Most of Nakodo is generated from a content registry: 63 of its 76 public pages are written as data and rendered by one component. That is the setup people mean when they say "programmatic SEO", and the standard worry about it is duplication. I had never measured ours, so I did.

Method

Fetch all 76 pages from the sitemap, strip script, style, svg, template and the whole head, normalise whitespace, split on words, then shingle each page into overlapping eight-word sequences. Two pages sharing an eight-gram are saying the same thing in the same order, which is close enough to what a search engine's near-duplicate detection is looking at.

words = re.findall(r"[A-Za-z0-9£%'’-]+", visible_text) grams = {tuple(words[i:i + 8]) for i in range(len(words) - 7)} 

The corpus: 126,382 words across 76 pages, median 1,706 a page, from 405 on the shortest to 3,392 on how it works.

The duplication profile is bimodal

102,240 distinct eight-grams exist on the site. Here is how many pages carry each one:

Appears on Distinct eight-grams
exactly 1 page 97,503
2 to 5 pages 4,035
6 to 20 pages 535
21 to 50 pages 2
51 to 75 pages 38
all 76 pages 127

95.4% of the phrasing on this site exists in exactly one place. Almost nothing lives in the middle of that table: only two sequences in the entire corpus appear on between 21 and 50 pages. Text here is either unique or it is site furniture, and the furniture is countable.

The furniture is 141 words, plus a 44-word call to action

Growing the shingle length until the all-76 set collapses shows what the universal text actually is:

n= 8: on all 76 pages = 127 sequences n= 25: on all 76 pages = 105 n= 60: on all 76 pages = 70 n=120: on all 76 pages = 10 

Reconstructed, it is the marketing header and the footer: the nav, the five footer columns of guides and niches and audiences, the company links, the legal line. Exactly 141 words of every page are covered by text that appears on all 76. Only 98 distinct lowercase words appear on every page.

One tier down there is a second block, on 72 of the 76 pages, 44 words long:

Find creators who already fit your brand. Nakodo reads your website, searches YouTube, Instagram and TikTok, scores each creator from 0 to 100 with the reasons, and emails the best fits for you. Free plan, no card. Start free. How it works. Read next 

That is the standing call to action at the bottom of every content page. 185 words of furniture per page is cheap on a 1,706-word median, but the share varies by a factor of five:

Page Words Covered by all-76 text
/privacy 2,661 5.3%
/glossary 2,335 6.0%
/ 1,835 7.7%
/pricing 1,451 9.7%
/guides 1,065 13.2%
/youtubers 580 24.3%

Six pages are mostly made of other pages

Strip the all-76 furniture, then ask for each page what share of its remaining eight-grams appear somewhere else on the site. The median is 13.2%. Six pages are over 50%:

Page Non-furniture eight-grams Share found elsewhere Biggest overlap
/youtubers 446 68.4% /tiktokers 19%
/pricing 1,242 67.7% the home page 68%
/guides 920 65.2% spread over 20 guides
/alternatives 271 57.9% each comparison page 44%
/ 1,678 55.0% /pricing 50%
/instagrammers 298 52.3% /tiktokers 29%

Every one of them is an index page, and the mechanism is the same in each case: an index lists its children, and the text it lists is each child's own title and description, which also appears on the child. /youtubers is two sentences of its own followed by eight niche cards, each card carrying the sentence that opens the page it links to. 583 words, of which roughly 440 belong to other pages.

The hub pages also overlap each other, because "Find creators who already fit your brand" and the pitch for scoring appear on all three of the platform indexes. That is the 51-to-75 bucket showing up again.

The home-and-pricing pair is a different case, and a deliberate one. 837 shared twelve-word sequences between them, which is the whole plan comparison and most of the plan FAQ:

For a brand running creator outreach every week. The first 7 days... and notes, asks you about anything else, and hands the conversation to... a published email is never contacted or counted, and neither is a... 

Prices and plan rules are on the home page because people ask for them before they click anything, and they are on the pricing page because that is where people go to compare. I would rather have that duplication than a home page that makes you navigate to find out what it costs.

The pages that owe nothing to anyone

At the other end, the share of each page's text found elsewhere:

/privacy 1.4% /terms 2.3% /glossary 3.2% /how-it-works 7.3% /guides/spark-ads-and-partnership-ads 8.8% /blog/ugc-creators-vs-influencers 9.5% 

The four most unique pages on the site are the four written entirely by hand, outside the registry. The generated pages come in at a median of 13.2% and the worst non-index generated page is at 30.3%: the four alternatives comparisons, which share a page structure and a set of stock arguments by design.

What I am changing

  1. The index pages get their own prose. Not more cards, a paragraph that only makes sense on that page. /youtubers at 24% furniture and 68% borrowed is the weakest page on the site by this measure, and it is the one a search engine is most likely to read as thin.
  2. The 44-word call to action stays. It is on 72 pages, it is the reason the site exists, and 44 words against a 1,706-word median is not what makes a page thin.
  3. This becomes a test. We already have a test that enforces a 600-word floor and title and description lengths on every registry page; the natural addition is a unique-share floor, computed the same way, failing the build when a new page is more than half somebody else's words. Measuring once tells you where you are. A test tells you when it changes.