AI detectors for admissions offices
An AI detector for admissions offices is filtered here on four fields, and 6 of the 72 cards in the index state all four: a data retention statement, a false positive figure from a study indexed here rather than from the vendor, an input limit that fits a 650-word essay in one check, and an institutional or team plan. The data retention ranking, added 9 November 2026, sorts the first of those fields across every card, and the false negative rate ranking shows how much AI text the same tools missed in the same studies. 26 cards carry a false positive figure, but three come from the vendor's own page, two more state no per-check ceiling, and two, GLTR and the GPT-2 Output Detector, publish no retention statement and no institutional plan, which leaves Winston AI, GPTZero, Copyleaks, Sapling, ZeroGPT and Turnitin, every field read on 8 November 2026 from the page recorded on the card. The view opens with what the Common App and College Board publish, because admissions readers ask first whether anyone in the application chain runs a detector at all. This page sits under AI detectors by use case beside the views for school districts and recruiters, and the same cards can be compared side by side on the full board.
- Field 1Retention statement as published
- Field 2False positive figure and study
- Field 3650-word essay in one check
- Field 4Institutional plan as stated
Retention comes first because an application essay is a minor's personal statement and the office must know whether submitted text is stored, trained on or deleted. The false positive figure must come from a study in the studies index, because a vendor's own rate was measured on text the vendor chose. The input limit must take 650 words in one check, the length this view tests for a personal essay; at 5.5 characters a word that is about 3,600 characters. The institutional plan must exist as a stated tier, because an office buys seats, not a personal subscription. The rules are in the methodology, and the false positive rate ranking orders every card on the same evidence field.
Which fields decide fit for an admissions office?
| Detector | Retention statement as published | False positive figure and study | 650-word essay in one check | Institutional plan as stated |
|---|---|---|---|---|
 Winston AI | Stores submitted text; deleting a report deletes the content; no training on identifiable content without consent (policy of July 17, 2026) | 0 false accusations in 18 human-written documents, Weber-Wulff 2023 | Yes: 200,000 characters per scan on paid plans | Enterprise: Contact sales |
 GPTZero | Dashboard inputs stored and used in aggregate; API inputs not stored; FERPA, GDPR and CCPA compliance stated (policy of May 20, 2026) | 0.00% at naive threshold 0.5, RAID 2024 | Yes: 150,000 characters with a free account; 10,000 without | Professional Team Plan: 500,000 words per member per month; min 2 seats |
 Copyleaks | May use data to train its models; direct customers may opt out; API content not stored (policy revised August 24, 2026) | 0 of 42 human papers, Walters 2023 | Yes: no character max on the web platform; 25,000 characters free | Education and Enterprise: Custom |
 Sapling | Content stored and used only to improve the service; API zero_retention flag; results cached 3 days | 0 of 42 human papers (31% uncertain), Walters 2023 | Pro or Enterprise only: 100,000 characters; free check covers 2,000 | Enterprise: starts at 10 seats, $15/seat/month |
 ZeroGPT | No training on private user content; retained only as long as needed to operate the service; hosted in Germany (policy of October 07, 2026) | 0 false accusations in 18 human-written documents, Weber-Wulff 2023 | Yes: 15,000 characters per detection on FREE | EDU: C$32.99 /month, billed yearly only, C$395.88 /year |
 Turnitin | AI results retained in the similarity report; institutions choose whether papers enter the repository; full deletion on request | 0 of 42 human papers, Walters 2023 | Yes: 300 to 30,000 words of prose | Licence for Turnitin Originality; no individual access |
Values as each vendor or study states them, read 8 November 2026.
What do the Common App and College Board publish about AI in applications?
The Common App's Fraud Policy, dated August 2023 on the page, lists among its examples of fraud the misrepresentation as one's own work of "the substantive content or output of an artificial intelligence platform, technology, or algorithm", and names no detection tool anywhere in its text (commonapp.org/fraud-policy, read 9 November 2026). The same policy states that Common App does not itself judge, evaluate or verify information about a student-applicant and that any decision on an application rests with the member college. A Common App applicant support article, last modified 21 September 2023, tells applicants that the personal essay must be their own work and not to use AI software such as ChatGPT or Bard to write the essay (appsupport.commonapp.org, read 9 November 2026).
The College Board publishes AI guidance for its AP program rather than for admissions. Its AP Central page for the 2026-27 academic year sets course-level rules, requires AP Seminar and AP Research students to complete checkpoints with a teacher who affirms the work is authentic, and scores a task 0 when the checkpoints are missed; the page names no detector and reserves the right to investigate submissions and request interim work (apcentral.collegeboard.org, read 9 November 2026). The question page on what AI detector College Board uses reads the AP exam security policy, the Art and Design policy guide and the SAT security page as well, and none names a tool. Neither organisation publishes a detector policy for application essays, so any detector use sits with individual colleges, and this view is built for the office that decides to run one.
Why each card is in this view
- AIDI-D-0005Winston AI
privacy policy of July 17, 2026 states stored text, deletion with the report and no training on identifiable content without consent (privacy policy); 0 false accusations in 18 human-written documents in Weber-Wulff 2023 (arXiv); 200,000 characters per scan on paid plans (gowinston.ai/pricing), all read 8 November 2026.
- AIDI-D-0001GPTZero
dashboard inputs stored and used in aggregate, API inputs not stored, FERPA, GDPR and CCPA compliance stated (gptzero.me/faq, privacy policy); 0.00% of human-written documents flagged at threshold 0.5 in RAID 2024 (arXiv); Professional Team Plan from 2 seats (gptzero.me/pricing), all read 8 November 2026.
- AIDI-D-0003Copyleaks
policy revised August 24, 2026 allows training use with an opt-out for direct customers; API content not stored (privacy policy, API page); 0 of 42 human papers flagged in Walters 2023 (publisher PDF); Education and Enterprise plans custom (copyleaks.com/pricing), all read 8 November 2026.
- AIDI-D-0010Sapling
entered text stored and used only to improve the service, API zero_retention flag (privacy, API docs); 0 of 42 human papers flagged in Walters 2023 with 31% uncertain; a 650-word essay exceeds the 2,000-character free check and fits the 100,000-character Pro check; Enterprise from 10 seats at $15 per seat per month (sapling.ai/pricing), all read 8 November 2026.
- AIDI-D-0011ZeroGPT
policy of October 07, 2026 states no training on private user content and retention only as long as needed (privacy policy); 0 false accusations in 18 documents in Weber-Wulff 2023 (arXiv); EDU C$32.99 per month billed yearly only, C$395.88 per year, in CAD to a Canadian viewer (zerogpt.com/pricing), all read 8 November 2026.
- AIDI-D-0002Turnitin
AI results retained within the similarity report, institutions choose whether papers enter the repository, full deletion on request (Turnitin FAQ); 0 of 42 human papers flagged in Walters 2023, whose author notes the papers could have trained the detector (publisher PDF); institutional licence only, all read 8 November 2026.
What the evidence says
What does the false positive evidence mean for applicants?
The results as evidence question, added 9 November 2026, lists the seven primary documents in the index on whether a score can stand alone, among them the College Board guidance this view reads.
The applicant population is the study population in Liang 2023. Across seven detectors, the paper reports an average false positive rate of 61.22% on 91 human-written TOEFL essays and near-perfect accuracy on 88 US 8th-grade essays; 18 of the 91 TOEFL essays were flagged as AI by all seven detectors, and GPTZero, Sapling, Originality.ai and ZeroGPT were among the seven (Liang 2023, arXiv, read 8 November 2026). An international applicant pool writing personal essays in a second language is the group this study describes.
Walters 2023 is the closest sample to an application essay: 42 human-written first-year composition papers and 84 GPT-written papers through 16 detectors. Copyleaks, Turnitin and Sapling each flagged 0 of the 42 human papers, Originality.ai 0 with 2 uncertain, and GPTZero and Scribbr 2% each (Walters 2023, publisher PDF, read 8 November 2026). Among the cards indexed here, the same study found detection of the 42 GPT-4 papers ranged from 33% for Sapling to 100% for Copyleaks, Turnitin and Originality.ai, so a clean human result and a reliable AI catch are separate questions.
A 0 on 18 or 42 documents is a small base: Weber-Wulff 2023 recorded 0 false accusations for Winston AI and ZeroGPT but 9 in 18 for GPTZero in the same test (Weber-Wulff 2023, arXiv, read 8 November 2026).
No detector score should be the only evidence in a misconduct or hiring decision. The Common App's own policy states that decisions on an application rest with the member college, and GPTZero's FAQ says results should not be used as the final verdict, both read 9 and 8 November 2026 respectively.
The same limits read for a longer document, a 6,000 to 10,000-word manuscript rather than a 650-word essay, are in the researchers view, added 9 November 2026, with Gao 2023's result on medical abstracts as its one study on scientific writing.
What a student can buy for their own draft is a different field from what an office licenses: the student plan question lists the four cards that price a student or academic tier for one person, and the nonprofit and education question the twelve that sell an institutional programme, both read 10 November 2026.
Sources
- https://www.commonapp.org/fraud-policy (read 9 November 2026)
- https://appsupport.commonapp.org/applicantsupport/s/article/Personal-Essay-760 (read 9 November 2026)
- https://apcentral.collegeboard.org/exam-administration-ordering-scores/administering-exams/exam-policies/artificial-intelligence-tools (read 9 November 2026)
- https://gowinston.ai/pricing/ (read 8 November 2026)
- https://gowinston.ai/privacy-policy/ (read 8 November 2026)
- https://gptzero.me/faq (read 8 November 2026)
- https://gptzero.me/pricing (read 8 November 2026)
- https://gptzero.me/privacy-policy.html (read 8 November 2026)
- https://copyleaks.com/ai-content-detector (read 8 November 2026)
- https://copyleaks.com/pricing (read 8 November 2026)
- https://copyleaks.com/privacy-policy (read 8 November 2026)
- https://copyleaks.com/api/ai-detector (read 8 November 2026)
- https://sapling.ai/ai-content-detector (read 8 November 2026)
- https://sapling.ai/pricing (read 8 November 2026)
- https://sapling.ai/privacy (read 8 November 2026)
- https://sapling.ai/docs/api/detector/ (read 8 November 2026)
- https://www.zerogpt.com/pricing (read 8 November 2026)
- https://www.zerogpt.com/privacy-policy (read 8 November 2026)
- https://guides.turnitin.com/hc/en-us/articles/28477544839821-Turnitin-s-AI-writing-detection-capabilities-FAQs (read 8 November 2026)
- https://www.scribbr.com/ai-detector/ (read 8 November 2026)
- https://originality.ai/pricing (read 8 November 2026)
- https://www.pangram.com/ (read 8 November 2026)
- https://proofademic.ai/ai-text-detector/ (read 8 November 2026)
- https://www.degruyterbrill.com/document/doi/10.1515/opis-2022-0158/pdf (read 8 November 2026)
- https://arxiv.org/pdf/2304.02819v3 (read 8 November 2026)
- https://arxiv.org/pdf/2306.15666 (read 8 November 2026)
- https://arxiv.org/pdf/2405.07940 (read 8 November 2026)
Revision log: 2026-11: view created; 2026-11-09: College Board question page linked2026-11-09: researchers view linked (fifth wave)2026-11-09: the two character conversions moved from 6 to 5.5 characters a word under the one-rate rule (650 words about 3,600 characters; Sapling's 2,000 characters about 360 words)2026-11-09: results as evidence question linked (seventh wave)2026-11-10: twelfth-wave pages linked
FAQ: AI detectors for admissions offices
Do college admissions use AI detectors?
Neither the Common App nor the College Board publishes a detector policy for application essays. The Common App Fraud Policy, dated August 2023, treats passing off substantive AI output as one's own work as fraud and names no detection tool; the College Board's AI guidance covers AP coursework and uses teacher checkpoints rather than a named detector (commonapp.org/fraud-policy, apcentral.collegeboard.org, read 9 November 2026). Individual colleges set their own practice.
Which AI detectors fit admissions offices?
Six cards state all four fields: Winston AI, GPTZero, Copyleaks, Sapling, ZeroGPT and Turnitin, as read on 8 November 2026. Each has a false positive figure from Weber-Wulff 2023, Walters 2023 or RAID 2024, a published retention statement, a per-check limit that takes 650 words and a stated institutional or team tier.
Does a 650-word essay fit a free check?
On three of the six: GPTZero takes 150,000 characters with a free account, Copyleaks 25,000 characters and ZeroGPT 15,000 characters per detection on FREE, all read 8 November 2026 (gptzero.me/faq, copyleaks.com, zerogpt.com/pricing). Sapling's free check covers 2,000 characters, about 360 words at 5.5 characters a word; Winston AI's pricing table leaves the Free row of the characters-per-scan line blank; Turnitin has no free access.
Which detectors store an applicant's essay?
Winston AI, GPTZero (dashboard) and Sapling state that submitted text is stored; Copyleaks allows training use with an opt-out; ZeroGPT states retention only as long as needed to operate the service; Turnitin retains AI results within the similarity report, all read 8 November 2026 on the policy pages linked above. Read the retention field on each card before any contract that touches minors' writing.
What does the evidence say about international applicants?
Liang 2023 reports an average false positive rate of 61.22% across seven detectors on 91 human-written TOEFL essays, against near-perfect accuracy on 88 US 8th-grade essays (arXiv, read 8 November 2026). Four of the seven detectors tested have cards in this index.
What does an admissions office pay?
ZeroGPT states C$395.88 per year for EDU and Sapling $15 per seat per month from 10 seats for Enterprise (zerogpt.com/pricing, sapling.ai/pricing, checked 8 November 2026). Winston AI, Copyleaks and Turnitin quote on request, and GPTZero's Professional Team Plan starts at 2 seats with 500,000 words per member per month (gptzero.me/pricing).
