PlagiarismCheck.org AI detector (TraceGPT)

The free tier is small and unquantified for AI. The FAQ says "you can check a short text to see how it works for free", and the about page offers "Check 2 pages for free" for the plagiarism service (read 9 November 2026). No free AI word allowance or period is stated, so the card records none and PlagiarismCheck sits in the second table of the free tiers ranking.

The pricing page has four tabs and renders its cards by script from plan data embedded in the page; the organisation tab shows two quote-only tiers in server HTML, and the AI Checker figures below are the embedded data for that tab, in US dollars, read 9 November 2026. The unit is AI pages at 275 words a page.

The developer page prices nothing: API tokens are issued to registered accounts by support, so the card records no API price per 1,000 words and PlagiarismCheck is tabled apart on the API pricing ranking.

Retention is where the product page and the policy disagree. The AI detector page says "by default, we don't retain submitted documents, and educators can control their repository within integration" (read 9 November 2026). The privacy policy, in its version dated 19 November 2025, sets a table under "How long we keep your data": for individual users, "Documents Checked: Until you delete them, or 1 year of inactivity"; for institutional users, student and teacher data "Per institution's instructions"; and after deletion "Backup copies are deleted within 90 days" (plagiarismcheck.org/privacy-policy, read 9 November 2026). The same policy says: "We may collect and store the files you upload, download, or access with the Website." The card records no period: the product page's statement that submissions are not retained by default and the policy's one-year clause govern different documents and conflict, the reading the index gave Sensity and Pindrop, so both sentences are quoted and the data retention ranking lists PlagiarismCheck in its second table rather than at 365 days (ruling of 10 November 2026); the data deletion question the deletion clause.

The headline is a sentence-level figure. The AI detector page reads "99% (accuracy)" in its stat strip and then "99,23%, to be precise. That's the percentage of sentences correctly classified as AI across the seven most recent AI models we tested" (read 9 November 2026). The benchmark post behind it reports 33,699 sentences in 709 documents from seven models, 99.23% classified as AI "measured per sentence", with per-model recall from 98.31% (Gemini Flash 3.5) to 99.87% (Gemini 3.1 Pro), generated and tested in July 2026 (benchmark post, read 9 November 2026). The false positive claim has two levels: "Out of 2.805 human-written documents checked, we wrongly flagged 1.04% of sentences in them. However, at the document level, this error rate drops to zero." The post gives the base, 280,781 human sentences from the British Academic Written English corpus of 2004 to 2007 plus an internal corpus, and the threshold table: 5.45% of human documents flagged at a 5% threshold, 0.64% at 20% and 0.00% at more than 50%.

The post's own qualifications are unusually direct. "Most detectors on the market claim 98-99.98% (accuracy). However, those numbers, including ours, are marketing figures, based on the vendors' own tests." "All published figures are internal, not verified by independent study, yet." And on paraphrase: "All the figures above apply to unedited AI output. We have not yet published the data for text run through tools designed to disguise AI use", which sits beside the product page's line that the tool "can recognize machine-generated content even when attempts are made to humanize AI text" and the FAQ's "The checkers are able to catch some of it, but the detection accuracy drops" (read 9 November 2026). The card quotes all three. The page's standing caveat is "Compass, not a verdict": "No AI detector is perfect; neither is ours."