Plagiarism checker — finds matching published text, never makes accusations

Does this text appear elsewhere on the web?

Paste text or upload a document. We search the public web for it and report every exact match as a quote with its source URL — verifiable evidence, not a verdict. No match is never proof of originality.

Checks the public web. 3 credits per completed check — an abstention or a search outage costs nothing. Nothing you submit is stored unless you rate the result.

How it works

About the plagiarism checker

The plagiarism checker finds whether submitted text matches text already published on the public web. Paste text, a public page URL, or upload a PDF, .docx, .txt, or .md file — it selects the most distinctive passages, searches for them, reads the candidate pages it finds, and reports the exact matches it confirms as quotes you can verify: the overlapping text from both sides, with the source URL. Searches and page reads are capped, so it reports what it found, not everything that exists.

It reports overlap, never accusations. Matching published text can be legitimate quotation, syndication, boilerplate, or your own earlier work — whether reuse was appropriate is a judgement about attribution and intent that only a person can make. And because it searches the public web within capped queries, finding nothing is never a certificate of originality.

  1. 01

    Paste text, a URL, or upload a file

    At least a paragraph of running text. A page URL works too — that page and its own site are excluded from matches.

  2. 02

    Distinctive passages are searched

    The most identifying phrases across the whole document are selected and searched on the public web, within a capped number of queries.

  3. 03

    Candidate pages are read and aligned

    Pages the search finds are fetched and compared word by word. A match is only reported when it is confirmed against the actual page text — never from a search snippet.

  4. 04

    You get matches with sources

    Each confirmed match is a quote with its source URL and position in your text. A separate paraphrase tier — clearly-labelled model judgements — is currently disabled while its evaluation completes.

What you get back

Overlap band and share

None, low, notable, or high — with the share of your words that matched published text.

Exact matches with quotes

The overlapping text from your document and from the source page, with the URL. Verifiable by opening the source.

Possible paraphrases, labelled

Reworded overlap flagged by a model — a judgement, not a quote, listed separately and never counted like an exact match. This tier is disabled until its measured evaluation passes; results say so honestly.

Honest scope

What was searched, which found pages could not be read, and an abstention when the text is too short, not prose, or not yet a supported language.

Common uses

  • Checking an essay, article, or report for copied passages before it is submitted or published
  • Finding where a suspicious passage was published, with the URL as evidence
  • Editorial and agency workflows: screening contributed copy for passages matching published web pages
  • Checking your own work does not accidentally duplicate published text

A plagiarism-style check finds matching published text — it does not judge intent, attribution, or whether reuse was legitimate. Use matches as evidence for a human decision, and never treat “no match” as proof of originality.

Questions

Is this the same as Turnitin?
No. Turnitin and similar academic tools match against private databases of submitted student work and paywalled journals. This checker searches the public web — pages a search engine can reach. It can find text copied from websites, news, blogs, documentation and public papers, but it cannot see into private essay databases. The two answer different questions, and neither certifies originality.
Does “no match” mean the text is original?
No. It means no matching published text was found within the searched scope — a capped number of web searches over the most distinctive passages. Paywalled sources, academic databases, private documents, and anything search engines have not indexed are outside that scope. Treat “no match” as an absence of evidence, never as a certificate.
What is the difference between an exact match and a paraphrase flag?
An exact match is deterministic: the same run of words (allowing case, punctuation and accent differences) appears in your text and on the source page, and both quotes are shown so you can verify it yourself. A paraphrase flag is a model’s judgement that a passage expresses the same specific content in different words — it can be wrong, it is clearly labelled, and it is never counted like an exact match. The paraphrase tier is currently disabled while its evaluation completes; until then results report exact matches only, and say so.
Is a match proof of plagiarism?
No. A match proves the same text appears at the source URL — nothing more. It does not say who published first, whether the passage is a legitimate quote, common boilerplate, syndicated content, or the same author’s earlier work. Plagiarism is a judgement about attribution and intent; this tool hands a person the evidence and never makes that judgement.
Can I check a PDF or Word document?
Yes — upload the file instead of pasting. PDF, DOCX, TXT and Markdown are read, and the running prose inside is checked the same way as pasted text. Very long documents are sampled across their whole length, and the result reports how much was covered.
What does it cost?
A completed check costs 3 credits. If the text is too short, not prose, or not yet a supported language — or if web search is unavailable — the check abstains and costs nothing. Every response carries its own price in the Credits-Cost header.
Do you store what I submit?
Not unless you rate the result. Checking is stateless: your text is assessed and not retained. Distinctive phrases from it are sent to web-search providers as queries, and passages may be compared by a model for the paraphrase tier. If you rate the result, we keep the text alongside your rating for 120 days to improve the check.
What languages does it support?
English-language prose. Text mostly in another language abstains with an honest message rather than returning an unreliable result.

Developer recipes & tools