How much of your page is actually your page

Some templates are structurally prone to this, and recognising yours in the list is faster than measuring.

• Location and service-area pages. The classic case, and the one this check was effectively built for. Twelve pages, one paragraph, a swapped city name, and the same 600 words of trust signals underneath. Each page is over any reasonable length threshold and has almost nothing of its own.
• Product pages with a manufacturer description. The unique region is a title, a price and a specification table. The rest is a description shared with every other retailer, plus your own returns policy, delivery terms and cross-sell blocks. Two problems at once, and only one of them is visible to this check.
• Category and tag archives. A heading, a list of links to other pages, and the site chrome. There is often no unique prose at all, which is defensible for a category people navigate and hard to defend for the four hundred tag pages a CMS generated automatically.
• Documentation with a large shared sidebar.An honest 200-word reference page sitting inside a 900-item navigation tree. The content is correct and useful, and the ratio looks alarming. This is the most common false alarm in the check, and the fix is marking the tree as navigation so it can be recognised as chrome rather than as text.

The last one matters because it points at the honest limit of the measurement: a low ratio is a question, not a verdict. On a location page the answer is that the page has nothing to say. On a reference page the answer is that the page says one thing precisely and the template is heavy.

Yes, and this is one of the checks where the AI answer is stronger than the search answer. Retrieval works on passages rather than pages: a system splits your document into chunks and decides which chunk answers a question. A page whose unique region is one short band produces exactly one chunk worth retrieving, and a great many chunks made of your navigation and your returns policy.

Those boilerplate chunks are not harmless. They are near-identical across your whole site, so they compete with each other and with themselves, and a retrieval system that keeps picking your footer is a retrieval system that never reaches your content. Sites with a low uniqueness ratio effectively bury their own material in copies of their own chrome.

The remedy is the same one that helps human readers and it is not a schema or a tag. Mark the repeated regions with the elements that name them, so a chunker can drop them, and give the unique region enough substance to be worth one good chunk. That is the same argument segmentation makes, arrived at from the other end.

You do not need a crawler to get within a few points of the real number. Take two pages that share a template and are about different things, and count what they have in common.

words() {
  curl -s "$1" \
    | sed -e 's/<script[^>]*>.*<\/script>//g' \
          -e 's/<style[^>]*>.*<\/style>//g' \
          -e 's/<[^>]*>/ /g' \
    | tr -s ' \n' '\n' | sed '/^$/d' | sort -u
}

words https://example.com/plumbers/leeds > /tmp/leeds.txt
words https://example.com/plumbers/york  > /tmp/york.txt

shared=$(comm -12 /tmp/leeds.txt /tmp/york.txt | wc -l)
total=$(wc -l < /tmp/leeds.txt)

echo "shared with the sibling page: $shared of $total words"
echo "roughly $(( 100 - shared * 100 / total ))% of this page is its own"

This counts distinct words rather than running text, so treat the output as an indicator and not a measurement. Anything under 20 percent unique on two pages that are supposed to be about different things is a template producing pages rather than a person writing them.