Nobody Copies Anymore. Everybody Agrees. I Measured It.

The copy-the-top-result playbook is being retired. I wanted to see the body.

For a decade the content playbook was one sentence long. Read the top ten, write the eleventh, make it longer. It worked because the ranking systems rewarded coverage, and coverage is easy to fake. Nobody said this out loud in a proposal, but the outline tool in every SEO suite was built around it. (The tool is still there. It is not well.)

Two things changed the price of that. A survey of 131 SEO experts published this month put original research and first-party data at the top of the content-quality factors. And Google Research published a paper on Retrieve-for-Train on September 15 that trains the sub-queries behind AI search to avoid what it calls paraphrastic collapse, near-synonymous queries that all say the same thing, and scores their breadth with a Vendi Score. Read that carefully. It is about the questions the machine asks, not a filter on your page. But the consequence is the same: the engine reads a topic from many angles, and a page that covers the one angle everyone else covers is competing for one sub-query out of many.

I did not want to argue about it. I wanted to measure how alike the top results actually are. So I took two questions my clients get asked on the phone every week and pulled the page text behind the top ten.

Two questions, twelve pages, zero copied sentences, one outline

Question one: how much does a car wrap cost. Seven organic results. Two were a Reddit thread and an Edmunds page my scraper barely got text from, so five pages got measured. Shared sentences between any two pages: zero. Shared five-word phrases: half a percent at most. By the plagiarism checker's standard, five independent works of literature.

By any other standard, one document. Five of five cover cost, four of five cover paint versus wrap, four of five cover how long it lasts, four of five bring up the installer. And the numbers repeat like a chorus. $1,500, $2,000, $2,500, $3,000, $4,000, $5,000, and $10,000 each appear on two different domains. Nobody copied. Everybody agreed, in the same order, with the same figures, and called it their take.

Google's AI Overview for that question said a full wrap runs $2,500 to $6,000 and cited three sources: two YouTube videos and one written page, the one with 21 dollar figures on it, the most specific page in the set. It did not cite 3M, which makes the vinyl, or Edmunds, which is Edmunds. The brand did not win. The number did.

Question two: how long does it take to remodel a kitchen. Six results, and the top two are a Reddit thread of 46 words and a Facebook post, because the searcher wants a person who did it, not a person who wrote about it. Then HGTV, Sweeten, 180 Kitchens, and a regional design-build firm. All five written pages cover the timeline and mention a contractor. Three cover permits. Zero share a sentence. The AI Overview cited HGTV, a YouTube video, and the 1,883-word page with the most detailed week-by-week timeline. Again the most specific page, not the most famous one.

Sameness is not plagiarism anymore. It is agreement.

The old originality test was a similarity score. Pass it and you were original. Every page in my sample passes it. And every page in my sample is interchangeable, because they were all built from the same outline, the same competitor pages, and the same three cost guides that everyone cites and nobody checks. A model reading seven versions of the same outline needs one, and it picked the one with the most checkable detail both times.

So the question for a page is no longer whether it is different enough to avoid a penalty. It is what the page has that no other page on the internet has. For most of my sample the honest answer is a logo. (The logo was doing a lot of work. It is tired.)

This lines up with the citation test I ran last week, where three engines agreed on one wrap shop page for one reason: it stated the price in its first sentence. That piece is here. Specificity got cited. Coverage got ignored.

What original looks like when you are a wrap shop, not a research lab

One. Your own numbers, from your own jobs. Not the industry range. The range of your last forty tickets, with the median and the two outliers explained. A remodeler who writes that its own kitchens finished in a stated number of days on site, from its own schedule, has published a figure that does not exist anywhere else on the internet. That is original research. It cost nothing. It was in the calendar.

Two. The question the query data proves people ask and nobody answers well. From one client's Search Console this month: thirteen phrasings of can you wrap a car with dents or scratches, 209 impressions in 28 days, one click, average position 8.4. Thirteen phrasings, one question, and the shop's answer sits eighth. An original answer here is not clever. It is a photo of a dented door, before and after, with the price and one sentence about what vinyl will and will not hide.

Three. A named human who will defend the claim. The survey ranked this alongside first-party data for a reason. A model corroborates. An author with a profile, a license number, and a photo on the job is corroboration. An anonymous services page is a claim.

Four. Publish less. The practitioners winning this are not writing more pages. They are writing the one page with the thing in it, then letting the engines do what they did in my sample, which is skip the famous page for the specific one.

Find the thin spot, then check the draft before it ships

The part of this I could not do by hand is knowing where demand exists but coverage is thin. In the Semrush Content Toolkit, Topic Finder groups the question universe around a term into topics with demand attached, so you invest the original work where there is a gap instead of adding an eighth version of the cost guide. The question report for kitchen remodel is instructive: do you need a permit to remodel a kitchen gets 320 searches a month at a keyword difficulty of zero. Zero. For car wrap, how long does it take to wrap a car gets 1,000 a month at a difficulty of two. Those are unclaimed. Nobody is agreeing there yet.

Then the check. SEO Writing Assistant scores a draft for originality and readability inside the editor, which is where the rewrite-of-the-top-result gets caught, before it ships and joins the chorus. And the AI Visibility features in Semrush One tell you whether the original piece is earning the citation, which is the only score that matters now.

Seven pages, one outline. Be the eighth page or be the one with the number.

The reason your own site is the least trusted voice about your business, and how outside sources decide what is true, is in the entity corroboration field guide. How an assistant builds its shortlist in the first place is in the shortlist piece.

Twelve pages is not a study. It is two phone questions and an afternoon. But the result did not need a bigger sample: the pages were original by the checker and identical by the reader, and the machine picked the one that added a number. Original is the part you can add and nobody else can. Everything else is agreeing in a different font.