Plagiarism Checker

Paste one document and search its passages against indexed public web pages. See the matching sources and the exact text worth reviewing.

Document to check 0 words

No word limit · files up to 20 MB are read here · extracted passages are sent securely for public-web search

Three steps

How an online check actually runs

There is one document input. A file is converted to text in your browser, then query-sized passages go to the search endpoint and are checked against indexed public pages.

01Add the document

Paste a document of any length, or choose a PDF, Word, RTF, Markdown or text file. The browser extracts readable text before the online check begins.

02Passages are searched online

The text is divided into meaningful exact-phrase queries and a server-side search provider looks each one up. A long document is sent in parts, because one request can only issue so many searches.

03Sources and passages are reported

The result shows a weighted similarity percentage, every passage checked, source titles and links, and the original document with passages that returned sources highlighted.

Try it on a document
Three outcomes

The three shapes an online result takes

A similarity figure on its own says very little. Correct quotations, standard wording and copied paragraphs can return the same source links. The result therefore keeps the passages and their context beside the percentage.

Substantial overlap
74 typical share

Long runs, from one source

Several long passages return the same page or closely related pages. That pattern deserves careful review, especially when the wording is not quoted or attributed.

Scattered overlap
18 typical share

Short runs, spread thin

Quotations, references, legal language, definitions and standard technical phrases often return sources. They are real overlaps but are not misconduct by themselves.

Nothing matched
02 typical share

No indexed result returned

None of the exact passages returned a public page from the configured search index. Private repositories, paywalled archives and unindexed pages remain outside that result.

In the wild

What people bring to a comparison

Almost always a document and a suspicion about where part of it came from. These are the cases that arrive, and every one of them is a comparison against sources somebody already has.

  • Coursework against sources

    An essay and the three articles on the reading list it was supposedly written from.

  • Two submissions

    Two students, one paragraph, and a conversation nobody wants to have.

  • Recycled coursework

    Work submitted again a year later, to a marker who kept the original.

  • Draft against a published article

    A filed piece and the published article it reads uncomfortably close to.

  • Contract and proposal text

    A bid that reuses a competitor clause for clause.

  • Ghostwritten work

    A submission and the file a freelancer sent the week before.

  • Website and product copy

    Copy lifted wholesale from a competitor, down to the typos.

  • Research manuscripts

    A methods section that appeared first in somebody else preprint.

  • Self-repetition

    A dissertation that says the same thing twice, three chapters apart.

  • Grant applications

    An application that reuses a funded proposal without saying so.

  • Translated reuse

    Two reports from the same office, one clearly built from the other.

  • Checking your own draft

    A draft checked against its own sources before anybody else sees it.

Reading the number

What the percentage says, and how firmly

The figure is the share of checked words that sit inside a passage for which the web search returned at least one indexed public source. A passage returned by three pages still counts once, so the overall figure cannot exceed one hundred per cent.

It is a search-coverage measurement, not a judgement about authorship. The marker at thirty per cent is a reading aid rather than a decision line. A correctly cited paper can cross it, while rewritten copying can stay below it.

The five public-web similarity bands, what each means and what to review next
Share Band What it means What to do next
0 – 5 Nothing matched No checked passage returned a source from the configured public-web index. Do not call it cleared. The source may be private, unindexed, paywalled, translated or paraphrased.
5 – 15 Incidental overlap A small number of passages returned indexed pages, often because of quotations or shared phrasing. Open the sources and confirm that quoted or borrowed wording is attributed correctly.
15 – 30 Worth a look Several passages have online matches, so their location and context start to matter. Check whether each match is quoted, cited, common language or presented as original work.
30 – 60 Substantial overlap A large share of the checked words belongs to passages that returned public sources. Review the leading source links, dates and highlighted blocks. Do not decide from the number alone.
60 – 100 Largely the same text Most checked passages returned one or more indexed sources. Establish publication order and attribution before treating the overlap as misuse.

The bands are fixed and published, so repeated checks can be read consistently even when the underlying index changes.

A high figure is not an accusation and a low one is not a clearance. The result covers only the public pages returned for the exact passages searched at that moment.

Under the hood

What the web search is actually doing

The document is normalised and divided into passages that usually contain about twenty-six words. The split prefers sentence endings and never creates a query below eight words. Every word remains covered, so a short tail at the end cannot silently disappear from the score.

Each passage is sent as an exact quoted query through the Original or AI server endpoint. The endpoint uses a server-held search credential, accepts only same-origin JSON requests and returns no cached response. The credential is never placed in page code or local storage. A long document is sent in parts, because one request can only issue so many searches before a serverless function runs out of room.

If the provider returns public pages, that passage is marked and the page titles, addresses and snippets are shown. The overall percentage is weighted by words rather than by sentence count, so a four-word fragment cannot influence the score as much as a full paragraph.

This does use the network. Files are parsed locally, but their extracted text is divided into passages and sent through our Cloudflare-hosted endpoint to the configured search provider. We do not add that text to a submissions archive. The provider and ordinary hosting logs remain subject to their own policies.

What is measured, and what is not

  • Measured: the word-weighted share of passages that return at least one indexed public page
  • Measured: which public URLs, titles and snippets the provider returned for each exact query
  • Not measured: private submissions, institutional archives, paywalled journals or unindexed pages
  • Not measured: which text came first, who wrote it, or whether a quotation is permitted
  • Not measured reliably: paraphrase, translation, shared ideas or wording changed throughout
What it catches

Where it holds up, and where it slips

Distinctive verbatim passages on indexed public pages are the strongest case for this method. Exact quoted search removes much of the topic-level noise that would appear if ordinary keyword results were treated as copying.

Search coverage is never complete. Indexes omit private repositories, some paywalled pages, recently published text and pages blocked from crawling. Exact phrases also miss real paraphrase. A zero result means the provider returned no page for these queries; it does not mean the writing has been certified as original.

What an online check can miss

  • Paraphrase when enough wording has changed to break an exact phrase
  • Translated passages, which no longer share the searched wording
  • Ideas, structure, data or arguments reused without the same sentences
  • Private submissions, subscription databases and pages excluded from the provider's index
  • Text inside images and scanned pages without a readable text layer
  • Very short copied phrases that are too common for a useful exact search

A flagged passage may be a quotation, reference entry, shared prompt, legal formula or standard language in a technical field. Those are genuine search matches but are not automatically plagiarism, which is why the output keeps every source link available for inspection.

A missed passage may come from a private or unindexed source, may have been translated, or may be paraphrased too heavily for an exact query. Search outages and exhausted quota are handled as errors; they are never converted into an invented unique result.

Read the highlighted passages, not the percentage. Where the matches fall, how long they run, and whether they are attributed will tell you everything the number cannot.

What it will not do

  • Search institutional submission archives, private drives or every subscription journal
  • Catch paraphrase, translation or ideas reliably when the wording has changed
  • Say which document came first, or identify who wrote either version
  • Read text inside images or scanned pages that have no text layer
  • Decide whether a match is a valid quotation, a citation problem or plagiarism
  • Stand alone in an academic, employment or disciplinary decision

Where this fits alongside an institutional checker

This checker is useful for finding likely public sources and opening them immediately. A school or publisher may also have licensed journals and a private archive of earlier submissions. Those collections can find material that a public search index cannot, so the two results can differ without either one being fabricated.

What you get

A result you can inspect, not just a percentage

The useful part of an online plagiarism checker result is the evidence behind it. Original or AI shows the searched passages, returned pages, per-source coverage and highlighted document together, with provider failures kept separate from originality findings.

Matches shown in place

The document view marks each query-sized passage that returned at least one public page, so the percentage can be traced back to visible text.

Source links and context

Returned page titles, addresses and search snippets are listed by coverage. Open the original page to decide whether the wording is quoted, licensed, common or misused.

Useful input formats

Paste text or choose PDF, Word .docx, RTF, Markdown and plain text. File extraction happens in the browser, and a document of any length is split into parts for searching.

Long documents in one report

A document past the size one request can carry is split, searched in parts and merged back into a single result, with every source counted once.

No submissions archive

Original or AI does not add checked documents to a private corpus or compare future visitors against them. Extracted passages do travel to the search service for this requested check.

Saved on your device

Only a compact result summary is kept in your browser history. The document text and source snippets are not written into that saved result.

Formats, limits and requirements

Input
Pasted text, or one PDF, Word (.docx), Markdown, RTF or plain text file up to 20 MB
Text limit
No limit; 20 words is the minimum for a meaningful phrase
Document input
One document at a time
Search passages
Normally about 26 words; never fewer than 8
Search coverage
Indexed public web pages returned for exact quoted passage queries
Processing boundary
File extraction in browser; passage search on the server and configured provider
Requirements
A current browser with JavaScript and a working network connection
Privacy

What stays local, and what is searched online

The privacy boundary is different from the media detectors. A chosen file is opened and converted to text in your browser, so its binary contents are not uploaded. When you press the check button, the extracted text is divided into passages and sent to our Cloudflare Pages Function, which forwards exact queries to the configured search provider. Original or AI does not add the document to a submissions archive, but this is network processing and should not be described as fully local.

Read locally The browser extracts readable text from the chosen file.
Searched online Exact passages go through our endpoint to the search provider.
Not archived by us Saved history contains the score and source count, not document text.

Read the privacy policy

By eye

What to look at before you even run a check

A source search is useful, but copied work also has changes a reader notices before any tool runs. Those changes can guide the source links and highlighted passages you inspect most carefully.

  1. A register that shifts between paragraphs, and shifts back
  2. Vocabulary well above or below the rest of the document
  3. British and American spelling alternating within a section
  4. References cited in the text that are missing from the list
  5. Formatting that changes mid-document: fonts, spacing, quotation marks
  6. A sentence that answers a question the document never asked
  7. Figures and examples that do not match the stated topic or year
  8. Paragraphs that read fluently but contradict something said earlier
  9. Sudden precision: exact numbers in a document that otherwise generalises
  10. A conclusion that does not follow from the argument above it

Every one of these has an innocent explanation, most often a document written over several weeks. Use them to guide review, not to reach a conclusion about the writer.

Search the public web
Side by side

Public search and institutional checking side by side

A public-web checker and an institutional system search different collections. The first can show open source links immediately; the second may also cover licensed journals and a private submission archive.

Capability Original or AI Institutional checker
File parsing happens in the browser Yes Extracted text is searched online Partly Depends on the service
No Original or AI submissions archive Yes Text is processed for the requested search Partly Institutional archives may retain submissions
Clear online word limit Yes No word limit No Usually a few hundred words
Matched passages shown in place Yes Highlighted in the text Partly A percentage and links
Coverage attributed per source Yes Partly
Counts a shared passage once Yes Cannot exceed 100% No Shares often sum past 100%
Provider failures shown as errors Yes Never converted into a unique score Partly Behaviour varies
Reads PDF and Word before transmission Yes Partly Often parsed on the server
Works without an account Yes Partly Often after a sign-up wall
Searches indexed public web pages Yes Exact quoted passage queries Partly Coverage depends on provider
Searches journal and archive databases No Institutional services only Yes Available to licensed institutional services
Catches paraphrase No Word matching cannot No Sometimes implied

The institutional column describes common licensed systems, not every product. Database coverage, retention and result logic vary, so consult the policy of the specific service used by your school, publisher or employer.

Check a document online
Questions

The things people ask before trusting it

Using it

Does this plagiarism checker search the internet?
Yes. After a file is converted to text in your browser, the checker divides that text into exact passages and sends them through our server endpoint to a configured public-web search provider. It does not ask for a second source document. Coverage is limited to pages present in that provider's index.
Is it free, and is there a word limit?
Yes. It is free and there is no word limit; twenty words is the floor, because a shorter phrase matches almost anything. A long document is split into parts behind the scenes and the results are merged into one report, so a dissertation reads as a single result rather than a dozen you have to add up yourself.
Do my documents get stored anywhere?
No. Original or AI does not add document text to a submissions database. A file is parsed locally, but extracted passages are transmitted for the requested public-web search. The compact browser history record contains the score, band and source count rather than the document. Search providers and hosting logs operate under their own policies.
What file types can it read?
PDF, Word .docx, Markdown, RTF and plain text files up to 20 MB are parsed in the browser. Old binary .doc files, Pages and OpenDocument files are refused rather than read badly. A scanned PDF contains pictures of words, so it needs OCR before this checker can search it.
Can I check several files at once?
No. The online checker takes one document at a time. This keeps source attribution and search failures attached to the correct text and makes provider usage predictable. For a long paper, check sections separately and inspect the returned links rather than averaging the percentages.

What the result means

What does the percentage actually measure?
The share of checked words that belong to a passage for which the exact web query returned at least one indexed source. The score is weighted by words and each passage counts only once even when several pages return it. It measures search overlap, not intent or academic misconduct.
Is a high percentage proof of plagiarism?
No. Correctly attributed quotations, a standard assignment prompt and copied work can all produce a high public-web similarity percentage. Open the source, compare its date and context, and check how the wording is cited before deciding what the match means.
Can it tell me which document came first?
No. A result says the search provider returned a page for the submitted wording. It does not establish publication order or prove that the checked author copied that page. Dates, archives, version history, drafts and citations are needed to establish direction.
How are the search passages chosen?
The splitter aims for about twenty-six words and prefers a nearby sentence ending. It can shorten a passage to fit the provider's query limits, but never below eight words. A short final tail is rebalanced so every submitted word remains represented in the weighted score.
Why might several links show the same passage?
Public wording is often syndicated, quoted or copied across many sites. The overall percentage counts that passage once, while the source section lists the returned pages separately. Publication dates and page quality help identify the likely original; result order by itself does not.

Limits and next steps

Will it catch paraphrased text?
No, not reliably. The checker deliberately uses exact quoted passages to avoid treating pages on the same topic as copied. Rewriting throughout, translation and idea-level reuse may therefore return no exact source even when attribution should be reviewed. Use the result to investigate wording, not to clear paraphrased reuse automatically.
Should I use this instead of my institution's checker?
No. Use it as a public-source review, not as a replacement. Institutional systems may search licensed journals and earlier submissions that a general web index cannot see. They may also use different phrase, fingerprint and paraphrase methods, so their percentages are not directly interchangeable with this one.
Can I check my own work before submitting it?
Yes, if you are comfortable sending extracted passages for public-web search. Original or AI does not build a submissions archive from checked text, but the check is not fully offline. Do not submit confidential, embargoed or legally restricted writing unless its handling rules permit this network processing.
Does it detect AI-written text?
No, that is a different measurement and a different tool. Generated text is usually original in the sense this checker means: it matches no existing source. The AI content detector on this site reads how writing is constructed instead, and running both is the sensible thing to do.
Are results saved anywhere?
Yes. A compact summary is saved in this browser so the results list can show the file label, percentage, band and source count. The document text, search snippets and full passage report are not put into that history record. Clearing site data removes the saved summary.

Search it before you send it

Paste one document and review exact passages against indexed public web pages. Source links, passage results and the highlighted document stay visible together.

Check for plagiarism

No sign-up. No word limit. No Original or AI submissions archive.

Also worth reading