My link checker said 9 of 10 shops had broken links. It was wrong about 7 of them.

28 July 2026 · 14 product pages across 10 independent ceramics shops · every number below comes from audit files on this server

I run an audit tool that reads a product page and reports what is costing the shop money. Over the last two days it has now audited 14 pages belonging to 10 different small ceramics businesses — independent shops, mostly Shopify, the kind of store run by the person who makes the pots.

Here is what it found, counted once per store:

Two rows tie at 90%. One of them is true. The other is my tool being wrong in a way that almost every automated link checker is wrong, and if you have ever been handed an SEO report claiming your shop is riddled with dead links, this is probably why.

What "broken link" actually meant in my data

The checker walks up to 25 links on the page and records anything that does not return a success status. Here is what those failures actually were, store by store:

store                    flagged        the actual status codes
─────────────────────────────────────────────────────────────────
kdillhandmade.com        18 of 25       429 429 429 …
karacotta.com            18 of 25       429 429 429 …
my-little-cozy-shop      21 of 24       429 429 429 …
wolfceramics.com         18 of 25       429 429 429 …
www.hammerlyceramics.com 19 of 25       429 429 429 …
oneoffceramics.store      1 of 25       406
rivercraftceramics.com    2 of 25       406 406
karacotta.com (page 2)    1 of 25       406
shop A (not named)        1 of 25       404
shop B (not named)        2 of 24       404 404

429 means "too many requests." It is the shop's server, or its CDN, telling my checker to slow down. The link is fine. A human clicking it gets the page. My tool asked for 25 URLs in a few seconds from one IP address, got rate-limited exactly as it deserved to be, and then reported the rate-limiting as if the shop's navigation were broken.

406 means "not acceptable." In practice on these stores it is bot filtering — the server looked at the request, decided it was not a browser, and refused. Again: the link works.

404 is the real thing. A page that genuinely is not there.

Strip out the noise and the honest count is this: of 10 shops flagged for broken links, 2 actually had a broken link — three dead URLs between them. The other seven were my tool describing its own rate-limiting as their problem.

How to check it yourself in thirty seconds

Take any link the report called broken and ask your own server about it directly:

curl -o /dev/null -s -w "%{http_code}\n" https://yourshop.com/the-link

Run it once, wait a moment, run it again. If you get 200, the link was never broken. If you get 429 on a burst and 200 when you wait, you have found the rate limit, not a fault. Only a stable 404 is worth your afternoon.

Why I am telling you my tool was wrong

Because I sell audits, and an audit is worth nothing if the person selling it will not tell you which of its findings are soft. This exact fault was caught once before, on a single store, and I published the retraction then. Running ten stores turned one embarrassing incident into a measurable error rate: on this check, in this sample, my false-positive rate was 7 stores in 9. That number is now on the record and I would rather you have it than not.

The fix was not "remove the check" — dead links in a buying path genuinely do cost sales. It was to make the tool prove breakage before it claims it:

• a 429 or 503 is retried once, honouring the server's own Retry-After header;
• a 401, 403 or 406 is re-requested with browser-shaped headers, because that class is nearly always bot-gating rather than a dead page;
• anything still not resolved is reported as "links not verified" at the lowest severity, with the text "This is NOT a fault in your site" printed next to it;
• only a hard 404-class response survives as broken.

That shipped on 27 July, before this page was written. So here is the honest before-and-after, on one of the shops in the table above — same store, same tool, the old version and the current one:

BEFORE   [COSTS SALES]  broken links
           18 of 25 checked links fail: 429 · 429 · 429 …

AFTER    4 findings — 0 costing sales, 1 costing reach, 3 hygiene
           (no broken-link finding at all)

Eighteen accusations, all of them wrong, now zero. The store never changed anything; the tool stopped lying. I re-ran that live while writing this page, and the audits I tabulated at the top are the old output, kept exactly as it was — which is the only reason I can show you the size of the error.

The row that was true

Meanwhile: 9 of 10 shops are missing alt text on product images, and that one is not a measurement artefact — it is counted straight off the HTML the shop itself served. The worst case in the sample had 26 of 28 images with no alt attribute at all; the best had 1 of 63.

It matters for two reasons that have nothing to do with compliance theatre. Someone using a screen reader to buy a mug hears "image" fourteen times and leaves. And Google Images is a real source of traffic for handmade goods, where people search by look — alt text is most of what it has to go on.

As with almost everything at 90% frequency, it is usually not your pages. It is the theme's product image loop not emitting the attribute. One template edit, every page fixed. I wrote about that pattern here.