Blog · 22 November 2026 · 5 min read

We asked 527 small shops for a page that does not exist. 35 answered 200 OK

Every HTTP response carries a status code. People never see it; search engines, link checkers and uptime monitors read nothing else. On 22 November we requested one made-up URL from each of 527 small Polish online shops. A missing page should answer 404. Thirty-five shops answered 200 – “here is your page”.

The numbers

Each shop got one request for a random path such as /strona-ktorej-nie-ma-x7k2p9qa, and we followed redirects. Ten shops could not be assessed: nine served a bot-verification screen to our client and one redirected to nowhere. That leaves 517.

Response to a URL that does not existShopsShare
404 – correct48293.2%
Redirect to the homepage, 200193.7%
200 at the requested URL (shop page or an error message)101.9%
Redirect to a dedicated error page that itself answers 20040.8%
Redirect to some other URL, 20020.4%

That is 35 shops, about one in fifteen, with what is usually called a soft 404.

We then asked those 35 a second time, with a differently shaped path: inside a subdirectory and ending in .html. Twenty-nine answered 200 again. Six returned a proper 404 this time – their behaviour depends on what the wrong URL looks like. One probe is not enough; rewrite rules often treat /foo and /foo/bar.html differently.

The usual caveat: this is not a random sample. It is a hand-built list of small shops by trade, without big chains. Method at the end.

What a soft 404 breaks

Search engines have to guess

Google’s documentation on HTTP status codes is plain about 4xx: such URLs are not indexed, and URLs already in the index are removed. With a 200, Google has only the content to go on. If it looks like an error or an empty page, Search Console reports a soft 404. If it does not – and a homepage served for a retired product URL does not – the crawler has one more URL with content it already knows.

None of the 35 pages carried noindex. On one shop the page served at the made-up URL declared that made-up URL as its own canonical.

Your tooling goes blind

Link checkers, uptime monitors and the “404 errors” report in analytics all key on the status code. If the site answers 200 to everything, a typo in a newsletter link, a deleted category and an old URL from a printed flyer all look healthy. The same applies to anything that probes for a file. When we looked for exposed files on small shops, a check had to match the file’s content, not the status code – on a soft-404 site a status-only check “finds” every file you ask for.

Visitors lose the thread

Someone follows a link to a specific product and lands on the homepage with no explanation. A real error page says the page is gone and offers search and categories. A homepage does neither.

Test every site you maintain

One site, one command:

curl -s -o /dev/null -w "%{http_code}\n" https://example.com/no-such-page-12345

404 or 410 is correct, 200 is a soft 404, 301/302 means a redirect – add -L to see where it ends. For a list of sites, this script probes both path shapes and follows redirects:

#!/usr/bin/env bash
# usage: ./check404.sh sites.txt   (one origin per line, e.g. https://example.com)
while read -r site; do
  [ -z "$site" ] && continue
  for path in "/no-such-page-$RANDOM" "/no-such-dir-$RANDOM/page.html"; do
    out=$(curl -s -o /dev/null -L --max-redirs 6 -m 20 \
      -w '%{http_code} %{num_redirects} %{url_effective}' "$site$path")
    code=${out%% *}
    case "$code" in
      404|410) verdict=ok ;;
      200)     verdict=SOFT-404 ;;
      *)       verdict=check ;;
    esac
    echo "$verdict $site$path -> $out"
  done
done < "$1"

Output for a healthy site and for one that redirects missing pages to the homepage:

ok https://webskan.com/no-such-page-24448 -> 404 0 https://webskan.com/no-such-page-24448
SOFT-404 https://shop.example/no-such-page-9411 -> 200 1 https://shop.example/

Fixing it

  • Turn off “redirect 404s to the homepage”. It was the most common pattern in our data (19 of 35) and usually comes from an option in a redirect or SEO plugin, a theme setting or the shop platform. A 301 is right only when the old URL has a specific successor.
  • Serve the error page at the failing URL, without a redirect. If the site sends visitors to /404.html or /page-404/, that page exists, so the server truthfully answers 200. In Apache, point to a local path: ErrorDocument 404 /404.html. With a full URL starting with https://, Apache – per its own documentation – sends a redirect and the client never receives the original error status. The nginx equivalent is error_page 404 /404.html;.
  • Single-page apps and catch-all routes: a wildcard route that renders the app shell for every path returns 200 for everything. Set the status in the server-side handler for unknown routes.
  • Re-test both path shapes after the change.

Check a site now

The free Webskan quick scan runs Lighthouse, security headers, basic SEO checks and a probe for files that should not be public on any public site. No sign-up; results in about 30 seconds. The quick scan does not judge 404 handling – use the script above for that.

Shipping sites with a coding agent? Give it the Webskan MCP server – free for 2 whole-site audits a day, Developer plan for 30. The whole-site audit opens a missing URL in a real mobile browser and reports what it gets: journey.404.soft, journey.404.redirect, journey.404.shows_home.

Method and caveats

  • The list: 527 small Polish online shops that answered a homepage request on 18 November 2026; hand-built by trade, no big chains.
  • On 22 November 2026 each shop got one GET for a random extensionless path; we followed up to six redirects and counted the final status. The 35 shops that answered 200 got a second request for a random path in a subdirectory ending in .html. The other shops were not asked twice, so shops that fail only on .html paths may exist and are not counted.
  • Nine shops served a verification screen instead of a page and one answered with a redirect that had no target; they are not assessed. A person in a normal browser might see something different there.
  • Statements about Google rely on the “HTTP status codes” page in Google Search Central; about Apache, on the ErrorDocument directive in the 2.4 documentation. We do not name the shops.