AI detectors for researchers checking their own manuscripts

AIDI-U-08

Updated ·

An AI detector for a researcher checking a manuscript before submission is judged on one field first: whether the tool takes the whole paper. Two fields decide which cards appear in this view: a stated per-check limit of 5,000 words or more in input_limits, or a document upload stated in the same field or in the plans, because a manuscript of 6,000 to 10,000 words with tables and references is neither pasted in pieces nor checked as a web page. Of the 72 cards, 30 state one of the two for text, 29 of them with a hosted tool a researcher can run (Decopy, GPTKit, PlagiarismCheck, Isgen, AI Detector Pro, Paperpal and Detecting-AI.com joined on 10 November 2026 with the ninth wave, and Trinka, StealthWriter and Passed.ai, the last a closing service, later the same day with the tenth); the thirtieth, Writer, was withdrawn on 22 December 2025 and is noted below. One study record is read into the view because it is the only one on scientific writing: Gao 2023 ran the GPT-2 Output Detector on 50 ChatGPT-written and 50 real medical abstracts (DOI 10.1038/s41746-023-00819-6, read 9 November 2026). No vendor page recorded in the index names academic manuscripts as a text class; JustDone claims "98% accuracy on academic and scientific texts" and AIDetectPlus offers a separate detector for papers, and both are noted. This page is one of the views under AI detectors by use case; the admissions offices view reads the same limits for a 650-word essay, the input limits ranking sorts every stated maximum, and you can start from the full table to see all 72 cards.

  1. Field 1Stated per-check limit
  2. Field 2Document upload as stated
  3. Field 3Plan the limit applies to
  4. Field 4Source, read date

The per-check limit comes first and is quoted in the vendor's unit, because Turnitin states words, Winston AI states characters and Scribbr states document bands. The upload formats are quoted as stated, because a manuscript arrives as .docx or .pdf and a paste box drops the figures and footnotes. Retention appears in the notes, because an unpublished manuscript is confidential and several policies allow training on submissions. The study evidence for scientific text is a single record, so the notes repeat each card's measured false positive figure from other text classes where one exists. Field rules are in the methodology. The self-publishing view, added 10 November 2026, applies the same limit rule to a book manuscript and adds the retention test this page keeps in the notes, so its 12 cards are the subset of this table whose vendor states what happens to the text.

Which fields decide fit for a manuscript check?

AI detectors for researchers checking their own manuscripts: the cards in this view and the fields that decide fit
DetectorStated per-check limitDocument upload as statedPlan the limit applies toSource, read date
Winston AIUp to 200,000 characters per scan.docx, .png and .jpgEssential, Advanced, Elitegowinston.ai/pricing, 8 November 2026
PangramWords per text check not publishedPDF, DOCX, RTF; up to 100 files at a timeAll planspangram.com, 8 November 2026
GPTZeroUp to 150,000 characters per textBatch of up to 250 files; file types not publishedFree accountgptzero.me/faq, 8 November 2026
CopyleaksMaximum 2,000 pages, no character maximum (web platform)Documents through the platform; OCR on gif, png, bmp, jpg, jpegCredits, 1 credit per 250 wordscopyleaks.com, 8 November 2026
Phrasly"no character limits on detection" (web tool); 15,000 characters per API requestUpload a file; file types not publishedFree web toolphrasly.ai/ai-detector, 8 November 2026
SaplingUp to 100,000 characters per checkPDF or Word document with selectable textPro and Enterprisesapling.ai/ai-content-detector, 8 November 2026
JustDoneUp to 15,000 words in one scanPDF, Word or TXTPremium (free scan without an account)justdone.com/ai-detector, 8 November 2026
Originality.aiMaximum words per scan not published; 1 credit per 100 wordsdocx, pdf, doc, or a URLAll plansoriginality.ai/pricing, 8 November 2026
ZeroGPT15,000 characters free; 100,000 PRO and PLUS; 150,000 MAX; 250,000 EDU; 350,000 EXPERT.txt, .docx, text-based .pdfBy planzerogpt.com/pricing, 8 November 2026
Proofademic3,500 words (Essential), 8,000 (Premium), 25,000 (Professional) per request.docx, .txt, PDF; Batch ScanPaid plansproofademic.ai/ai-text-detector, 8 November 2026
Turnitin300 to 30,000 words of prose; file under 100 MB.docx, .pdf, .txt, .rtfInstitutional licence through Turnitin Originalityguides.turnitin.com, 8 November 2026
AIDetectPlusNo per-document maximum published; 1 credit per word.docx, .pdf, .txt, .html; separate detectors for essays, papers, resumes and PDFsSigned-in detectorsaidetectplus.com, 9 November 2026
Undetectable AIUp to 10,000 words per check (web); under 30,000 words per API requestPaste or upload a fileFree web toolhelp.undetectable.ai, 9 November 2026
Smodin50,000 characters per text or document.txt, .doc, .docx, .pdfFree web toolsmodin.io, 9 November 2026
CompilatioNo document size limit published; Studium meters 1 credit per 250 wordsDocuments uploaded to Magister+ or StudiumMagister+ (institutions), Studium (students)compilatio.net/studium, 9 November 2026
Quetext1,000 words per free scan; 100,000 or 200,000 words a month on paid plans.doc, .docx, .pdf, .txt in bulk uploads of 20 or 100 filesEssential and Professionalquetext.com/integrations, 9 November 2026
TruthScanUnder 30,000 words per API requestWord, PDF and Google Docs document processing for enterprise usersAPI and enterpriseTruthScan API documentation, 9 November 2026
ScribbrPremium check bands: up to 7,499 words, 7,500 to 49,999, 50,000+Document check (per-check pricing by document length)Premium check, $19.95 to $39.95scribbr.com/plagiarism-checker, 8 November 2026
QuillBotNo word limit for detection scans (Premium); 1,200 words freeUpload a document; file types not published; Premium bulk uploadPremiumquillbot.com, 8 November 2026
DecopyUp to 200,000 characters per check.txt, .pptx, .ppt, .doc, .docx, .pdfWeb tool, free basic detection without an accountdecopy.ai/ai-detector, 9 November 2026
GPTKit2048 characters as a guest; 10,240 to 153,600 characters per request on paid plans.html, .txt, .doc, .docx, .pdf, .odt, .rtf and othersUnlimited (Personal, Team, Institution)gptkit.ai, 9 November 2026
PlagiarismCheckUp to 97 pages per check, 275 words a pagedoc, docx, pdf, txt, odt, rtf through the APIEvery planplagiarismcheck.org/ai-detector, 9 November 2026
Isgen60,000 characters per submission (Academic); unlimited (Ultimate)pdf, docx, doc, 10 MB eachAcademic, Ultimateisgen.ai/pricing, 9 November 2026
AI Detector Pro40,000 characters per paid scan (about 6,000 to 7,000 words per the FAQ)Documents uploaded in the web app; no format list publishedBasic, Pro, Unlimitedaidetector.pro/pages/faq, 10 November 2026
Paperpal1,200 words per free scan; Prime 25,000 words a month, Pro unlimited.docx up to 25 MBFree, Prime, Propaperpal.com/tools/ai-detector, 10 November 2026
Detecting-AI.com15,000 characters per free check; 1,000,000 characters per scan on paid plansTXT, PDF, DOCX, read in the browserWriter, Pro, Teamdetecting-ai.com, 10 November 2026
TrinkaNo per-check ceiling published; the AI Detector Report costs 1 credit per 5,000 words; Premium Plus and Confidential Data include 240 and 1,000 pages a year"Upload your MS Word or PDF file to check for AI content"; the marketplace accepts Word and PDF up to 100 MB and 25,000 wordsEvery plan (Basic: 4 free credits every 30 days)trinka.ai/features/ai-content-detector, 10 November 2026
StealthWriter5,000 words per input on every paid plan; 1,000 on Free; 500 per scan on the no-account demoPaste only; no upload formatStarter and abovestealthwriter.ai/pricing, 10 November 2026
Passed.aiUp to 5,000 words per Academic Honesty Report; no per-check limit for the in-app tool"drag and drop your files into our app"; no format namedIndividual subscription and pay-as-you-go reports, no longer soldweb.archive.org, pricing, 17 April 2026, 10 November 2026

Values as each vendor states them, read 8 and 9 November 2026. Writer's detector, which took up to 5,000 words per check, was sunset on 22 December 2025 and is not listed (archived writer.com page, read 9 November 2026). Walter Writes (750 to 2,000 words per request, no upload stated), Hive (text in 2,048-character chunks, no upload), GLTR and the GPT-2 Output Detector (pasted text only) state neither field and are absent; the GPT-2 Output Detector enters the evidence section through Gao 2023.

AI detectors for researchers checking their own manuscripts: register plate with the page title and its register ID
The view plate: register ID and title, drawn from the index data.

Why each card is in this view

  1. AIDI-D-0005Winston AI

    200,000 characters per scan and .docx upload on paid plans (gowinston.ai/pricing, read 8 November 2026); 0 false accusations in 18 human documents in Weber-Wulff 2023 and 0.55% at threshold 0.5 in RAID 2024. The policy stores submitted text until the report or account is deleted.

  2. AIDI-D-0006Pangram

    PDF, DOCX and RTF upload with up to 100 files at a time (pangram.com, read 8 November 2026); 0.0000 of 1,992 human passages flagged in Jabarian and Imas 2025 and 0 of 40 human papers in Van Vlasselaer 2026, which also ran it alone on 1,163 theses and saw it flag 45.5%. Detection is trained on a dataset that excludes user submissions.

  3. AIDI-D-0001GPTZero

    150,000 characters per text and a batch of up to 250 files with a free account (gptzero.me/faq, read 8 November 2026); false positive results from 0.00% at threshold 0.5 in RAID 2024 to 9 of 18 in Weber-Wulff 2023. Dashboard inputs are stored and used in aggregate; API inputs are not stored.

  4. AIDI-D-0003Copyleaks

    up to 2,000 pages per document with no character maximum (copyleaks.com, read 8 November 2026); 0 of 42 human papers in Walters 2023, 5 of 10 in Perkins 2024. The policy allows training on data unless a direct customer opts out.

  5. AIDI-D-0012Phrasly

    no character limit on the web detector and file upload (phrasly.ai/ai-detector, read 8 November 2026); no study record. The detector page says text is deleted immediately after analysis.

  6. AIDI-D-0010Sapling

    100,000 characters per check and PDF or Word upload on Pro (sapling.ai/ai-content-detector, read 8 November 2026); 0 of 42 human papers in Walters 2023 with 31% uncertain. Content is stored to improve the service; the API offers zero retention.

  7. AIDI-D-0013JustDone

    15,000 words per scan and PDF, Word or TXT upload (justdone.com/ai-detector, read 8 November 2026); the vendor claims "98% accuracy on academic and scientific texts", the one card that names scientific text, with no study record behind it (JustDone legit question).

  8. AIDI-D-0004Originality.ai

    docx, pdf and doc upload with no stated word maximum (originality.ai/pricing, read 8 November 2026); 0 of 42 human papers in Walters 2023, 0.25% at threshold 0.5 in RAID 2024, 7 of 114 in Orenstrakh 2023. Scan history may train the models unless the user opts out.

  9. AIDI-D-0011ZeroGPT

    15,000 to 350,000 characters per detection by plan with .txt, .docx and text-based .pdf upload (zerogpt.com/pricing, read 8 November 2026); 0 of 18 in Weber-Wulff 2023, 2% of 42 in Walters 2023, 1.42% at threshold 0.5 in RAID 2024. Content is kept only as long as needed to operate the service.

  10. AIDI-D-0014Proofademic

    3,500 to 25,000 words per request by plan and .docx, .txt and PDF upload with Batch Scan (proofademic.ai/ai-text-detector, read 8 November 2026); no study record.

  11. AIDI-D-0002Turnitin

    300 to 30,000 words with .docx, .pdf, .txt and .rtf upload, through an institutional licence only (guides.turnitin.com, read 8 November 2026); 0 of 42 in Walters 2023, 0 of 18 in Weber-Wulff 2023, 0 of 40 in Van Vlasselaer 2026. A researcher without an institutional account cannot run it.

  12. AIDI-D-0018AIDetectPlus

    .docx, .pdf, .txt and .html upload with a separate detector for papers and no per-document maximum (aidetectplus.com, read 9 November 2026); no study record.

  13. AIDI-D-0024Undetectable AI

    10,000 words per web check and under 30,000 words per API request, with file upload (help.undetectable.ai, read 9 November 2026); no study record on the index at the read date.

  14. AIDI-D-0029Smodin

    50,000 characters per text or document with .txt, .doc, .docx and .pdf upload (smodin.io, read 9 November 2026); no study record, and the terms grant the vendor a perpetual licence over user content.

  15. AIDI-D-0032Compilatio

    documents uploaded to Magister+ by institutions or Studium by students, metered at 1 credit per 250 words (compilatio.net/studium, read 9 November 2026); 0 of 18 in Weber-Wulff 2023.

  16. AIDI-D-0033Quetext

    bulk uploads of 20 or 100 files in .doc, .docx, .pdf and .txt on paid plans, 1,000 words per free scan (quetext.com/integrations, read 9 November 2026); no study record.

  17. AIDI-D-0034TruthScan

    under 30,000 words per API request and Word, PDF and Google Docs document processing for enterprise users (TruthScan API documentation, read 9 November 2026); no study record; content retained 30 days by default.

  18. AIDI-D-0009Scribbr

    premium checks priced by document band up to 50,000+ words, with the free detector stating no limit (scribbr.com/plagiarism-checker, read 8 November 2026); 2% of 42 human papers in Walters 2023. "we do not store or share your data".

  19. AIDI-D-0007QuillBot

    no word limit on Premium detection scans and document upload, 1,200 words free (quillbot.com, read 8 November 2026); no study record; no retention clause on the detector page.

  20. AIDI-D-0043Decopy

    200,000 characters and six upload formats without an account (decopy.ai/ai-detector, read 9 November 2026); the FAQ says content is not permanently saved and history can be deleted; no study record.

  21. AIDI-D-0045GPTKit

    document upload in many formats and up to 153,600 characters on the Institution plan (gptkit.ai, read 9 November 2026); content removed from the server after detection per the FAQ; 0 of 114 false positives in Orenstrakh 2023.

  22. AIDI-D-0046PlagiarismCheck

    97 pages of 275 words per check and six API formats (plagiarismcheck.org/ai-detector, read 9 November 2026); checked documents are kept until deleted or a year of inactivity for individual users; 0 of 18 in Weber-Wulff 2023.

  23. AIDI-D-0047Isgen

    60,000 characters per submission on Academic and unlimited on Ultimate, with pdf, docx and doc uploads (isgen.ai/pricing, read 9 November 2026); the home page FAQ says no data is stored while the pricing page states Document History periods; no study record.

  24. AIDI-D-0048AI Detector Pro

    40,000 characters per paid scan (aidetector.pro/pages/faq, read 10 November 2026); uploaded text is deleted at the beginning of the fourth month after entry per the privacy policy; no study record.

  25. AIDI-D-0050Paperpal

    a .docx upload up to 25 MB and a detector page that states no storage of drafts, with ISO 27001 and ISO 42001 certifications on the data security page (paperpal.com/tools/ai-detector, paperpal.com/data-security, read 10 November 2026); the vendor describes the training set as scholarly samples; no study record.

  26. AIDI-D-0051Detecting-AI.com

    1,000,000 characters per paid scan with TXT, PDF and DOCX read in the browser before the text is sent (detecting-ai.com, read 10 November 2026); anonymous submissions are not added to account history per the privacy policy; no study record. Content Detector AI's archived page stated no limit but the address redirects, and Fast-DetectGPT's demo publishes no figure, so neither joins.

  27. AIDI-D-0055Trinka

    a Word or PDF upload with a PDF report back, metered at 1 credit per 5,000 words, from a vendor built for academic manuscripts; the trust centre states a 90-day deletion default and the privacy policy reserves the right to save documents (trinka.ai/trust-center, read 10 November 2026); no study record, and the vendor's 99.9% rests on a self-submitted leaderboard entry.

  28. AIDI-D-0056StealthWriter

    5,000 words per input on paid plans meets the limit rule, by paste only; the privacy policy states submitted text is not stored (stealthwriter.ai/privacy, read 10 November 2026); the detector is the second tab of a humanizer and no study tested it.

  29. AIDI-D-0060Passed.ai

    file upload and 5,000-word reports meet the rule on the archived pricing page, but the service is closing and registration is closed, so the row is a record for existing customers (passed.ai, read 10 November 2026); no study record; the privacy policy disposes of submitted content "typically 90 to 120 days" after processing.

What the evidence says for scientific writing

One record in the index scored scientific text. Gao 2023 asked ChatGPT to write 50 abstracts in the style of five high-impact medical journals from real titles, and ran them and the 50 originals through the GPT-2 Output Detector: an AUROC of 0.94, with a median "fake" score of 99.98% on the generated abstracts and 0.02% on the originals, and sensitivity of 86% with specificity of 94% at the paper's 1.46% cutoff (DOI 10.1038/s41746-023-00819-6, Results and Figure 2, read 9 November 2026; Gao 2023 record). The GPT-2 Output Detector card is a 2019 RoBERTa model trained on GPT-2 output, run locally from MIT-licensed weights with the hosted demo paused, which is why it does not sit in the table above: it takes pasted text and states no limit. Blinded reviewers in the same study caught 68% of the generated abstracts and wrongly called 14% of the real ones generated, and the lower quartile of ChatGPT's abstracts scored 12.73% "fake", so a quarter of them sat near the human range. The index's other human-rater record points the same way for a different reader: in Fleckenstein 2024, 89 pre-service teachers identified 45.1% of ChatGPT essays and 200 experienced teachers 37.8%, with confidence near 80% in both groups (DOI 10.1016/j.caeai.2024.100209, Tables 1 and 4, read 9 November 2026), so a reviewer's own judgement is not the check a researcher can fall back on either. A researcher who cannot send a manuscript to a vendor's server has the offline question, added 9 November 2026, which lists the fourteen cards that run locally or on-premise, three of them open-source research tools.

The larger caution for a researcher is the false positive record on formal prose. Liang 2023 reported an average false positive rate of 61.22% across seven detectors on 91 essays by non-native English writers, against near-perfect accuracy on 88 US 8th-grade essays (arXiv 2304.02819, read 8 November 2026; Liang 2023 record), and a manuscript written in a second language in the constrained register of a journal article is the profile that study flagged. Turnitin's own research reported 0.014 for ELL writers against 0.013 for native writers on its current model (guides.turnitin.com, read 8 November 2026), a vendor figure. Van Vlasselaer 2026 ran Pangram alone on 1,163 real theses and it flagged 45.5% of them (DOI 10.1007/s40979-026-00226-w, Table 6, read 9 November 2026), a result the authors could not resolve without knowing how the theses were written.

Confidentiality is the third field. Of the 19 cards in the table, Scribbr and Phrasly state that submissions are not stored or are deleted on analysis, Pangram and Winston AI state no training on user submissions, and Originality.ai, Copyleaks and Sapling state that submitted text may be used to improve or train their models unless the user opts out, all as read on 8 November 2026 and set out in the do AI detectors store your text question. An unpublished manuscript pasted into a detector that trains on submissions has been disclosed. No detector score should be the only evidence in a misconduct, hiring or publishing decision, and a journal's AI policy, not a score, is the document a researcher should read first.

A reviewer reading a proposal rather than an author checking a manuscript has the grant reviewers view, added 9 November 2026, which adds the retention field to the same document limits and sets the NIH and NSF rules on uploading proposals to outside AI tools before any card.

Sources

  • https://gowinston.ai/pricing/ (read 8 November 2026)
  • https://www.pangram.com/ (read 8 November 2026)
  • https://gptzero.me/faq (read 8 November 2026)
  • https://copyleaks.com/ai-content-detector (read 8 November 2026)
  • https://phrasly.ai/ai-detector (read 8 November 2026)
  • https://sapling.ai/ai-content-detector (read 8 November 2026)
  • https://justdone.com/ai-detector (read 8 November 2026)
  • https://originality.ai/pricing (read 8 November 2026)
  • https://www.zerogpt.com/pricing (read 8 November 2026)
  • https://proofademic.ai/ai-text-detector/?utm_source=aidetectorindex.com&utm_medium=owned&utm_campaign=for_researchers (read 8 November 2026)
  • https://guides.turnitin.com/hc/en-us/articles/28477544839821 (read 8 November 2026)
  • https://aidetectplus.com/ (read 9 November 2026)
  • https://help.undetectable.ai/en/article/detector-api-1cf74il/ (read 9 November 2026)
  • https://smodin.io/ai-content-detector (read 9 November 2026)
  • https://www.compilatio.net/en/product/studium/ (read 9 November 2026)
  • https://www.quetext.com/integrations (read 9 November 2026)
  • https://truthscan.com/truthscan-ai-text-detection-api-documentation (read 9 November 2026)
  • https://www.scribbr.com/plagiarism-checker/ (read 8 November 2026)
  • https://quillbot.com/ai-content-detector (read 8 November 2026)
  • http://web.archive.org/web/20251209115654/https://writer.com/ai-content-detector/ (read 9 November 2026)
  • https://doi.org/10.1038/s41746-023-00819-6 (read 9 November 2026)
  • https://arxiv.org/pdf/2304.02819v3 (read 8 November 2026)
  • https://doi.org/10.1007/s40979-026-00226-w (read 9 November 2026)

Revision log: 2026-11-09: view created at index version 2026.11.62026-11-09: offline question linked (seventh wave)2026-11-09: eighth-wave pages linked2026-11-10: seven ninth-wave cards added (Decopy, GPTKit, PlagiarismCheck, Isgen, AI Detector Pro, Paperpal, Detecting-AI.com), 26 hosted cards2026-11-10: Trinka, StealthWriter and Passed.ai added (tenth wave), 29 hosted cards; Check For AI's archived file upload left out as a withdrawn tool2026-11-10: self-publishing view linked

FAQ: AI detectors for researchers

Which AI detectors accept a full manuscript?

Nineteen cards with a hosted tool state a per-check limit of 5,000 words or more or a document upload, as read on 8 and 9 November 2026, from Winston AI's 200,000 characters per scan to Scribbr's 50,000-word premium band; Turnitin and Compilatio require an institutional account. Writer, a twentieth, was withdrawn in December 2025.

Has any study tested AI detectors on scientific papers?

One in the index: Gao 2023, where the GPT-2 Output Detector reached an AUROC of 0.94 on 50 ChatGPT-written and 50 real medical abstracts, with 86% sensitivity and 94% specificity at the paper's cutoff (DOI 10.1038/s41746-023-00819-6, read 9 November 2026). No indexed study scored full manuscripts.

Will a detector flag a manuscript by a non-native English writer?

Liang 2023 found an average false positive rate of 61.22% across seven detectors on 91 TOEFL essays (arXiv 2304.02819, read 8 November 2026). The study tested 2023 versions and essays rather than papers; vendors have since published claims of improvement, recorded on each card.

Does any vendor name academic manuscripts?

None of the 72 vendor pages recorded names manuscripts as a text class. JustDone claims "98% accuracy on academic and scientific texts" and AIDetectPlus offers a detector for papers, both without a study record (justdone.com/ai-detector, aidetectplus.com, read 8 and 9 November 2026).

Is a pasted manuscript kept or used for training?

It depends on the card: Scribbr and Phrasly say no storage or immediate deletion; Pangram and Winston AI say no training on submissions; Originality.ai, Copyleaks and Sapling allow training unless the user opts out, all as read on 8 November 2026. The data retention ranking sorts every stated period.

Should a researcher rewrite a paper to lower its score?

The vendors say a score is a signal, not a verdict: QuillBot's page says "if you used AI in your work, disclose it rather than revising to change a score", and Originality.ai says an AI score should not be used alone (quillbot.com, originality.ai, read 8 November 2026). The journal's disclosure policy is the document that governs.

Related reading