AI detectors ranked by launch year, with every change of hands and withdrawal the cards record

Codequiry is the oldest detector in the index with a stated launch year, 2017, the year its about page gives for the founding of the Chennai company behind it (codequiry.com/about, read 9 November 2026); Turnitin Clarity is the newest at 2025, "a paid add-on for all Turnitin Feedback Studio customers starting in the third quarter of 2025" (turnitin.com, read 9 November 2026). Twenty-nine of the 72 cards state a launch year on a vendor page, a press release or the paper that introduced the method, and 14 of the 29 launched in 2023, the year after ChatGPT; the other 43 cards state none and sit in a second table with what their pages say instead. Seven of the 72 tools have since been withdrawn by their makers and five have changed hands, each dated from the vendor's own page or the Wayback Machine index on the card.

Updated Vendor pages read 8 to 10 November 2026

Short answer

The sort is the launch year each card states, earliest first, a description of what 29 vendors and paper authors say about when their detector appeared and not a measure of maturity or quality: Codequiry 2017; Sensity 2018; the GPT-2 Output Detector and GLTR 2019; Deepfake-O-Meter and Reality Defender 2021; Originality.ai, BrandWell, Illuminarty and FakeCatcher 2022; fourteen cards in 2023, from GPTZero and Turnitin to SynthID; four in 2024; and Turnitin Clarity in 2025, all read 8 to 10 November 2026. Ties within a year are broken by study records, then name. The events column records what the card says happened after launch: a rename, an acquisition, a closure or a withdrawal, each with its date and source.

How we ranked

AI detectors ranked by launch year, with every change of hands and withdrawal the cards record: register plate with the page title and its register ID
Launch year the vendor or paper states per detector, from the index data file.

Which AI detectors launched first, and which have since changed hands or closed?

AI detectors ranked by launch year, with every change of hands and withdrawal the cards record: the decision table
RankDetectorLaunch yearSource and read dateEvents since launch, with date and sourceBest forMain limitation
1Codequiry2017About page: the company was founded in 2017 (codequiry.com/about, read 9 November 2026)None recorded; the code detector sits inside a plagiarism productReaders who want the oldest dated code detectorA founding year, not a detector launch date; no study record
2Sensity2018About page, dated 3 February 2026: the company was founded "late in 2018" after a deepfake detector trained in March 2018 (sensity.ai, read 9 November 2026)None recordedReaders who want the earliest dated deepfake vendorA founding year; no study record
3GPT-2 Output Detector2019Release post of 5 November 2019 for the 1.5B GPT-2 model and its RoBERTa detector (openai.com, read 9 November 2026)Hosted demo paused on Hugging Face at the read date; the GitHub repository archived with a last commit of 13 December 2023; code and weights public under the MIT License (openai-openai-detector.hf.space, read 9 November 2026)Readers who want the oldest detector with 13 study recordsBuilt for GPT-2 text; 0.9503 of 1,992 human passages flagged in Jabarian and Imas 2025
4GLTR2019ACL 2019 demo paper, June 2019 (arXiv 1906.04043, read 9 November 2026)None recorded; the demo at gltr.io still ran on 9 November 2026Readers who want the oldest open visual method99.3% of RAID's human documents flagged at threshold 0.5
5Deepfake-O-Meter2021Platform paper: version 1.0 was presented at the 2021 IEEE Security and Privacy Workshops (arXiv 2404.13146, read 10 November 2026)Version 2.0 described in the 2024 paper; 41 hosted detectors on 10 November 2026Researchers who want several published methods on one uploadNo privacy policy or terms page; no study of the platform
6Reality Defender2021Company page: founded in 2021 (realitydefender.com/company, read 9 November 2026)None recordedEnterprise buyers who want a dated vendorA founding year; no study record
7Originality.ai2022Accuracy page release history: model 1.1 beta in November 2022 (originality.ai/blog/ai-accuracy, read 8 November 2026)Passed.ai's team joined Originality.ai and Passed.ai is closing, per every page of passed.ai on 10 November 2026 (passed.ai)Readers who want a dated model history from launch to 5.0The vendor's compilation of 17 studies restates its own headline figures
8BrandWell2022Archived detector FAQ: the first version was created in October 2022; the about page dates Content at Scale's launch to September 2022 (web.archive.org, 9 August 2026, read 9 November 2026)Renamed from Content at Scale to BrandWell, the name Weber-Wulff 2023 and Walters 2023 tested it under; the detector page has redirected to a Moxby marketplace Mod since 29 August 2026 (brandwell.ai, read 9 November 2026)Readers tracing a Content at Scale result to today's product0 of 9 edited texts classified correctly in Weber-Wulff 2023
9Illuminarty2022Terms and Privacy Policy dated 16 November 2022; the release post of 18 March 2023 updates an existing service (illuminarty.ai, read 9 November 2026)None recordedReaders who want a dated image detector with one record72.40% of 280 human artworks called AI in Ha and Passananti 2024
10FakeCatcher2022Intel newsroom post of 14 November 2022 announcing the productised detector (intel.com, read 10 November 2026)Never sold: no product page, price, sign-up or API existed on 10 November 2026Readers tracing the 96% figure to its dateA status record, not a product
11GPTZero2023Vendor page: launched in 2023 (gptzero.me, read 8 November 2026)None recordedReaders who want the most-tested 2023 launch, 16 records50% false positives in one study beside 0.08% in the vendor's own benchmark
12Turnitin2023Launch post: the English detector launched on 4 April 2023 (turnitin.com, read 8 November 2026)Vanderbilt disabled the indicator on 16 August 2023; Turnitin Clarity added as a paid add-on in 2025 (Vanderbilt statement)Institutions that want the licence's dated historyAll 40 fully AI papers scored under 20% in Van Vlasselaer 2026
13DetectGPT2023Paper first posted 26 January 2023 (arXiv 2301.11305, read 9 November 2026)Hosted demo gone by 26 November 2023 with no statement; MIT-licensed code remains (github.com/eric-mitchell/detect-gpt, read 9 November 2026)Researchers who want the first curvature method13 of 36 AI documents undetected in Weber-Wulff 2023
14Pangram2023About page: Pangram Labs founded in Brooklyn in 2023 (pangram.com/about-us, read 8 November 2026)None recordedReaders who want a 2023 launch with four records and two model cardsThe vendor's own DAMAGE paper tested its research models, not the product
15RADAR2023Paper first posted 7 July 2023, NeurIPS 2023 (arXiv 2307.03838, read 10 November 2026)None recorded; the repository's last push was 18 September 2025Researchers who want an adversarially trained 2023 method51.87% of human essays flagged at the default threshold in DAMAGE 2025
16AI or Not2023Press page: June 2023 coverage of the web app under its earlier name, Optic (aiornot.com/press, read 9 November 2026)Renamed from Optic, the name Ha and Passananti 2024 tested it underReaders tracing an Optic result to today's product24.47% of human artworks called AI in the one record
17AI Writing Check2023Vendor notice: "When we launched this tool in January 2023, the only Generative AI tool available was ChatGPT" (aiwritingcheck.org, read 10 November 2026)Withdrawn: "August 2023 Update - AI Writing Check is no longer available", the notice in place from the capture of 12 September 2023 (aiwritingcheck.org, read 10 November 2026)Readers who want the shortest-lived tool in the index, seven monthsWithdrawn; one averaged record
18Fast-DetectGPT2023Paper first posted 8 October 2023, ICLR 2024 (arXiv 2310.05130, read 10 November 2026)None recordedResearchers who want the faster curvature method23.2% of RAID's human documents flagged at threshold 0.5
19Wordtune2023Vendor blog post of 21 August 2023, "Introducing the Wordtune AI Content Detector" (wordtune.com, read 10 November 2026)Withdrawn: the page has returned 404 since the Wayback capture of 26 July 2025 with no statement (Wordtune card)Readers tracing a 2023 or 2024 Wordtune percentageWithdrawn; one record on one text
20AI Detector Pro2023About page: the first public version was released in January 2023 (aidetector.pro/pages/about, read 10 November 2026)None recordedReaders who want a dated small-vendor launchNo study record
21Codeleaks2023Press release of 9 November 2023 launching Codeleaks (copyleaks.com, read 9 November 2026)AI code verdict withdrawn from the API: standalone endpoint deprecated 28 August 2025, verdict removed from the submit endpoints 16 December 2025 (Codeleaks card)Readers who want a dated code launch and its dated retreatThe AI verdict for code no longer exists in the API
22ElevenLabs Classifier2023Launch post of 15 June 2023, "Introducing the ElevenLabs AI Speech Classifier" (web.archive.org, 30 August 2023, read 10 November 2026)The post's address now redirects to the product page; an in-app Audio Detector was addedReaders who want a dated self-detector from a voice generatorScores ElevenLabs audio only; no study record
23Ghostbuster2023Paper first posted 24 May 2023, NAACL 2024 (arXiv 2305.15047, read 10 November 2026)The demo moved to babbage-002 and davinci-002 after OpenAI deprecated the paper's models in January 2024Researchers who want the Berkeley feature methodNo study record; 74.7% on 91 TOEFL essays is the authors' own
24SynthID2023DeepMind blog post of 29 August 2023 announcing SynthID for images (deepmind.google, read 9 November 2026)SynthID-Text published in Nature on 23 October 2024Readers who want a dated watermark launchA watermark detector for Google outputs only
25Binoculars2024Paper first posted 22 January 2024, ICML 2024 (arXiv 2401.12070, read 10 November 2026)None recordedResearchers who want the newest open method with two records28.23% true positive rate after 19 humanizers in DAMAGE 2025
26Hiya Deepfake Detector2024Press release of 17 October 2024 launching the free extension (hiya.com, read 10 November 2026)Built on the AI Voice Detection technology Hiya acquired with Loccus.ai on 25 July 2024 (hiya.com, read 10 November 2026)Readers who want a dated free voice checkNo study record; the 99% is stated for the platform
27McAfee Deepfake Detector2024Press release of 21 August 2024 launching the detector on Lenovo AI PCs (mcafee.com, read 10 November 2026)Extended to HP Copilot+ PCs on 3 January 2025; the $9.99 launch offer replaced by McAfee plansReaders who want a dated on-device voice toolNo study record; English audio only
28Pindrop Pulse2024Pulse Inspect launched in Preview on 15 August 2024 (pindrop.com, read 9 November 2026)None recordedReaders who want a dated enterprise voice launchNo study record; the 99% rests on vendor product data
29Turnitin Clarity2025Blog post: a paid add-on starting in the third quarter of 2025 (turnitin.com, read 9 November 2026)None recordedReaders who want the newest dated launchA process record, not a classifier; no study record

Each year is the vendor's or the authors' own statement, read on the date shown; a founding year stands where the vendor gives no detector launch date, and the row says so.

Cards with no stated launch year

DetectorWhat the card records insteadEvents since, with date and sourceSource, read date
Scribbr and QuillBotNo launch year on either page; Scribbr is "part of QuillBot, within the broader Learneo family of businesses (founded 2012)"Both sit under Learneo, the ownership the shared owner question mapsscribbr.com/about-us, read 8 November 2026
Passed.aiNo launch year; the Terms are dated 25 January 2023Closing: "Passed.ai is closing, the team has joined Originality.ai"; registration closed by 10 November 2026passed.ai, read 10 November 2026
TruthScanNo launch year; the footer names Undetectable Inc. dba TruthScan, the company behind the Undetectable AI detectorShares an owner with Undetectable AI, per the Terms of 27 August 2026truthscan.com/terms, read 9 November 2026
WriterNo launch year on the archived pageWithdrawn: sunset on 22 December 2025 with the AI detection API endpoint; the page has redirected to writer.com since 24 December 2025web.archive.org, 9 December 2025, read 9 November 2026
WritefullNo launch year on the archived pageWithdrawn: "Our GPT Detector has been discontinued", between 31 December 2024 and 7 March 2025web.archive.org, 31 December 2024, read 10 November 2026
Check For AINo launch year; the Terms are dated 1 January 2023Withdrawn: sunset between 3 June and 23 July 2023, "a free nonprofit project for individuals"web.archive.org, 14 March 2023, read 10 November 2026
GrammicaNo launch year; the first Wayback capture of the detector is dated 10 March 2023, a capture date rather than a statementWithdrawn: removed in the site relaunch between the captures of 15 December 2025 and 12 January 2026; the address returned 307 to the home page on 10 November 2026web.archive.org, 31 July 2025, read 10 November 2026
GPT RadarNo launch year; the footer read Copyright 2022 and the first capture is dated 10 January 2023, capture dates rather than a statementWithdrawn: 301 to neuraltext.com on 14 August and 8 October 2026 per the Wayback CDX index; the host did not resolve on 10 November 2026web.archive.org, 12 July 2026, read 10 November 2026
IvyPanda and SEO.aiNo launch year; first Wayback captures of 19 March 2023 and 3 October 2023, capture dates rather than statementsNone recordedivypanda.com and seo.ai, read 10 November 2026
Grammarly, Grammarly Authorship, Winston AI, Copyleaks, Phrasly, Sapling, JustDone, ZeroGPT, Proofademic, Walter Writes, Hive, AIDetectPlus, WasItAI, Undetectable AI, Smodin, Surfer, Compilatio, Quetext, Sightengine, Deepware, Resemble Detect, Decopy, Content Detector AI, GPTKit, PlagiarismCheck, Isgen, Paperpal, Detecting-AI.com, Monica, Trinka, StealthWriter, Ircam AmplifyNo launch year for the detector on the pages each card cites; a company founding year, where one exists, is not entered as a detector launchContent Detector AI's page has redirected to a Moxby marketplace Mod since 18 August 2026, a move the card records rather than a withdrawalThe vendor page recorded on each card, read 8 to 10 November 2026

The withdrawals total seven: AI Writing Check (August 2023), Check For AI (June to July 2023), Writefull (early 2025), Wordtune (July 2025), Writer (22 December 2025), Grammica (January 2026) and GPT Radar (August 2026). Five changes of hands or names: Content at Scale became BrandWell, Optic became AI or Not, Hiya acquired Loccus.ai, Passed.ai's team joined Originality.ai, and TruthScan and Undetectable AI share Undetectable Inc. as owner. Scribbr and QuillBot sit together under Learneo.

One of the plan rankings in the rankings hub, beside independently tested, which counts study records, and the shared owner question, which maps who owns what. The shut-down question quotes each withdrawal notice in full; this page sets the launch year beside the withdrawal date so a reader can see how long each tool ran. The AI detector index carries every card.

Which AI detectors should you check first?

  • Ifyou want the longest-running detector with a study record: the GPT-2 Output Detector, released in 2019 and tested in 13 records, though it was built for GPT-2 text.
  • Ifyou want the newest launch with any record: Binoculars, posted January 2024, with RAID 2024 and DAMAGE 2025.
  • Ifyou want to know how long a withdrawn tool ran: AI Writing Check ran seven months, from January to August 2023; Wordtune about two years, from August 2023 to July 2025.
  • Ifyou want a dated launch from a vendor still selling the product: Turnitin (4 April 2023), Codeleaks (9 November 2023), Pindrop Pulse (15 August 2024) and McAfee (21 August 2024) each date theirs to the day.

What does each card state, tool by tool?

1

Codequiry

Verdict
Fits readers who want the oldest dated card, on code rather than text.
Value
2017: "founded in 2017", the about page of the Chennai company (codequiry.com/about, read 9 November 2026); the year is a founding, not a detector launch.
Best for
Instructors who want to know how long the code product has existed.
Main limitation
No study in the index tested Codequiry; the AI code claim of 90% caught and 1.3% falsely flagged is the vendor's own.
Read the Codequiry card
2

Sensity

Verdict
Fits readers who want the earliest dated deepfake vendor.
Value
2018: the about page, dated 3 February 2026, says the company was founded "late in 2018" after a detector trained in March 2018 (sensity.ai, read 9 November 2026).
Best for
Enterprise readers who want a vendor with a stated history.
Main limitation
A founding year; no study in the index tested Sensity.
Read the Sensity card
3

GPT-2 Output Detector

Verdict
Fits readers who want the oldest detector that studies keep testing.
Value
2019: OpenAI released the RoBERTa detector with the 1.5B GPT-2 model on 5 November 2019 (openai.com, read 9 November 2026); the hosted demo is paused and the repository archived.
Best for
Researchers who want a dated open baseline with 13 records.
Main limitation
The model card warns against using it as a ChatGPT detector; Jabarian and Imas 2025 flagged 0.9503 of 1,992 human passages at threshold 0.5.
Read the GPT-2 Output Detector card
4

GLTR

Verdict
Fits readers who want the oldest open visual method.
Value
2019: the ACL demo paper of June 2019 (arXiv 1906.04043, read 9 November 2026).
Best for
Readers who want to see token likelihoods rather than a score.
Main limitation
99.3% of RAID's human documents flagged at threshold 0.5.
Read the GLTR card
5

Deepfake-O-Meter

Verdict
Fits readers who want a dated research platform.
Value
2021: version 1.0 presented at the 2021 IEEE Security and Privacy Workshops, per the 2024 platform paper (arXiv 2404.13146, read 10 November 2026).
Best for
Researchers running several published methods on one file.
Main limitation
No privacy policy or terms page; no indexed study scored the platform.
Read the Deepfake-O-Meter card
6

Reality Defender

Verdict
Fits enterprise readers who want a dated vendor.
Value
2021: "founded in 2021" on the company page (realitydefender.com/company, read 9 November 2026).
Best for
Buyers who want a company history before a sales call.
Main limitation
A founding year; no study record and no published rate.
Read the Reality Defender card
7

Originality.ai

Verdict
Fits readers who want a dated model history from beta to 5.0.
Value
2022: the accuracy page's release history starts with model 1.1 in November 2022 (originality.ai/blog/ai-accuracy, read 8 November 2026). Event: Passed.ai's team joined Originality.ai and Passed.ai is closing (passed.ai, read 10 November 2026).
Best for
Publishers who want to know which model a 2023 result scored.
Main limitation
The vendor's own compilation of 17 studies restates its headline figures and omits five index records.
Read the Originality.ai card
8

BrandWell

Verdict
Fits readers tracing a Content at Scale result to today's name.
Value
2022: the archived FAQ dates the first version to October 2022 and the about page dates Content at Scale's launch to September 2022 (web.archive.org, 9 August 2026, read 9 November 2026). Events: renamed BrandWell; the detector page has redirected to a Moxby Mod since 29 August 2026.
Best for
Readers who meet the Content at Scale name in a 2023 study.
Main limitation
0 of 9 manually edited texts classified correctly in Weber-Wulff 2023.
Read the BrandWell card
9

Illuminarty

Verdict
Fits readers who want a dated image detector with one record.
Value
2022: the Terms and Privacy Policy are dated 16 November 2022 and the 18 March 2023 release post updates an existing service (illuminarty.ai, read 9 November 2026).
Best for
Readers reading an image verdict against the one record.
Main limitation
72.40% of 280 human artworks called AI in Ha and Passananti 2024.
Read the Illuminarty card
10

FakeCatcher

Verdict
Fits readers who want the date behind a 96% figure.
Value
2022: Intel's newsroom post of 14 November 2022 (intel.com, read 10 November 2026). Event: never sold; no product page, price or sign-up existed on 10 November 2026.
Best for
Readers tracing an announcement.
Main limitation
A status record; nothing to test.
Read the FakeCatcher card
11

GPTZero

Verdict
Fits readers who want the most-tested 2023 launch.
Value
2023: launched in 2023 per the vendor page (gptzero.me, read 8 November 2026); 16 study records.
Best for
Teachers who want the longest outside record of a 2023 tool.
Main limitation
50% false positives in one study beside 0.08% in the vendor's own benchmark.
Read the GPTZero card
12

Turnitin

Verdict
Fits institutions that want a licence's dated history.
Value
2023: the English detector launched on 4 April 2023 (turnitin.com, read 8 November 2026). Events: Vanderbilt disabled the indicator on 16 August 2023; Turnitin Clarity added in 2025.
Best for
Administrators dating a policy to a product change.
Main limitation
All 40 fully AI papers scored under 20% in Van Vlasselaer 2026.
Read the Turnitin card
13

DetectGPT

Verdict
Fits researchers who want the first curvature method.
Value
2023: the paper first posted 26 January 2023 (arXiv 2301.11305, read 9 November 2026). Event: the hosted demo was gone by 26 November 2023 with no statement.
Best for
Readers who want the method behind Fast-DetectGPT.
Main limitation
13 of 36 AI documents undetected in Weber-Wulff 2023.
Read the DetectGPT card
14

Pangram

Verdict
Fits readers who want a 2023 launch with four records and two model cards.
Value
2023: Pangram Labs founded in Brooklyn in 2023 (pangram.com/about-us, read 8 November 2026).
Best for
Readers who want a dated model card beside outside records.
Main limitation
DAMAGE 2025 tested the vendor's research models, not the product.
Read the Pangram card
15

RADAR

Verdict
Fits researchers who want an adversarially trained 2023 method.
Value
2023: the paper first posted 7 July 2023, NeurIPS 2023 (arXiv 2307.03838, read 10 November 2026).
Best for
Readers who want paraphrase robustness by design.
Main limitation
51.87% of human essays flagged at the default threshold in DAMAGE 2025.
Read the RADAR card
16

AI or Not

Verdict
Fits readers tracing an Optic result to today's product.
Value
2023: the press page links June 2023 coverage of the web app as Optic (aiornot.com/press, read 9 November 2026). Event: renamed from Optic to AI or Not.
Best for
Readers reading Ha and Passananti 2024 under the old name.
Main limitation
24.47% of human artworks called AI in that record.
Read the AI or Not card
17

AI Writing Check

Verdict
Fits readers who want the shortest-lived tool in the index.
Value
2023: "When we launched this tool in January 2023, the only Generative AI tool available was ChatGPT" (aiwritingcheck.org, read 10 November 2026). Event: withdrawn in August 2023 with the notice "AI Writing Check is no longer available".
Best for
Readers tracing a spring 2023 verdict.
Main limitation
Seven months of service; one averaged record.
Read the AI Writing Check card
18

Fast-DetectGPT

Verdict
Fits researchers who want the faster curvature method.
Value
2023: the paper first posted 8 October 2023, ICLR 2024 (arXiv 2310.05130, read 10 November 2026).
Best for
Readers who want DetectGPT's idea at a fraction of the cost.
Main limitation
23.2% of RAID's human documents flagged at threshold 0.5.
Read the Fast-DetectGPT card
19

Wordtune

Verdict
Fits readers tracing a withdrawn free widget.
Value
2023: the vendor blog post of 21 August 2023 (wordtune.com, read 10 November 2026). Event: the page has returned 404 since the Wayback capture of 26 July 2025.
Best for
Readers who meet a Wordtune percentage from 2023 or 2024.
Main limitation
Withdrawn; one record on one text.
Read the Wordtune card
20

AI Detector Pro

Verdict
Fits readers who want a dated small-vendor launch.
Value
2023: the about page dates the first public version to January 2023 (aidetector.pro/pages/about, read 10 November 2026).
Best for
Readers who want to know how long the tool has run.
Main limitation
No study in the index tested AI Detector Pro.
Read the AI Detector Pro card
21

Codeleaks

Verdict
Fits readers who want a dated code launch and its dated retreat.
Value
2023: the press release of 9 November 2023 (copyleaks.com, read 9 November 2026). Events: the standalone AI code endpoint deprecated 28 August 2025, the verdict removed 16 December 2025.
Best for
Developers dating a workflow to an API change.
Main limitation
The AI verdict for code no longer exists in the API.
Read the Codeleaks card
22

ElevenLabs Classifier

Verdict
Fits readers who want a dated self-detector from a voice generator.
Value
2023: the launch post of 15 June 2023 (web.archive.org, 30 August 2023, read 10 November 2026).
Best for
Checking whether a clip came from ElevenLabs.
Main limitation
Scores ElevenLabs audio only; no study record.
Read the ElevenLabs Classifier card
23

Ghostbuster

Verdict
Fits researchers who want the Berkeley feature method.
Value
2023: the paper first posted 24 May 2023, NAACL 2024 (arXiv 2305.15047, read 10 November 2026). Event: the demo moved to babbage-002 and davinci-002 after OpenAI deprecated the paper's models in January 2024.
Best for
Readers who want a method with released datasets.
Main limitation
No study record; the 74.7% on 91 TOEFL essays is the authors' own.
Read the Ghostbuster card
24

SynthID

Verdict
Fits readers who want a dated watermark launch.
Value
2023: the DeepMind blog post of 29 August 2023 (deepmind.google, read 9 November 2026). Event: SynthID-Text published in Nature on 23 October 2024.
Best for
Readers checking Google-generated content.
Main limitation
Detects Google's own watermark only.
Read the SynthID card
25

Binoculars

Verdict
Fits researchers who want the newest open method with records.
Value
2024: the paper first posted 22 January 2024, ICML 2024 (arXiv 2401.12070, read 10 November 2026).
Best for
Readers who want RAID and DAMAGE results for a method they can run.
Main limitation
28.23% true positive rate after 19 humanizers in DAMAGE 2025.
Read the Binoculars card
26

Hiya Deepfake Detector

Verdict
Fits readers who want a dated free voice check.
Value
2024: the press release of 17 October 2024 (hiya.com, read 10 November 2026). Event: built on the technology Hiya acquired with Loccus.ai on 25 July 2024.
Best for
A one-second voice check in the browser.
Main limitation
No study record; the 99% is stated for the platform.
Read the Hiya Deepfake Detector card
27

McAfee Deepfake Detector

Verdict
Fits readers who want a dated on-device voice tool.
Value
2024: the press release of 21 August 2024 (mcafee.com, read 10 November 2026). Events: extended to HP Copilot+ PCs on 3 January 2025; the $9.99 launch offer replaced by McAfee plans.
Best for
Alerts while watching video on a supported PC.
Main limitation
No study record; English audio only.
Read the McAfee Deepfake Detector card
28

Pindrop Pulse

Verdict
Fits readers who want a dated enterprise voice launch.
Value
2024: Pulse Inspect launched in Preview on 15 August 2024 (pindrop.com, read 9 November 2026).
Best for
Newsrooms dating a verification workflow.
Main limitation
No study record; the 99% rests on vendor product data.
Read the Pindrop Pulse card
29

Turnitin Clarity

Verdict
Fits readers who want the newest dated launch.
Value
2025: "a paid add-on for all Turnitin Feedback Studio customers starting in the third quarter of 2025" (turnitin.com, read 9 November 2026).
Best for
Institutions weighing the add-on against the detector.
Main limitation
A process record, not a classifier; no study record.
Read the Turnitin Clarity card

How we ranked

Field sorted
The year the vendor page, press release or paper the card cites states the detector launched; a company founding year stands in only where the vendor dates nothing else, and a Wayback Machine capture date is never entered
Direction
Earlier first
Tie rule
More study records first, then name
Qualification
A launch year field that is not empty; a card with none is listed apart with what its pages say and any withdrawal or change of hands the card records
Qualified
29 of 72 detectors

Field sorted: the launch year field on each card, a year stated on the vendor page, press release or paper the card cites, read 8 to 10 November 2026. A vendor's founding year stands in for a detector launch only where the vendor dates nothing else, and the row says which it is; a Wayback Machine first-capture date is never entered, because a capture dates the archive and not the launch (GPT Radar, Grammica, IvyPanda and SEO.ai are in the second table for that reason). Earlier years rank first; ties within a year are broken by study records, then name. A card with no stated year sits in the second table with what its pages say instead. The events column is read from the card's ownership, plans and status fields and names the date and source of each rename, acquisition, closure or withdrawal; a page that redirects to a marketplace listing is recorded as a move, not a withdrawal, unless the vendor says the tool has stopped.

Why numbers differ between sources

A launch can be dated three ways and the index takes the vendor's own. BrandWell's about page says Content at Scale launched in September 2022 and its archived FAQ says the first detector version was created in October 2022; the card carries 2022 and quotes both. Reality Defender and Codequiry date their founding, not a product; Originality.ai dates a beta model; GPTZero and Pangram date a year with no month. Research methods are dated by the paper's first arXiv posting, not its conference, so Binoculars is 2024 (posted January 2024, ICML 2024) and Fast-DetectGPT 2023 (posted October 2023, ICLR 2024). A press report or a review site never sets the year, which is why 43 cards state none: their pages do not say.

The 2027 census tabulates this ranking's years, 29 stated and 43 not, and the Q3 2027 index update records the launch year metric as one of the two fields added in the quarter, both read 10 November 2026.

Questions about when AI detectors launched

Which AI detector launched first?

Codequiry, with a 2017 founding year on its about page (codequiry.com/about, read 9 November 2026), is the earliest dated card; the GPT-2 Output Detector, released on 5 November 2019, is the earliest dated text detector and the only one from before 2021 with a study record.

How many AI detectors launched in 2023?

14 of the 29 dated cards, from GPTZero and Turnitin (4 April 2023) to SynthID (29 August 2023), per the launch year field of each card read 8 to 10 November 2026; 4 launched in 2024 and 1, Turnitin Clarity, in 2025.

How many AI detectors have been withdrawn?

7 of the 72 cards record a withdrawal by the maker, each dated on the card: AI Writing Check (August 2023), Check For AI (June to July 2023), Writefull (between 31 December 2024 and 7 March 2025), Wordtune (404 since 26 July 2025), Writer (22 December 2025), Grammica (between 15 December 2025 and 12 January 2026) and GPT Radar (301 since 14 August 2026). The shut-down question quotes each notice.

Which AI detectors changed hands or changed name?

5 events on the cards read 8 to 10 November 2026: Content at Scale became BrandWell, Optic became AI or Not, Hiya acquired Loccus.ai on 25 July 2024, Passed.ai's team joined Originality.ai as the service closes, and TruthScan and Undetectable AI share Undetectable Inc. as owner; Scribbr and QuillBot sit together under Learneo. The shared owner question maps each.

Which is the shortest-lived AI detector in the index?

AI Writing Check, 7 months: launched in January 2023 and withdrawn in August 2023 by the vendor's own notice (aiwritingcheck.org, read 10 November 2026). Check For AI ran about six months from its Terms of 1 January 2023 to its sunset in July 2023, but states no launch date.

Why do 43 cards state no launch year?

Because the pages each card cites say nothing about when the detector appeared, and a Wayback Machine capture date is not a statement: GPT Radar's first capture of 10 January 2023 and SEO.ai's of 3 October 2023 date the archive, not the launch. A company founding year is entered only where the vendor dates nothing else and the row says so.

Does an older launch mean a better detector?

No. The GPT-2 Output Detector, dated 2019, flagged 0.9503 of 1,992 human passages at threshold 0.5 in Jabarian and Imas 2025 (NBER w34223, read 9 November 2026), and GLTR, also 2019, flagged 99.3% of RAID's human documents at threshold 0.5. The independently tested ranking and the false positive rate ranking sort on results.

How often is this ranking updated?

The ranking was created on 10 November 2026 from the launch year field of 72 cards and changes whenever a card gains a stated year or records a new withdrawal or change of hands. The next scheduled recheck of every card's fields is 8 December 2026 (methodology).

Sources and revisions

Sources

  1. Codequiry about pagecodequiry.com/about
  2. Sensity about pagesensity.ai/about-sensity-ai
  3. OpenAI, GPT-2 1.5B releaseopenai.com/index/gpt-2-1-5b-release
  4. Gehrmann, Strobelt and Rush 2019, GLTRarxiv.org/abs/1906.04043
  5. Deepfake-O-Meter v2.0 platform paperarxiv.org/abs/2404.13146
  6. Reality Defender company pagerealitydefender.com/company
  7. Originality.ai accuracy page and release historyoriginality.ai/blog/ai-accuracy
  8. BrandWell AI Content Detector (archived 9 August 2026)web.archive.org/web/20260809041054/https:...
  9. Illuminarty release post, 18 March 2023illuminarty.ai/en/posts/big-spring-update...
  10. Intel newsroom, 14 November 2022intel.com/content/www/us/en/newsroom/news...
  11. GPTZero home pagegptzero.me
  12. Turnitin launch post, 4 April 2023turnitin.com/blog/the-launch-of-turnitins...
  13. Mitchell et al. 2023, DetectGPTarxiv.org/abs/2301.11305
  14. Pangram about pagepangram.com/about-us
  15. Hu, Chen and Ho 2023, RADARarxiv.org/abs/2307.03838
  16. AI or Not press pageaiornot.com/press
  17. AI Writing Check noticeaiwritingcheck.org
  18. Bao et al. 2023, Fast-DetectGPTarxiv.org/abs/2310.05130
  19. Wordtune blog, 21 August 2023wordtune.com/blog/ai-content-detector
  20. AI Detector Pro about pageaidetector.pro/pages/about
  21. Copyleaks press release, 9 November 2023copyleaks.com/about-us/press-releases/cop...
  22. ElevenLabs AI Speech Classifier launch post (archived 30 August 2023)web.archive.org/web/20230830030350/https:...
  23. Verma et al. 2023, Ghostbusterarxiv.org/abs/2305.15047
  24. Google DeepMind, SynthID announcementdeepmind.google/blog/identifying-ai-gener...
  25. Hans et al. 2024, Binocularsarxiv.org/abs/2401.12070
  26. Hiya press releases, 25 July and 17 October 2024hiya.com/newsroom/press-releases/hiya-acq...
  27. McAfee press release, 21 August 2024mcafee.com/en-us/newsroom/press-releases/...
  28. Pindrop, Pulse Inspect launchpindrop.com/company-news/pindrop-launches...
  29. Turnitin Clarity announcementturnitin.com/blog/turnitin-clarity-bringi...
  30. Scribbr about pagescribbr.com/about-us
  31. Passed.ai closing noticepassed.ai
  32. TruthScan termstruthscan.com/terms
  33. Writer AI Content Detector (archived 9 December 2025)web.archive.org/web/20251209115654/https:...
  34. Writefull GPT Detector (archived 31 December 2024)web.archive.org/web/20241231044606/https:...
  35. Check For AI (archived 14 March 2023)web.archive.org/web/20230314171740/https:...
  36. Grammica AI Detector (archived 31 July 2025)web.archive.org/web/20250731232452/https:...
  37. GPT Radar (archived 12 July 2026)web.archive.org/web/20260712054433/https:...

Revision log

  1. Ranking created from the launch year field of 72 cards at index version 2026
  2. Twelfth-wave pages linked

Corrections log

Related reading