It depends entirely on which faymas site you are standing on. Measured on 14 October 2026 with one script across 144 pages: on a median faymas.in prompt page, 67.8% of its five-word sequences appear in no other page in a 100-page sample, and two pages picked at random share 4.3% of theirs. On faymas.net's 21 model-version pages the same two numbers are 35.8% exclusive and 28.1% shared — every one of the 210 pairs among those pages overlaps by more than 20%, and 21 of them carry only 9.5 pages' worth of distinct text. faymas-in.com's three posts are the least repetitive things we measured, at 96.2% exclusive.
This guide analyses published prompt text. No generation tests were run.
How we measured it
One script, one set of definitions, three domains. That was the constraint from the start, because a uniqueness figure is worthless if each site gets measured a different way.
Here is the whole method. Fetch a page. Strip script, style and noscript blocks. Remove tags. Unescape entities. Split what is left into lowercase word tokens. Then walk the token list and take every overlapping run of five words as a "sequence" — a five-word shingle. A page of 904 visible words yields 784 of them.
From there, three numbers. Overlap: for any two pages, the count of sequences they share divided by the count in either one. That is the Jaccard index, and it is the fairest answer to "how much of these two pages is the same page". Exclusive share: of this page's sequences, how many turn up in no other page in the set at all. Page-equivalents: the size of the combined sequence pool divided by the size of one median page, which tells you how many pages' worth of unrepeated text a set actually holds.
The samples: 100 prompt pages drawn from faymas.in's sitemap on 12 October, all 41 URLs in faymas.net's sitemap, and all three posts that faymas-in.com's WordPress API returns. One hundred and forty-four pages. Every figure in this piece comes out of that run and nothing else.
Three sites, three answers
| Set | Pages | Visible words (median) | Overlap between two random pages (median) | Exclusive to that page | Pages' worth of text |
|---|---|---|---|---|---|
| faymas.in prompt pages | 100 | 904 | 4.3% | 67.8% | 73.6 |
| faymas.net, all URLs | 41 | 764 | 4.9% | 42.2% | 24.1 |
| faymas.net model-version pages | 21 | 745 | 28.1% | 35.8% | 9.5 |
| faymas.net /prompts/ pages | 12 | 886 | 6.8% | 64.8% | 9.5 |
| faymas.net top-level pages | 8 | 721 | 4.2% | 91.5% | 7.0 |
| faymas-in.com posts | 3 | 3,732 | 2.4% | 96.2% | 2.3 |
Read down the exclusive column. It is the one that answers the question, and the three domains do not land near each other.
faymas.in sits at roughly two-thirds. faymas.net's model pages sit at roughly one-third. faymas-in.com sits at nearly all of it. Three sites sharing one word in their name, three completely different relationships between a URL and the text behind it.
faymas.in: 100 pages really are close to 100 pages
This was not the result we expected. A site with 16,762 prompt URLs, all built from one template, all carrying the same twelve related cards, looked like a candidate for heavy boilerplate. It is not.
The median faymas.in page keeps 67.8% of its sequences to itself. Two pages chosen at random share 4.3%. Across all 4,950 pairs in the sample, only 18 exceed 50% overlap — that is 0.4% of the pairs. The shell is small too: just 47 five-word sequences appear in at least 90% of the 100 pages, which is 6% of a median page. The navigation, the footer and the boilerplate amount to less than a twentieth of what the server sends.
Put differently: those 100 pages contain 57,683 distinct sequences between them, against 784 on a single median page. That is 73.6 pages' worth of unrepeated text in 100 URLs.
| Pages sampled | Distinct sequences found | New sequences per added page |
|---|---|---|
| 10 | 7,976 | 705 |
| 20 | 15,025 | — |
| 40 | 27,185 | — |
| 60 | 37,300 | — |
| 80 | 48,980 | — |
| 100 | 57,683 | 372 |
The return is falling. The first ten pages contributed 705 new sequences each; the last ten contributed 372. That is the shape you would expect once a site starts reusing phrasing across prompts, and it is the honest limit of this measurement: we sampled 100 URLs out of 16,762 and we will not pretend that ratio speaks for the rest. If the last-decile rate of 372 held flat across the full catalogue, 16,762 URLs would hold roughly 7,900 pages' worth of distinct text — our arithmetic, not our measurement, and stated only so the range is visible.
The most repetitive corner of the sample is worth naming. The closest pair we found — 73.3% overlap — is /prompt/realistic-cinematic-portrait-57b5b6c9 against /prompt/dp-portrait-de7fa9bd, with /prompt/bg-ed36a6f9 close behind. Short generic slugs, portrait subjects, near-identical prompt bodies. Recycled pages exist on faymas.in. They are the exception, not the rule, and at 18 pairs out of 4,950 they are not what the catalogue is made of.
faymas.net: the model pages are the repetitive half
Split faymas.net's 41 URLs by shape and the site stops looking uniform. Twenty-one of them are model-version pages — /nano-banana/nano-banana-2, /wan/wan-2-6, /happyhorse/happyhorse-1-1 and so on, one route per model build. Those 21 pages behave nothing like the rest of the site.
Their median pairwise overlap is 28.1%, seven times the site-wide median. The mean is 29.8%. All 210 pairs among them exceed 20% overlap, 101 exceed 30%, and nine exceed 40%. The most similar pair, /happyhorse/happyhorse-1-1 against /happyhorse/happyhorse-1-0, shares 44.9% of its sequences.
Every one of those top pairs is a version number apart. Two builds of happyhorse. nano-banana-2 against nano-banana-pro. wan-2-6 against wan-2-5. That is the explanation, and it is a reasonable one — successive versions of the same model genuinely deserve similar pages. But the consequence is measurable: 21 URLs hold 6,990 distinct sequences, and a median page there holds 733. Nine and a half pages of text, spread across 21 URLs.
The rest of faymas.net is not like that. Its twelve /prompts/ pages keep 64.8% of their sequences exclusive and only one pair out of 66 breaks 20% overlap. Its eight top-level pages are the most distinct group on the site at 91.5% exclusive. So the repetition is not a site-wide habit. It is concentrated in the one section where the site has decided to publish a page per model build.
faymas-in.com: three posts, almost no repeats
Three posts is too small a set to generalise from, and we will say so. But the direction is unambiguous: 96.2% of the sequences on a median post appear in no other post, pairwise overlap sits at 2.4%, and the shared furniture is 100 sequences out of 3,374 — under 3% of a page.
Those posts are also seven times longer than anything on the other two domains, at a median 3,732 visible words. Long and unrepeated is a combination the other two sites do not produce anywhere. It is also the combination that costs the most to write, which is probably why there are only three of them.
What the shared shell actually is
Across all three domains, the part of a page that every other page also carries is tiny. Forty-seven sequences on faymas.in — 6% of a median page. Forty-eight on faymas.net — 6.6%. A hundred on faymas-in.com — 3%.
So the familiar criticism of large prompt libraries — that they are one template stamped thousands of times — does not hold at the level of wording. The chrome is real and it is small. What differs page to page is the body text, and on faymas.in most of that body text is genuinely different from the page next door.
That is worth stating plainly, because it changes what the catalogue number means. Sixteen thousand seven hundred and sixty-two URLs is not sixteen thousand copies of one page. It is closer to sixteen thousand thin pages that mostly say different things.
What this measurement cannot tell you
We read server responses. We did not run a browser. Anything a site assembles after the page loads is invisible to us, which is also true for any crawler that does not execute JavaScript.
Wording is not value. Two pages can be 96% distinct and both useless, and two pages can share 45% of their sequences and one of them be the better page. This measures sameness of text. It does not measure whether any of it is good, correct, or worth copying.
The faymas.in figure is a 100-page sample of a catalogue numbered 16,762 URLs when we counted it on 12 October 2026. A hundred cannot speak for sixteen thousand six hundred and sixty-two. The faymas.net and faymas-in.com figures are complete: that is every URL either site publishes.
And the three-post set from faymas-in.com is too small to carry any claim about that site's habits. It is reported because it was measured, not because it generalises.
Where the sameness actually hurts
Set the two problems side by side and they are different problems. faymas.in has a reach problem we have already measured elsewhere: it holds a five-figure catalogue and shows you twenty of it at a time. faymas.net has a sameness problem, concentrated in one directory: 21 URLs that carry nine and a half pages of text between them.
For a reader, that has a practical shape. If you open three faymas.net model pages expecting three different things, you will find three pages that agree on roughly a third of their wording — and the closer the version numbers, the more they agree. If you open three faymas.in prompt pages, you will mostly get three different prompts.
For anyone publishing in this space, the lesson is narrower and more useful: the page-per-build pattern is the one that produces thin pages, and it produces them predictably. A route per model version is easy to generate and hard to differentiate. It is also the only place in these 144 pages where the URL count and the text count come apart by more than half.
FAQ
How much of a Faymas page is unique to that page?
On faymas.in, 67.8% of a median page's five-word sequences appear in no other page in a 100-page sample. On faymas.net's 21 model-version pages the figure is 35.8%, and on faymas-in.com's three posts it is 96.2%.
Are the 16,762 faymas.in prompt pages just copies of each other?
No. Two pages picked at random from a 100-page sample share a median 4.3% of their sequences, and only 18 of 4,950 pairs exceed 50% overlap. The 100 pages contain 73.6 pages' worth of unrepeated text.
Which faymas site has the most repetitive pages?
faymas.net, and specifically its model-version pages. All 210 pairs among those 21 pages overlap by more than 20%, 101 by more than 30%, and the 21 pages carry 9.5 pages' worth of distinct text between them.
How much of a Faymas page is shared navigation and footer?
Very little. Sequences appearing in at least 90% of pages number 47 on faymas.in (6% of a median page), 48 on faymas.net (6.6%) and 100 on faymas-in.com (3%).
How was uniqueness measured?
One script stripped script and style blocks, removed tags, split the visible text into lowercase word tokens, and took every overlapping run of five words as a sequence. It then compared sequence sets pairwise and counted how many sequences on each page appeared nowhere else.
Does this say whether the prompts are any good?
No. It measures sameness of wording only. Two pages can be 96% distinct and both unhelpful, and the measurement says nothing about output quality, accuracy, or whether a prompt is worth copying.