to do.
Check 5: yesterday's answer is not today's answer
Three platforms on my original list were marked viable in a survey I'd done two weeks earlier. All three had changed:
- One now only accepts companies with a registered presence in a specific region — a rule that wasn't enforced when the original list was compiled.
- One's free tier lost its website field entirely.
- One turned out never to have had a company directory at all. The site only offers certification services; the "member directory" I'd noted was a page describing how to apply for a certificate.
None of these are dramatic changes. That's what makes them dangerous. A directory list is a snapshot, and snapshots rot silently.
If you're maintaining a list like this, every row needs a date and a piece of evidence attached. Not "DA 55, accepts Chinese factories" — that's a claim with no expiry and no proof. Instead: "2026-11-16, a[href*='example.com'] = 1, [rel*='nofollow'] = 0, sample profile: <url>."
Mistake #5: storing conclusions without dates or evidence. Six months later you won't remember whether "works" meant "I confirmed it" or "I assumed it."
The audit script
Here's the shape of what I ended up using. It runs inside the page context, so it has to be pasted into devtools or driven by a browser automation tool — not curl.
function auditLink(domain) { const anchors = [...document.querySelectorAll(`a[href*='${domain}']`)]; return { count: anchors.length, dofollow: anchors.filter(a => !a.rel.includes('nofollow')).length, samples: anchors.slice(0, 3).map(a => a.outerHTML.slice(0, 200)), // if count === 0, also check whether the string appears as text: textOnly: document.body.innerText.includes(domain), }; } The textOnly flag is the useful one. If count === 0 but textOnly === true, you're looking at exactly the failure mode from Mistake #1 — the domain is on the page, but nothing links to it.
Run it against a third-party sample profile, not your own. If you haven't registered yet, you need to know what the platform does for companies that have. Pick a company that's clearly active there, audit their profile, and that's your answer.
What 37 directories actually produced
| Result | Count |
|---|---|
| No link field at all | 9 |
| Login wall | 4 |
| Won't accept companies in our region | 8 |
| Paid only | 5 |
| Wrong audience / wrong ecosystem | 5 |
| Actively harmful (link farms, PBN-adjacent) | 5 |
| Blocked by bot protection, unverifiable | 1 |
| Confirmed dofollow | 0 |
Zero out of thirty-seven.
And the last one to fall is the interesting one, because it looked like the winner for two weeks.
There's an open-data platform for factory supply chains. Their company profiles do carry a Company Website link, and on the profile I audited, that link was dofollow. So I submitted our facility, waited for the claim to clear, and got access to the edit form — ready to add our URL.
The edit form has ten fields. Here they are, verbatim:
Location Name Address Country Sector(s) Product Type(s) Location Type(s) Processing Type(s) Number of Workers Parent Company Industry Classification (ISIC) No website field. Not hidden under an advanced tab, not gated behind a paid tier — it does not exist in the current form.
The dofollow link I had audited belonged to a profile created under an older version of the platform, or one that carried the URL inside its description body. Either way, the field a new applicant sees is gone.
Mistake #6: auditing someone else's profile tells you what the platform did, not what it will do for you.
That one cost two weeks and a claim process. The fix is cheap, though: find a profile that was created recently, not one that ranks well. Recent profiles show you the current form.
The short version
If you take one thing from this:
Search results tell you what text is on the internet. They do not tell you what links to you.
To know that, you need three things, in this order:
- A real
<a href>in the rendered DOM - No
nofollowin itsrel - Visible without logging in
Anything short of that is a guess wearing a table's clothing.
The site this audit was for is a woven fabric mill in Shaoxing, China — Shaoxing Mujia Textile, 897 pages across 10 languages. If you're doing the same audit for a manufacturing or B2B business, the failure modes are probably identical; the directory names just differ.
