Originality is not new ideas. It is detail nobody else had.

It is not about a novel thesis. It is about whether the page contains a fact a competitor could not write without doing what you did: a measurement, a process, a failure, a price, a customer's exact words.

A page can be entirely conventional in argument and still original, because the evidence is yours. A page can also be radical in argument and unoriginal, because the argument came from somewhere and no evidence supports it.

This definition matters because it is achievable. Nobody can be told to have an original idea, and almost anyone can be told to write down what they measured this week.

It also survives scrutiny. A page with your data is defensible even if the prose is ordinary, while a beautifully written page with nothing behind it is not.

Plagiarism is copying text or ideas without attribution. Originality failure is worse disguised and harder to detect: nothing was copied, and nothing was added either.

A generated page is typically a recombination of public material, presented without sources and without the structure that made any of the sources useful. It is not theft; it is dilution.

This is why plagiarism checkers return clean on pages that still read as hollow. They measure overlap with other text, and the failure is an absence rather than an overlap.

Judge the page by what a reader gains from it and what they could not have found in five other results. If the answer is nothing, the page fails on originality whatever a similarity score says.

A model produces the most probable continuation, given the prompt and the text so far. The most probable text about a subject is the text most often written about it, which is the consensus view.

It is also trained to be inoffensive and broadly applicable, so it hedges where you would commit and generalises where you would name a number. Both behaviours are safety features that become a quality problem in publishing.

The prompt usually supplies nothing to work with. A brief that names a keyword and a word count gives the model no material it does not already have, so the only possible output is a rearrangement of what everybody has.

This is testable in ten minutes. Give the same model a brief with one measurement and one constraint, and you get a page that says something the generic version cannot. The difference was never in the model.

You do not need all four. One is usually enough to make a page worth reading, and two makes it citable.

Unoriginality has a vocabulary. When a sentence reaches for a vague quantity or an unnamed source, it is usually because the writer did not have the figure or the citation.

Studies show, experts say, many businesses, a variety of strategies, best practices: each is a placeholder where a fact should be, and each is detectable at a glance by a reader who is looking for substance.

The fix is not to delete the sentence, which leaves a gap, but to go and find the number. If the number cannot be found, the claim is not yours to make and the sentence goes.

A page that has three real figures and no vague claims is a different product from one with twelve vague claims and no figures, even at the same word count.

Five questions, asked of yourself or a colleague, will usually produce enough for several pages.

What did we measure or count, and what were the conditions? What surprised us, and what did we expect instead? What does everyone else get wrong about this, concretely? What does this cost in money, hours or risk, with a number? What would we tell a customer who asked this on a call?

Record the answers verbatim. The wording of a real answer is more useful than a polished paraphrase, and it is the part a reader recognises as human.

Keep the answers in a file. A single afternoon of questions can supply a quarter of pages, and the file is the difference between a content plan and a keyword list.

Editors smooth prose, and smoothing removes exactly what makes a page original: the odd detail, the exact number, the sentence that sounds like a person. Protect those before the edit, not after.

Mark the sentences that carry evidence and tell the editor they are fixed. A small convention, such as leaving them unmarked and instructing that no figure may be removed, saves the round trip.

Watch for the swap that editors make instinctively: turning a measured result into a general trend. It reads better and says less, and it is the moment a page becomes ordinary.

After editing, compare the finished page with the notes. If every number survived, the edit was a success even if the prose changed entirely.

A site can be original page by page and derivative as a whole, if every page covers the same consensus from a different angle. The set needs a position, not just pages.

The test is whether someone could summarise what your site believes about the subject that differs from the default. If the summary is the same as the first five results, the site has no position.

Practically, that means one page per real question, with the strong pages carrying the argument and the others supporting it. Fifteen strongly held pages beat eighty that agree with everything.

It also means deleting pages that exist only to occupy a phrase. They dilute the position and teach readers that the site is a catalogue.

Take a paragraph and replace its subject with a different subject from the same category: your product with a competitor, your city with another city, one integration with a different integration.

If the paragraph is still true and still useful, it was generic. Nothing in it depended on the specific case, which means the page did not need to exist in that form.

If the paragraph breaks, you have found the original material. A sentence that becomes false when the noun changes is a sentence that carries information.

Run the test across a page, then across a cluster. The paragraphs that survive the swap are the ones to keep, and the rest is the rewrite brief.

Three options, in order of preference. Add one form of originality from the list above and keep the page. Merge it with a page that already has evidence. Delete it and redirect.

The wrong option is to reword it. A paraphrase of the consensus is still the consensus, and it costs the same effort as adding a measurement to a page that already has a point.

If you cannot add anything, that is information: the page covers a question you have no standing to answer, and publishing it mainly advertises that.

Being absent from a topic is survivable. Being present and empty is the thing readers remember.