You have a site with 200 pages and a nagging feeling that some of them are living second lives on domains you have never heard of. Checking all 200 is not going to happen, and it does not need to. Scrapers do not copy sites evenly, they copy what ranks, which means a whole-site check is really a prioritization problem wearing a technical costume.
Here is the system I use on client sites: a priority list, a fast fingerprint sweep for breadth, full scans for the pages that matter, and a standing alert so you stop doing this manually. An afternoon covers a mid-sized site, and the quarterly repeat takes an hour.
Build the list before you check anything
Open your analytics and pull three sets: your most-visited pages, your highest-converting pages, and the evergreen pieces that earn steady search traffic. The overlap of those sets is your top tier, usually 10 to 20 pages. Everything else is second tier.
The logic is unsentimental. A scraped page costs you something only when the copy competes with the original, for rankings, for readers, or for the credibility of being first. Your contact page cannot be meaningfully stolen. Your best-performing guide is the single most likely thing on your domain to be living somewhere else, precisely because it performs.
The fingerprint sweep: fast, wide, surprisingly effective
For every page on the list, pick two or three distinctive sentences from the middle of the text, not the introduction. Openings get rewritten by even lazy scrapers, middles get pasted. A good fingerprint sentence is specific to you: an example with your numbers in it, a phrasing nobody else would produce, eight to fifteen words.
Search each one inside quotation marks. The quoted search asks for that exact string, so a clean result is meaningful evidence nothing indexed carries the sentence, and a hit takes you straight to the copy. Two or three fingerprints per page catches partial lifts, where someone took one section rather than the whole piece.
Bad fingerprints waste the sweep, so screen them. Anything that could plausibly appear on another site unchanged fails: product boilerplate, legal phrasing, a definition, a sentence built from the standard vocabulary of your industry. If a candidate sentence would read naturally on a competitor's page, pick another one.
This is quick in practice. Batching it, a spreadsheet of pages and fingerprint sentences, gets through ten pages in well under an hour, and the spreadsheet becomes the asset you reuse every quarter with fresh sentences. Log every hit in it as you go, with the URL, the date, and which bucket it landed in, so the next sweep starts from a record instead of a memory.
Full scans for the top tier
Fingerprints sample the text. For the pages where a miss actually costs money, scan the whole thing. Paste the full text into PlagiarismCheck.io, and before you run it, put your own domain in the exclude-URLs setting, otherwise the top match for every page will be the page itself and the report becomes noise. The 2,000-word cap means a long guide goes through in two or three chunks, which is the honest cost of a free tool, so budget for it rather than being surprised by it.
What you get back that a quoted search cannot give you: every overlapping source with a link to each, similar-text matching that catches light rewording, and a view of which sections travel and which stay home. Ten full scans covers most sites' entire top tier.
Set the alert and stop checking manually
The maintenance layer costs nothing and runs while you sleep. Create a Google Alert for one or two of your best fingerprint sentences, in quotation marks, and the day a new page carrying them gets indexed, you get an email. Rotate the monitored sentences when you publish something new that matters.
Add two passive tripwires from data you already have. A referrer you do not recognize sending trickle traffic is sometimes a scraper whose copied page kept your internal links, which means it is linking back at you. And a sudden ranking drop on a stable evergreen page is occasionally the moment a copy started competing, which turns an SEO mystery into a plagiarism check with a specific target.
Triage: not everything you find is theft, and not all theft is worth chasing
Sort every hit into one of three buckets before you act. Legitimate: syndication you agreed to, quotes with attribution and a link, feeds you enabled. Harmless-but-annoying: a dead blog with twelve readers that pasted your post in 2023. Costly: a live site republishing your content wholesale, ranking for your phrases, or stripping your attribution.
Only the third bucket earns your time. For those, the sequence is documented in the takedown walkthrough: archive the evidence, email the operator, notify the host, then delist. And before spending an afternoon on any of it, ask the colder question of whether the copy is hurting you at all, because duplication is a cost, not a penalty, and a copy that neither ranks nor converts is mostly an insult, not an injury.
The same check, pointed at a site you do not own yet
Run this system in reverse before you buy a site, take on a content client, or inherit a blog from a predecessor, because the question flips: is this site the original, or the copy? Fingerprint its top pages the same way and see whether the sentences exist elsewhere with earlier dates. Full-scan its cornerstone content and look at where the matches point.
A site whose flagship pages match older pages on other domains is carrying someone else's text, and whatever you paid for it, you also bought the takedown notice that may eventually arrive. Ten fingerprint searches during due diligence is the cheapest insurance in content acquisition, and almost nobody runs them.
What this system will miss, said plainly
Text checking finds text. A copy that has been competently rewritten shares no strings with your original and will pass both the fingerprint sweep and the full scan. A translated copy is invisible to any scan run in English. Structure theft, your outline and examples in fresh words, is detectable only by a person comparing the two pages.
Accept those blind spots rather than buying your way around them, because the misses share a feature: heavily rewritten copies rarely outrank originals, since the rewriting that beats detection also dilutes the thing that ranked. The copies that hurt are the lazy ones, and the lazy ones are exactly what this system catches.
The whole routine on one line
Quarterly: refresh the priority list, sweep fingerprints across it, full-scan the top tier with your domain excluded, keep one alert running, and act only on copies that compete. First run this week, an afternoon. Every run after, an hour. The point is not to find everything. It is to make sure that anything worth finding cannot stay hidden for long, and a free scanner, a search box, and a spreadsheet are genuinely all it takes.