A plagiarism checker compares your text with a database of existing material and highlights passages that match, so you can quote and credit them, rewrite them or remove them before publishing. Run one on every AI-assisted draft, because language models can occasionally reproduce text from their training data and writers can paste in source material without noticing. Keep two things separate: a plagiarism check finds copied wording, while an AI detector guesses whether text was machine-written, and independent research has found those guesses unreliable enough that they should never be the only evidence for a decision.

This guide explains how the checks work, where the copying risk in AI writing comes from, how to choose a tool and a workflow that fits a small editorial team. It is tool-neutral; product details are taken from each vendor's own pages and were checked in October 2026.

Plagiarism checks and AI detection are different tests

A plagiarism check (often called a similarity check) looks for overlapping text. If a sentence in your draft matches a sentence on a web page, in a journal or in a document the tool has indexed, it is flagged and the source is shown. The result is evidence you can verify: you can open the source and compare.

An AI detector estimates the probability that text was produced by a language model. There is no source to open, only a score.

Many products now offer both in one interface. Quetext, for example, describes its DeepSearch technology as powering both a plagiarism checker and an AI detector, and colour-codes exact matches separately from near-exact or "fuzzy" matches. Originality.ai markets AI detection, plagiarism checking and fact review to writers, editors and marketers. Grammarly's plagiarism checker pairs similarity checks with suggestions on when and how to acknowledge generative AI use. Read the two results as answers to different questions.

How reliable AI detection is

Peer-reviewed and preprint research gives reasons for caution. A team led by Debora Weber-Wulff (HTW Berlin) tested 12 publicly available detection tools and two commercial systems, Turnitin and PlagiarismCheck, and concluded that the tools were "neither accurate nor reliable", with a bias towards classifying output as human-written. They also found that obfuscation techniques, such as paraphrasing, significantly worsened performance.

A team led by Weixin Liang reported that several widely used detectors consistently misclassified writing by non-native English speakers as AI-generated, while native samples were identified accurately. The authors caution against using such detectors in evaluative or educational settings.

These studies tested particular tools at particular times, and vendors update their models. The practical lesson still holds: use an AI score as a prompt for a conversation or a closer read, not as proof. For a publisher, the more useful questions are whether the text is accurate, whether it copies anyone, and whether it is good.

Where plagiarism risk comes from in AI writing

Three sources account for most problems.

Memorised training text. Research has shown that language models can reproduce parts of their training data verbatim. Nicholas Carlini and colleagues extracted hundreds of verbatim sequences from GPT-2, even when a sequence appeared in only one training document. A follow-up study, Quantifying Memorization Across Neural Language Models, found memorisation grows with model size, with how often an example is duplicated in the training data, and with the amount of context in the prompt. For everyday writing the risk is highest with famous passages, standard definitions and boilerplate that appears many times online.

Source material in the prompt. If you paste a competitor's article and ask for "a version of this", the output may stay close to the original's wording and structure. A plagiarism checker catches the wording; only an editor catches a borrowed structure.

Stitched summaries. Asking a model to combine several sources can produce text that tracks each source closely, sentence by sentence.

Why it matters for publishers in the UK

Copying is a legal question as well as an editorial one. GOV.UK's copyright guidance explains that protection is automatic: there is no need to apply or pay, and there is no register of copyright works in the UK. Text you find online is protected by default, whether or not it carries a notice.

The law on AI and copyright is under active review. The government's consultation on copyright and artificial intelligence ran from 17 December 2024 to 25 February 2025, and a report and impact assessment from the Department for Science, Innovation and Technology were added on 19 March 2026 (checked October 2026). For publishing decisions, take legal advice rather than relying on a checker's score.

Search engines have their own rules. Google's spam policies describe abusive scraping as including copying content from other sites, modifying it only slightly ("for example, by substituting synonyms or using automated techniques") and republishing it. Its helpful content questions ask whether content that draws on other sources avoids "simply copying or rewriting those sources" and adds "substantial additional value and originality". A clean plagiarism score does not answer that second question.

How to check AI content for plagiarism

  1. Finish the edit first. Check the version you intend to publish, including headings, captions and the meta description.
  2. Run the full text. Short excerpts can miss matches that appear only in context.
  3. Review every match. Open the source. Decide whether the overlap is a common phrase (fine), a fact expressed in standard words (usually fine), or distinctive wording (quote it with attribution, rewrite it or cut it).
  4. Check quotations separately. Correctly quoted and attributed text will match its source; that is expected. Confirm the quote is accurate and the link points to the original.
  5. Look beyond wording. If the draft follows one source's structure and argument closely, rewrite from your own outline, even if the scan is clean.
  6. Keep a record. Save the report with the draft. If a question arises later, you can show what was checked and when.

Checkers cannot see everything. Paywalled articles, recent pages and documents outside a tool's index may not be compared. A clean result means "no match found in this database", not "original".

Choosing a plagiarism checker

Prices, limits and features change often, so check the vendor's current pages before you subscribe. Questions to ask:

  • What does it compare against? Open web only, or also academic databases, publisher content and your own previous submissions?
  • How does it show matches? Side-by-side source views and colour-coding for exact and near-exact matches make review faster.
  • Can you exclude sources? Excluding quoted material, your own site or a client's previous work avoids noise.
  • What happens to your text? Some services store submissions in a repository that future checks compare against. For unpublished client work, read the data policy and choose settings accordingly.
  • Does it handle your volume? Bulk upload, an API or integration with your editor matters for teams.
  • What does the AI score claim? If a tool offers AI detection, read how it describes accuracy and false positives, and treat the score in light of the research above.

No single tool suits every team. A freelance writer checking a few posts a month needs something different from an agency running hundreds of pages through a workflow.

A pre-publication workflow for AI-assisted drafts

  • Brief: record the sources you gave the model.
  • Draft: ask the model to work from your outline and your notes, not from a single pasted article.
  • Edit: add your own examples, analysis and structure.
  • Fact-check: Google's guidance on generative AI content says it is "critical to manually factcheck and review all AI-generated content for accuracy and trustworthiness before publishing".
  • Plagiarism check: run the final text, review every match and fix what needs fixing.
  • Sign-off: a named editor approves the piece and the saved report.

Our sister site AI Disraeli UK covers related editorial workflows for British teams. A checker is a safety net for wording. Originality, accuracy and fair credit still depend on the person who signs the piece off.