The check that finds nothing because nothing was checked against
You paste your essay into the box, leave the second field empty, run it, and the panel reports zero matched sentences and zero percent overlap. That is not a clean bill of health. It is a comparison against an empty string, and every text is original next to nothing at all.
This tool compares your draft with a source you supply. It does not search the web, does not query an academic database, and does not detect AI writing. Those are not omissions waiting to be fixed in a later version; they are what a tool running entirely inside your browser can possibly do, and knowing the boundary is what makes the results usable.
Why a browser cannot search the web for you
Finding matching passages across the world's published text needs three things a page in your browser does not have: a copy of that text, an index over it, and permission to use both. Commercial services crawl pages, license journal collections, and archive submitted papers, then keep server farms to search the result. Their reports are largely a statement about the size of their library.
A local tool has none of that. It could forward your draft to a search engine, but then your draft has left your device, which is the very thing many users of this page are trying to avoid. So the comparison stays local and the corpus is whatever you paste in. Anything you did not provide is invisible, and no percentage can tell you about it.
What the sentence match actually does
Your draft is split into sentences. Each one is reduced to its words, lowercased, and checked against the source, which has been flattened into a single stream of words the same way. A sentence matches when its complete word sequence appears somewhere in that stream.
| Draft sentence | Source contains | Result |
|---|---|---|
| "Participants were randomly assigned to two groups." | the same words | Match |
| "Participants were randomly allocated to two groups." | "assigned" | No match |
| "Improves recall fast." | the same three words | No match |
| "Morning practice improves recall." | those words inside a longer sentence | Match |
Three details do real work there. Punctuation and capitalization are ignored, so a match survives a changed comma or a missing full stop. Matching is by whole sentence rather than by fragment. And a draft sentence of fewer than four words is skipped before the comparison starts, which is why the three-word row above misses even though the words are right there in the source.
All or nothing, which is the property that shapes every result
A sentence either appears in the source or it does not. There is no partial credit within a sentence, and that single property explains most of what surprises people.
Change one word in a seven-word sentence and the match vanishes completely. "Participants were randomly assigned to two groups" against a draft that says "allocated" reports zero matched sentences and a word overlap of 0.0%, even though six of the seven words are identical. The overlap figure counts only words inside sentences that matched in full, so it moves in jumps rather than drifting.
Sentence splitting causes misses of its own. The sentence segmenter treats abbreviations as boundaries, so a draft reading "Dr. Smith wrote it. He finished the draft on Tuesday afternoon." is handled as three sentences: "Dr.", "Smith wrote it.", and the remainder. Two of the three fall under the four-word minimum and drop out of the comparison altogether. Against a source that was word-for-word identical, that draft reports one matched sentence out of three and a word overlap of 63.6%.
What it cannot tell you, and what to do instead
The comparison cannot see paraphrase, and it cannot see idea-level borrowing, where nothing was copied but the structure and the argument arrived wholesale from a source. It also cannot separate a properly quoted and cited sentence from a stolen one. Quotation marks are punctuation, and punctuation is stripped before comparison, so an attributed block quotation matches exactly like an unattributed one.
The definitions worth keeping straight are set out in what counts as plagiarism, which covers the attribution question this tool raises but cannot answer. Plagiarism and copyright infringement are also different offences. Plagiarism is a failure of attribution, judged by institutions and readers. Infringement is a legal matter about permission and the specific rights a creator holds. You can plagiarise a work you legally own, and you can infringe a copyright while citing the source correctly.
There is one more case this comparison handles unusually well, and most people never think to run it: your own earlier work. Reusing your own paragraphs without saying so is a real problem in academic and professional settings, and the source you need to paste is a document you already have. Self-plagiarism explained covers where reuse crosses that line.
Getting a result worth trusting
Run the comparison against the sources your draft was actually built from, one at a time or pasted together: the article you summarised, the brief you answered, the client's reference document, your previous draft. A clean result against those is meaningful, because you checked the texts that could plausibly be the origin of your phrasing, and finding nothing there is genuine information.
A clean result against some unrelated web page is not meaningful, because that page was never a candidate.
Expect quotation-heavy writing to score high and stay calm about it. A literature review that quotes properly will match its sources repeatedly, and the matched sentences the tool lists are the ones to leave alone once the quotation marks and citations are in place. Paste the reference list too, and every entry will match as well, since a citation is a copied string by design.
Frequently Asked Questions
Does this tool check my text against the internet?
No. It compares your draft only with the source text you paste into the second field. There is no web index behind it, no journal database, and no server involved, so a failure to find matches against material you never supplied is expected rather than suspicious.
Why did a sentence I copied not show up as a match?
Two likely reasons. Sentences under four words are skipped entirely, and a sentence with even one changed word no longer matches, because matching requires the complete word sequence. A three-word phrase lifted verbatim is invisible to this comparison by design.
Does it detect paraphrased text?
No. Paraphrase replaces words, and this tool matches word sequences, so a genuine restatement scores zero. It catches copy-and-paste and near-verbatim borrowing. The rewritten kind is the case that no local word-matching tool can reach.
Why is my word overlap 63% when I wrote the passage myself?
Sentence splitting, most often. Abbreviations, initials, and numbered lists can cause the segmenter to cut one sentence into two, and any fragment under four words drops out of the comparison. Compare the counted-sentence figure against the number of sentences you actually wrote.
Can I check a document against a PDF or a Word file?
Only after extracting the text yourself. Paste the plain text of the source into the field. The tool compares characters rather than document structure, so a page number or a running header caught in the middle of your paste becomes part of the word stream you are matching against.
Paste the source, not a sample of it
The next time you check a draft, change one habit: spend the extra minute collecting the real sources, paste them all in, and then read the matched sentences rather than the percentage. That list is the deliverable. Any sentence on it you cannot point to a quotation mark or a citation for is the work still to do; everything else is accounted for.