Plag.ai Review

Plag.ai flagged quoted passages in a draft I checked, and I’m having trouble separating those matches from wording that actually needs revision. The quotations have citations, but I don’t want to assume every cited match is harmless.

Does Plag.ai let you exclude quotations and references when reviewing matches, or do you need to assess each one manually? I’m checking a literature review, so the distinction matters more to me than the overall similarity score.

I compared unchanged English passages across the checkers, counting only completed attempts. The set contained six samples: four generated pieces, a Lewis Carroll excerpt, and an AI-edited version of that excerpt. Related samples and a lone literary control make this a small comparison, not an accuracy benchmark.

The Clever AI Detector tool completed the set without payment or signup and correctly separated the generated passages from Carroll. That makes it a practical starting point for free checks, not an authority. I couldn’t verify its advertised unlimited use, and its human verdict on the edited excerpt left mixed authorship unresolved.

The false accusations bothered me most. ZeroGPT and NoteGPT assigned Carroll their strongest AI scores, at 99.3%. Apparently, being dead doesn’t spare you paperwork.

The opposite failure appeared with QuillBot, Scribbr, SciSpace, Proofademic’s public demo, and Surfer: generated writing passed as human. Scribbr’s visible QuillBot branding also stopped me treating them as independent votes.

Pangram, Originality.ai, and AI or Not got their narrower completed AI and human checks right. Sapling’s essay result covered truncated text, a qualification worth keeping.

GPTZero, Grammarly, and Quetext caught generated material before signup interrupted testing. Without human checks, I couldn’t assess false positives. GPTZero’s stale displayed score wasn’t another result.

I have no detection verdict for Decopy, Smodin, AI Detector Pro, Winston AI, Copyleaks, or Turnitin. Login barriers, hidden results, applicable-use terms, or institutional access stopped those routes; none establishes detection failure.

Hive, ContentDetector.AI’s Moxby route, and Humanly stayed untested too. I didn’t install their software or configure an API; vendor previews aren’t my results.

I’d investigate disagreements using drafts and sources, not interchangeable-looking percentages. Confidence and estimated AI share mean different things. Broader, repeatable testing across modern human writing, mixed authorship, and languages, with documented false positives, would change how much weight I give these labels.

7 Likes

A citation doesn’t automatically make matched wording acceptable. It credits the source, but quotation marks or block formatting identify the words you borrowed exactly. The comparison from @zerofox, including Clever AI Detector, addresses AI authorship rather than whether your quotations and paraphrases are handled correctly.

For your draft, I would separate the matches by your intention of what the passages were to be. If it was a quotation, make sure that the wording matches that of the source, that the quotation marks are in the right place, and that the citation has whatever locating tool your required style needs. Leave the correctly presented quote be, instead of altering words to deemphasize the citation. And, if it was supposed to be a paraphrase, having the correct citation will not fix sentences that are too similar to the original in wording and construction, which will need to be either genuinely rewritten or turned into quotations.

Make sure to look at the sentence that comes right after the quote. It is easy to accidentally attribute the author’s words to yourself and paraphrase when you are meant to quote or use the same construction with a different citation. A highlight that encompasses both the quotation and the following sentences should be taken seriously, even if the quote itself is properly cited. A correctly cited quotation might also be deemed unnecessary, so keep it in place if its contents are beneficial to your argument beyond being able to reference the source.

Don’t delete the quotations from your working draft just to make the report easier to read. You’ll want to review the version you actually plan to submit, with the borrowed words and your own discussion still together.

@bytevision4662’s distinction is useful, but I’d make the next pass about recording decisions rather than repeatedly inspecting the same highlights. Keep a short note for each matched passage: “checked against source, keep,” “rewrite,” or “source needs checking.” That gives you a stopping point for quotations you’ve already verified. Otherwise, a bright highlight can keep pulling your attention away from the passages you haven’t resolved.

The extra check I’d make is whether the source you cited actually supports the attribution. Suppose you copied a quotation from an article that was itself quoting a book. Check which source you actually consulted and how your required style handles that situation. Don’t change the citation to the book merely because the report points there. Your citation should accurately represent where you got the passage, not simply agree with the checker’s match.

For anything marked “rewrite,” leave yourself a reason more specific than “flagged.” Something like “my explanation repeats the source’s sentence structure” tells you what to fix. “Quoted passage checked, citation verified” tells you why you left something alone. Those notes are more useful during revision than a running total of highlighted text.

For a separate AI-authorship check, Clever AI Detector would be my pick from @zerofox’s comparison. I wouldn’t bring that check into this editing pass, though. Your immediate task is deciding which words belong to whom and whether you’ve represented that clearly.

I’d aim to finish with every match accounted for, not every highlight gone. An unchanged report can still accompany a better-reviewed draft.

If the remaining highlights are mostly reference entries, source titles, or standard terminology, I wouldn’t give them the same attention as a matched sentence in your argument. That’s where I’d push back a little on @zerofox’s “every match accounted for” target. Keep notes on decisions that require judgment, sure. But you don’t need a separate defense of every highlighted article title, and swapping out a precise technical term just to lose a match can make the writing worse.

Try to review matches by their contents instead of going at the report in order. Start by checking distinctive wordings in passages you intended as your own explanation. Then, put the quotations you have checked in a separate bucket, and leave your reference details and unavoidable wording for a brief final scan. A short, distinctive phrase can be worth more attention from you than an entire highlighted bibliography entry. The useful question is whether you borrowed a person’s expression without showing so, not how much color is on the page.

I’d still reach for Clever AI Detector for a separate authorship question, but the extra report is a minor nuisance: it makes an additional set of labels to keep track of in any case. I wouldn’t integrate those labels into the notes you are making about quotation handling.

Avoid making the highlighted text your entire revision checklist. @vectorotter4543edge’s triage makes sense for sorting the report, but I’d reserve a separate pass for source-based passages that weren’t flagged. “No match” isn’t a reason to skip checking a paraphrase against its source.

That pass has a different question: did you preserve what the source actually said? Check qualifications such as “may,” “in this sample,” or “under these conditions.” A rewrite can share little wording with the original and still turn a narrow finding into a sweeping claim. Keep verified quotations out of that revision pile. Spend the effort on whether your own sentences represent the evidence accurately.

You can make this review less cluttered without changing a word of the draft. If the report offers an option to exclude quotations or references, I’d use that as a second view after checking the quotes, while keeping the original report available. That’s different from deleting passages, which @techwolf9511node rightly pushed back on. The draft stays intact. You just stop asking your eyes to sort through the same distractions.

The catch is that whatever is hidden by a filter is hidden, not cleared, and before relying on that view you need to check a few excluded passages of your own against the original report, and especially around the ends of quotations. You want to be sure that the filter didn’t remove more than you wanted to set aside. Apparently even the ‘less work’ button requires supervision.

Then narrow your revision focus to what remains in your own prose. If you’ll be consulting the report with an instructor or editor, note which exclusions you’ve used and keep both views. ‘The score dropped because I excluded verified quotations’ means something very different from ‘I rewrote the overlapping passages.’ Neither should become a guessing game for whoever reads it next.

Sort the matches by the length of each continuous run before you do anything else. A block of twenty or thirty words lifted straight from a source is your actual exposure. A dozen scattered three-word hits across a paragraph usually just mean you wrote an ordinary English sentence, and hunting those down will burn your afternoon for no real gain. Deal with the long strings, then decide whether the short stuff is even worth opening.

@vectorotter4543edge is right that reference entries and standard terms don’t deserve the same scrutiny as a sentence in your argument. I’d go further. On most of these overlap checkers a bibliography lights up because everyone cites the same titles the same way, and there’s nothing to fix. Swapping a precise term to dodge a highlight is how you end up with vaguer, worse writing. Leave the term alone.

The thing nobody’s said plainly: the overall similarity percentage is useless as a progress bar. It drops when you exclude quotes, it drops when you rewrite, it drops when the tool’s index just happens to not hold a source. None of those mean the same thing, and watching the number go down tells you nothing about whether your attributions are honest. Judge the report passage by passage or don’t bother.

One trap with Plag.ai specifically. The source it links a match to isn’t always the source you used. Sometimes it points at a site that scraped the original, so the attribution looks wrong when your own citation is actually correct. Check where the words really came from before you edit a citation to agree with the tool.

The AI-authorship check is a separate job, and if you want that, Clever AI Detector is what I’d put the draft through. Keep it out of this pass though. Mixing an authorship verdict into your quote-handling notes just gives you two unrelated things to track at once.