The AI detector index
One entry per AI detector, fourteen fields in the same order, each value with its source and the date it was read: twenty-two cards and nine study records at index version 2026.11.2, twenty-two cards and seventeen records at version 2026.11.3 after eight records were added on 9 November 2026, and seventy-two cards and thirty-five study records at version 2026.11.13 after six more cards, then ten more, then four more (Codeleaks, Illuminarty, Pindrop Pulse and Resemble Detect) and two records (Pegoraro 2023 and Fleckenstein 2024) followed the same day. Where no evidence is published, the field is left empty. How an entry is made The twelfth wave, the last of the twelve-month plan, added no card and thirteen pages on 10 November 2026 at index version 2026.11.13: four questions (student plans, scanned PDFs, music detectors, nonprofit and education pricing), two comparisons (Sapling vs GPTZero, Pindrop vs Resemble Detect), four views (customer support QA, team seats and SSO, handwriting and OCR, mobile apps) and three updates, the 2027 census among them, which counts 72 cards, 12 withdrawn, unreachable or redirected, 35 with a study record and four planned rows held. A team field was written to every card the same day, and two file type counts were corrected.
Every detector, the same fields
Updated
| Register ID | Detector | Detects | Languages | Free tier | API | LMS | Independent studies |
|---|---|---|---|---|---|---|---|
| AIDI-D-0001 | GPTZero | Text, image | 5 | Yes | Yes | Yes | 16 |
| AIDI-D-0002 | Turnitin | Text | 7 | No | No | Yes | 5 |
| AIDI-D-0003 | Copyleaks | Text, code, image, video | 30 | Yes | Yes | Yes | 9 |
| AIDI-D-0004 | Originality.ai | Text, code | 30 | Yes | Yes | Yes | 7 |
| AIDI-D-0005 | Winston AI | Text, image, handwriting | 11 | Yes | Yes | Yes | 2 |
| AIDI-D-0006 | Pangram | Text, image, video | 24 | Yes | Yes | Yes | 4 |
| AIDI-D-0007 | QuillBot | Text, image | 6 | Yes | No | No | 2 |
| AIDI-D-0008 | Grammarly | Text | Yes | No | No | 0 | |
| AIDI-D-0009 | Scribbr | Text | 1 | Yes | No | No | 1 |
| AIDI-D-0010 | Sapling | Text | 16 | Yes | Yes | No | 3 |
| AIDI-D-0011 | ZeroGPT | Text, image, video | Yes | Yes | No | 7 | |
| AIDI-D-0012 | Phrasly | Text, image | 4 | Yes | Yes | No | 0 |
| AIDI-D-0013 | JustDone | Text | Yes | No | No | 0 | |
| AIDI-D-0014 | Proofademic | Text, image, voice | 23 | Yes | Yes | No | 0 |
| AIDI-D-0015 | Walter Writes | Text, image, voice | 20 | Yes | Yes | Yes | 0 |
| AIDI-D-0016 | Hive | Text, image, video, voice, music | Yes | Yes | No | 2 | |
| AIDI-D-0017 | GLTR | Text | Yes | Yes | No | 4 | |
| AIDI-D-0018 | AIDetectPlus | Text | Yes | No | No | 0 | |
| AIDI-D-0019 | GPT-2 Output Detector | Text | Yes | Yes | No | 13 | |
| AIDI-D-0020 | Codequiry | Code | Yes | Yes | Yes | 0 | |
| AIDI-D-0021 | WasItAI | Image | Yes | Yes | No | 0 | |
| AIDI-D-0022 | Reality Defender | Video, image, voice, text | Yes | Yes | No | 0 | |
| AIDI-D-0023 | Writer | Text | Yes | No | No | 5 | |
| AIDI-D-0024 | Undetectable AI | Text | 50 | Yes | Yes | No | 1 |
| AIDI-D-0025 | AI or Not | Image, video, voice, music, text | Yes | Yes | No | 1 | |
| AIDI-D-0026 | Grammarly Authorship | Text | Yes | No | Yes | 0 | |
| AIDI-D-0027 | Turnitin Clarity | Text | No | No | Yes | 0 | |
| AIDI-D-0028 | SynthID | Image, video, audio, text | Yes | No | No | 0 | |
| AIDI-D-0029 | Smodin | Text | 100 | Yes | Yes | No | 0 |
| AIDI-D-0030 | BrandWell | Text | 1 | Yes | No | No | 3 |
| AIDI-D-0031 | Surfer | Text | 1 | Yes | Yes | No | 0 |
| AIDI-D-0032 | Compilatio | Text, handwriting | 35 | No | Yes | Yes | 1 |
| AIDI-D-0033 | Quetext | Text | 25 | Yes | Yes | No | 0 |
| AIDI-D-0034 | TruthScan | Text, image, pdf documents, video, voice | 30 | Yes | Yes | No | 0 |
| AIDI-D-0035 | DetectGPT | Text | Yes | No | No | 4 | |
| AIDI-D-0036 | Sightengine | Image, video, deepfake, voice, music | Yes | Yes | No | 1 | |
| AIDI-D-0037 | Video | Yes | Yes | No | 0 | ||
| AIDI-D-0038 | Sensity | Video, image, voice, file forensics | No | Yes | No | 0 | |
| AIDI-D-0039 | Codeleaks | Code | No | Yes | No | 0 | |
| AIDI-D-0040 | Illuminarty | Image, text | Yes | Yes | No | 1 | |
| AIDI-D-0041 | Pindrop Pulse | Voice, video | 40 | No | Yes | No | 0 |
| AIDI-D-0042 | Resemble Detect | Voice, image, video, music | 54 | Yes | Yes | No | 0 |
| AIDI-D-0043 | Decopy | Text | Yes | No | No | 0 | |
| AIDI-D-0044 | Content Detector AI | Text | 1 | Yes | No | No | 3 |
| AIDI-D-0045 | GPTKit | Text | 1 | Yes | Yes | No | 2 |
| AIDI-D-0046 | PlagiarismCheck | Text | 2 | Yes | Yes | Yes | 1 |
| AIDI-D-0047 | Isgen | Text | 80 | Yes | Yes | No | 0 |
| AIDI-D-0048 | AI Detector Pro | Text | 7 | Yes | Yes | No | 0 |
| AIDI-D-0049 | Fast-DetectGPT | Text | Yes | Yes | No | 1 | |
| AIDI-D-0050 | Paperpal | Text | 1 | Yes | No | No | 0 |
| AIDI-D-0051 | Detecting-AI.com | Text | Yes | Yes | Yes | 0 | |
| AIDI-D-0053 | Binoculars | Text | Yes | No | No | 2 | |
| AIDI-D-0054 | Monica | Text | Yes | No | No | 0 | |
| AIDI-D-0055 | Trinka | Text | Yes | Yes | Yes | 0 | |
| AIDI-D-0056 | StealthWriter | Text | Yes | No | No | 0 | |
| AIDI-D-0057 | Wordtune | Text | 5 | Yes | No | No | 1 |
| AIDI-D-0058 | Check For AI | Text | 1 | Yes | No | No | 1 |
| AIDI-D-0059 | Writefull | Text | Yes | No | No | 2 | |
| AIDI-D-0060 | Passed.ai | Text | 1 | Yes | No | No | 0 |
| AIDI-D-0062 | ElevenLabs Classifier | Voice: elevenlabs-generated speech only | Yes | No | No | 0 | |
| AIDI-D-0063 | FakeCatcher | Video | No | No | No | 0 | |
| AIDI-D-0064 | Deepfake-O-Meter | Image, video, voice | Yes | No | No | 0 | |
| AIDI-D-0065 | Ircam Amplify | Music | No | Yes | No | 0 | |
| AIDI-D-0066 | GPT Radar | Text | Yes | No | No | 1 | |
| AIDI-D-0067 | Grammica | Text | Yes | No | No | 1 | |
| AIDI-D-0068 | IvyPanda | Text | Yes | No | No | 1 | |
| AIDI-D-0069 | SEO.ai | Text | 50 | Yes | No | No | 1 |
| AIDI-D-0070 | AI Writing Check | Text | Yes | No | No | 1 | |
| AIDI-D-0071 | RADAR | Text | Yes | No | No | 2 | |
| AIDI-D-0072 | Ghostbuster | Text | Yes | No | No | 0 | |
| AIDI-D-0073 | Hiya Deepfake Detector | Voice: the voice in a video or audio played in a chrome tab | Yes | Yes | No | 0 | |
| AIDI-D-0074 | McAfee Deepfake Detector | Voice: ai-generated audio in a video playing in the browser | 1 | Yes | No | No | 0 |
| No entry matches that name. See all detectors | |||||||
What the studies found
Each published study is its own record, with its sample, the detectors it tested and its results as reported.
All studies- AIDI-S-0001Weber-Wulff et al. 2023: testing of detection tools for AI-generated text2023Independent14 tested
- AIDI-S-0002Liang et al. 2023: GPT detectors are biased against non-native English writers2023Independent7 tested
- AIDI-S-0003Walters 2023: the effectiveness of 16 AI text detectors on student essays2023Independent16 tested
- AIDI-S-0004Elkhatat et al. 2023: evaluating AI content detection tools2023Independent5 tested
- AIDI-S-0005Perkins et al. 2024: AI detectors against adversarial techniques2024Independent7 tested
- AIDI-S-0006RAID: the shared benchmark for AI text detectors2024Independent6 tested
- AIDI-S-0007OpenAI's AI text classifier, 2023: the published accuracy record2023Vendor or lab1 tested
- AIDI-S-0008Turnitin AI detection accuracy: vendor claims and independent tests2023Vendor or lab1 tested
- AIDI-S-0009Chaka 2023: detecting AI content in ChatGPT, YouChat and Chatsonic responses with five tools2023Independent5 tested
- AIDI-S-0010Sadasivan et al. 2023: can AI-generated text be reliably detected? Recursive paraphrasing against four detector classes2023Independent7 tested
- AIDI-S-0011Krishna et al. 2023: paraphrasing evades detectors of AI-generated text, but retrieval is an effective defense2023Independent5 tested
- AIDI-S-0012Orenstrakh et al. 2023: detecting LLM-generated text in computing education, eight detectors on 164 submissions2023Independent8 tested
- AIDI-S-0013Gao et al. 2023: ChatGPT-generated scientific abstracts against the GPT-2 Output Detector and blinded reviewers2023Independent1 tested
- AIDI-S-0014Jabarian and Imas 2025: artificial writing and automated detection, four detectors on 1,992 passages2025Independent4 tested
- AIDI-S-0015Van Vlasselaer et al. 2026: who wrote this? GPTZero, Pangram, Copyleaks and Turnitin on 160 academic papers2026Independent4 tested
- AIDI-S-0016NIST AI 700-1, 2025: the GenAI text-to-text pilot, 348 detector submissions on AI and human summaries2025Vendor or lab3 tested
- AIDI-S-0017Ha and Passananti et al. 2024: organic or diffused? Hive, Optic and Illuminarty on 280 human artworks and 350 AI images2024Independent5 tested
- AIDI-S-0018Pegoraro et al. 2023: to ChatGPT or not to ChatGPT, 20 detection methods and no true positive rate above 47.3%2023Independent20 tested
- AIDI-S-0019Fleckenstein et al. 2024: do teachers spot AI? 89 novice and 200 experienced teachers on ChatGPT and student essays2024Human-rater study0 tested
- AIDI-S-0021Ibrahim 2023: GPT-2 Output Detector and Crossplag on 120 ESL student essays and 120 ChatGPT essays2023Independent2 tested
- AIDI-S-0023Kar et al. 2024: how sensitive are the free AI detector tools, ten tools on one ChatGPT article and three paraphrases2024Independent10 tested
- AIDI-S-0024Alshammari and Rao 2025: six AI text detectors on 49 DeepSeek-V3 answers, paraphrased and humanized2025Independent6 tested
- AIDI-S-0025DAMAGE 2025: nineteen AI humanizers against GPTZero, Binoculars, RADAR and a Pangram Labs model2025Independent5 tested
- AIDI-S-0026Epoch AI 2026: Pangram, GPTZero and Originality.ai on 495 human passages and 594 AI passages2026Data insight (not peer reviewed)3 tested
- AIDI-S-0027ASVspoof 5: 53 teams' deepfake speech detectors on 680,774 crowdsourced utterances and 16 attacks2026Challenge0 tested
- AIDI-S-0028SAFE 2025: twelve teams' synthetic voice detectors on 21 real sources, 13 TTS models and laundered audio2025Challenge0 tested
- AIDI-S-0029DFDC 2020: what 2,114 teams scored on 10,000 unseen deepfake videos2020Challenge0 tested
- AIDI-S-0030RAID image 2025: adversarial images cut Hive to 0.44 and Sightengine to 0.00 (accuracy)2025Independent4 tested
- AIDI-S-0031Vanderbilt 2023: why a university disabled Turnitin's AI detector, with the numbers it used2023Institutional statement0 tested
- AIDI-S-0032Vendor benchmarks: what GPTZero and Pangram publish about their own accuracy2026Vendor or lab5 tested
- AIDI-S-0033Rashidi et al. 2023: 14,400 pre-ChatGPT journal abstracts through the RoBERTa GPT-2 detector2023Independent1 tested
- AIDI-S-0034ASVspoof 2021: 54 teams' spoofed and deepfake speech detectors across logical access, physical access and deepfake tasks2023Challenge0 tested
- AIDI-S-0035ADD 2023: the second Audio Deepfake Detection Challenge, 145 teams across fake-audio game, region location and algorithm recognition2023Challenge0 tested
- AIDI-S-0036FaceForensics++ 2019: six forgery detectors and 204 human observers on 1.8 million manipulated face images2019Challenge0 tested
- AIDI-S-0037Vendor compilation: what Originality.ai's accuracy page and 17-study meta-analysis publish about its own detector2026Vendor or lab10 tested
Rankings derived from the index
- AIDI-R-01Independently testedAI detectors ranked by how many published studies tested them, 17 records, with every study linked, each result as reported, and the method and date stated.
- AIDI-R-02False positive rateAI detectors ranked by published false positive rate, each figure from a named study or vendor statement with its sample size and date, and the limits of each.
- AIDI-R-03Free tiersWhat each AI detector's free tier includes, in words or characters per check and per period as checked on the vendor page, ranked by what you can check free.
- AIDI-R-04LanguagesAI detectors ranked by the languages each vendor says it detects, including Spanish, with the vendor source and date per row and studies on non-English text.
- AIDI-R-05Input limitsAI detectors ranked by the largest words or characters a vendor states for one check, with the plan carrying each limit, the source per row and unlimited tiers.
- AIDI-R-06Data retentionAI detectors ranked by how long each vendor's policy says pasted text or images are kept, shortest first, with the clause quoted, the policy linked and dated.
- AIDI-R-07Studies by content typeThe 14 AI detectors that state image, voice, video, code or music detection, ranked by how many published studies scored them on those types; one record exists.
- AIDI-R-08Adversarial robustnessAI detectors ranked by the share of paraphrased, humanised or edited AI text each still caught in a published study, with the attack, base and date per row.
- AIDI-R-09False negative rateAI detectors ranked by lowest published false negative rate, the share of AI text a study found each missed, with base, threshold and false positive rate shown.
- AIDI-R-10Claim vs evidence gapEight AI detector vendors' headline accuracy claims set beside the best accuracy a published study prints for each tool, ranked by the gap in points, sourced.
- AIDI-R-11Platform integrationsAI detectors ranked by how many of eight integration surfaces each vendor names: Chrome extension, Google Docs, Word, LMS, Zapier, MCP, API, mobile app.
- AIDI-R-12File typesAI detectors ranked by how many upload formats each vendor names, from Sensity's 22 media containers to three-format text tools, each list quoted and dated.
- AIDI-R-13Minimum inputAI detectors ranked by the smallest text a vendor requires for one check, from Phrasly's 30 words to Turnitin's 300, with the vendor's wording per row.
- AIDI-R-14API pricingAI detector APIs ranked by the published price of a text check per 1,000 words, from AI or Not's $5 per million words to Copyleaks' plan credits.
- AIDI-R-15LMS coverageAI detectors ranked by how many LMS platforms each vendor names, from Compilatio's seven and Copyleaks' six to Grammarly Authorship's one; 11 of 42 cards.
- AIDI-R-16Free checks per dayAI detectors ranked by the free checks per day each vendor states, from Sightengine's 500 operations to QuillBot's 6 scans; 7 of 63 cards state a daily figure.
- AIDI-R-17Generators coveredAI detectors ranked by how many AI generator models each vendor names, from Resemble Detect's 62 to GLTR's one; 56 of 72 cards name one.
- AIDI-R-18Languages evidenceAI detectors ranked by how many indexed studies print a result on non-English, translated or non-native English text, GPTZero first with five; 19 of 72 cards.
- AIDI-R-19Code languagesAI code detectors ranked by how many programming languages each vendor names, Codequiry's 12 to Originality.ai's 7; 4 of 42 cards state code, 3 name languages.
- AIDI-R-20Launch year29 of 72 AI detectors state a launch year, from Codequiry in 2017 to Turnitin Clarity in 2025; 14 launched in 2023, seven are withdrawn, five changed hands.
Find detectors by use case
- AIDI-U-01BusinessWhich AI detectors in the index fit business use, on API, retention policy, team plans and price, plus how a manager can audit a report without over-reading it.
- AIDI-U-02School districtsWhich AI detectors in the index fit K-12 districts, on LMS integration, student data policy, false positive evidence and district pricing, every card linked.
- AIDI-U-03Copywriters and agenciesWhich AI detectors in the index fit agency and copywriting work, on bulk checks, API, team seats and price per word, with the studies that matter for edits.
- AIDI-U-04Job applicationsWhich AI detectors in the index fit screening cover letters and applications, what the false positive evidence means for candidates, and the fields HR needs.
- AIDI-U-05CodeEvery AI code detector in the index compared on languages supported, input limits, API, pricing as checked, and the published evidence for detecting AI code.
- AIDI-U-06Admissions officesWhich AI detectors in the index fit admissions work, on retention, study-sourced false positive evidence, a 650-word essay limit and an institutional plan.
- AIDI-U-07Marketplaces and review platformsWhich AI detectors fit short product reviews at volume: an API, a stated minimum under 300 characters or short-text handling, and a bulk or batch route.
- AIDI-U-08Researchers checking their own manuscriptsWhich AI detectors take a full manuscript: 29 text cards with a 5,000-word per-check limit or a document upload, plus the one study on scientific abstracts.
- AIDI-U-09Law firms and courts: an index fit check9 of 72 AI detector cards state a retention period or no-storage route plus an enterprise deployment or report export, read with the one institutional record.
- AIDI-U-10Grant and proposal reviewers18 of 72 AI detector cards take a full proposal and state a retention policy, read beside the NIH and NSF rules on uploading proposals to outside AI tools.
- AIDI-U-11Ai text detectors for newsrooms and fact desks5 of 72 AI detector cards state an API or browser extension and a minimum under 500 characters, read with the human-baseline and ChatGPT-sweep records.
- AIDI-U-12Localisation and translation teams14 of 72 AI detector cards state 10 or more languages; 6 have a study record on translated or non-native text, where accuracy fell from 94% to 69% in 2023.
- AIDI-U-13Government and public sector buyers19 of 72 AI detector cards state an on-premise deployment, a named data region or a SOC Type 2 or ISO 27001 statement, each with its source; 0 state FedRAMP.
- AIDI-U-14Self-publishing authors: the cards that take a manuscript and say what happens to it12 of 72 AI detector cards take a document upload or 5,000 words a check and state a retention policy: limits, formats, prices and the KDP rule as read.
- AIDI-U-15Customer support quality assurance: the cards with an api, a short-text statement, a bulk route and a retention policy2 of 72 AI detector cards state all four fields a support QA team needs: an API, a short-text statement under 500 characters, bulk checking and retention.
By feature
- AIDI-F-01APIWhich AI detectors in the index offer an API, with links to each vendor's docs, rate and input limits, pricing as checked, and the studies that tested each one.
- AIDI-F-02LMS integrationsWhich AI detectors in the index integrate with Canvas, Moodle, Blackboard, D2L or Google Classroom, how each integration works, and what institutions pay.
- AIDI-F-03ModalitiesA coverage matrix of the index: which detectors handle text, code, image, voice, video and music, as each vendor states, with links to every card and tool.
- AIDI-F-04Open sourceThe three AI detectors in the index that publish code or weights, GLTR, the GPT-2 Output Detector and SynthID Text, with licence, last commit and hosted extras.
- AIDI-F-05Published false positive rateThe 26 AI detector cards with a false positive figure, each marked study result or vendor statement, with value, base, threshold and date.
- AIDI-F-06No account20 of 72 AI detector cards state that a check runs with no sign-up or login, from Scribbr's unlimited free checks to GPTZero's 10,000-character widget.
- AIDI-F-07Bulk and batch17 of 72 AI detector cards state a bulk upload, batch API, ZIP or folder import, from GPTZero's 250 files to Sapling's 5; ceiling, plan and price for each.
- AIDI-F-08Compliance12 of 72 AI detector cards name a security or privacy standard on a vendor page: 8 SOC Type 2, 10 GDPR, 7 ISO 27001, 5 HIPAA, 3 FERPA, 0 FedRAMP, each dated.
- AIDI-F-09Model changelogs10 of 72 AI detector cards link a dated model changelog, release note or model card on a vendor page, from Turnitin to Pangram; latest entry per row.
- AIDI-F-10Institution-only7 of 72 AI detector cards sell the detector only by licence, demo, application or sales team, Turnitin to Ircam Amplify; 15 sell both routes.
- AIDI-F-11Team seats and SSO20 of 72 AI detector cards state a seat count, a per-seat price or SSO, SAML or SCIM: 17 state seats, 6 state SSO, 3 state both; 4 more name a team tier only.
- AIDI-F-12Handwriting and OCR3 of 72 AI detector cards state OCR of handwriting or image files: Winston AI, Compilatio, Copyleaks; 0 state OCR of a scanned PDF; 8 take images only.
- AIDI-F-13Mobile apps5 of 72 AI detector cards name a mobile app among their integrations: QuillBot, Grammarly, Walter Writes, Monica and Deepware, whose Play listing returned 404.
Questions about the index
How many AI detectors are in the index?
The index holds seventy-two detector cards and thirty-five study records at version 2026.11.13; sixteen cards date from 8 November 2026, and six cards and the ninth record were added on 9 November 2026, followed by eight more records, six more cards and then ten more cards (Smodin, BrandWell, Surfer, Compilatio, Quetext, TruthScan, DetectGPT, Sightengine, Deepware and Sensity) the same day. Every card carries the same fourteen fields in the same order, from what the tool detects and what it costs to the studies that tested it, and each row of the board above links to its card. New cards are added with a dated line in the corrections log.
Does the index test detectors itself?
No. Every result on the site comes from a published study, kept as its own record with its sample, method and date, or from the vendor's own page, quoted as a claim. The index runs no tests of its own and prints no score of its own; the methodology page sets out what a card records and what it does not.
Why is a field empty on some cards?
Because no source that meets the index's standard was found for it when the card was read: no vendor page states the value and no indexed study reports it. The row is left out rather than filled with an estimate, and the card's revision log shows when the field was last checked.
How are the rankings made?
Each of the twenty rankings sorts the cards on one field of the index data file (the twentieth, launch year, added 10 November 2026, sorts on the year the vendor states and dates every withdrawal), among them study record count, published false positive rate, false negative rate, free tier words per month, language count, stated data retention, named integration surfaces and stated upload formats. Only cards with a published value qualify, ties follow a rule stated on the page, and each page says how many of the seventy-two cards qualified. No overall ranking exists.
Which detectors have been independently tested?
35 of the 72, as of 9 November 2026: GPTZero in eleven study records, the GPT-2 Output Detector in nine, Copyleaks in six, Turnitin, Originality.ai and ZeroGPT in five each, GLTR in three, Pangram, Winston AI and Sapling in two each, and Hive and Scribbr in one. The ranking of independently tested detectors lists them in that order with every record linked.
How do I report an error?
Use the contact page and name the card or record, the field and the source that supports the change. A correction to a published figure is made on the page the day it is confirmed and written to the corrections log with the date, the page, what changed and the source that prompted it.




































































