AI text detectors for newsrooms and fact desks
An AI detector for a newsroom is judged on the shortest text it will score and on whether it runs where the desk works, because the item a fact desk checks is a reader submission, a quoted statement, a press release paragraph or a social post, rarely a 1,000-word essay, and the check has to happen in a browser tab or a pipeline, not a paste box visited one text at a time. Two fields decide which cards appear in this view: a text or image card whose api field is a yes or whose integrations name a browser extension, and a stated minimum text length under 500 characters in input_limits. Of the 72 cards, 5 state both on the pages read on 8 and 9 November 2026: QuillBot, Copyleaks, Phrasly, JustDone and AI or Not. Thirty-five more state an API or an extension and no minimum (eight of the nine cards added on 10 November 2026 with the ninth wave among them, and four of the twelve added with the tenth: Trinka and Ircam Amplify state an API, Monica and Passed.ai an extension, and none a floor; Wordtune's 250-character floor belongs to a withdrawn widget with no API or extension, so it stays out: GPTKit's "more than 50 words" and Content Detector AI's "above 250 words" are advice, not requirements, and none of the nine states a floor), and the minimum input ranking sorts the 10 cards that state a floor at all. Two records are read into the view because they describe the two things a desk compares a score against: Fleckenstein 2024 measured how well 289 teachers told ChatGPT essays from student ones, and Pegoraro 2023 ran 20 detection tools and methods against ChatGPT answers to real questions. This page is one of the views under AI detectors by use case; the short texts question reads the same floors for any reader, the marketplaces view adds a bulk route for review platforms, and the full index shows all 72 cards.
- Field 1Stated minimum per check
- Field 2API or browser extension as stated
- Field 3Modalities as stated
- Field 4Source, read date
Length first. A stated minimum is quoted in the vendor's unit and converted at the index's rate of 5.5 characters a word, because Copyleaks states characters and QuillBot states words; a vendor's "recommended" length is not a minimum and sits in the notes. The route column quotes the API or extension as stated, because a desk that checks twenty items an hour needs one that runs on the page or in the content system. The modality column is carried because a fact desk also receives images, and four of the five cards state image detection beside text. Field rules are in the methodology.
Which fields decide fit for a fact desk?
| Detector | Stated minimum per check | API or browser extension as stated | Modalities as stated | Source, read date |
|---|---|---|---|---|
 Phrasly | At least 30 words for accurate results on the web tool (about 165 characters); at least 50 words per Business API request | Business API: up to 15,000 characters per request, CSV bulk scans of up to 100 documents | Text, image | phrasly.ai/ai-detector, API docs, 8 November 2026 |
 AI or Not | Text from 250 characters (about 64 words) to 500,000 characters per API request | REST API v2; MCP server; Live Voice for Zoom | Image, video, voice, music, text | docs.aiornot.com, 9 November 2026 |
 Copyleaks | Minimum 255 characters on the web platform; 350 characters in the browser extension | API with SDKs; browser extension (Chrome, Edge, Firefox); WordPress add-on | Text, code, image, video | copyleaks.com/ai-content-detector, 8 November 2026 |
 JustDone | Minimum 350 characters for reliable results | Chrome Extension | Text | justdone.com/ai-detector, 8 November 2026 |
 QuillBot | Minimum 80 words per scan (about 440 characters) | Chrome and Edge extensions; macOS app; Word add-in | Text, image | quillbot.com/ai-content-detector, 8 November 2026 |
Values as each vendor states them, read 8 and 9 November 2026, sorted by the stated minimum from shortest to longest. Three cards state an API or extension and a minimum at or above 500 characters and are absent: Originality.ai (100 words, about 550 characters), Surfer (100 words) and Smodin (1,000 characters). Sapling's API flags results on text under 500 characters as low_confidence and recommends at least 300 characters, a recommendation rather than a floor, so Sapling sits with the 23 cards that state a route and no minimum: among them Pangram, Winston AI, GPTZero (10,000 characters per scan without an account, no minimum), ZeroGPT (a guidance of 150 to 200 words), Proofademic, Walter Writes, Hive, Undetectable AI (200 words recommended for the API), TruthScan (200 words recommended) and Grammarly.
Why each card is in this view
- AIDI-D-0012Phrasly
the shortest stated floor in the index at 30 words on the web tool and 50 per API request, with a Business API billed at $0.02 per 1,000 words (phrasly.ai/ai-detector, business.phrasly.ai, read 8 November 2026); no study record. The detector page states that text is deleted immediately after analysis.
- AIDI-D-0025AI or Not
text from 250 characters per API request beside image, video, voice and music endpoints, the one card in this table that covers the media a desk receives in the same API (docs.aiornot.com, read 9 November 2026); 1 image record (as Optic, in Ha and Passananti 2024) and no text record. Uploaded media are deleted after inference and the free text tool stores no data.
- AIDI-D-0003Copyleaks
255 characters on the platform and 350 in the extension, an API with SDKs and a WordPress add-on that puts the check inside a content system (copyleaks.com/ai-content-detector, read 8 November 2026); 0 of 42 human papers flagged in Walters 2023, 5 of 10 in Perkins 2024, and a card that lists both text and image detection.
- AIDI-D-0013JustDone
350 characters for reliable results and a Chrome extension, with a first scan that needs no account (justdone.com/ai-detector, read 8 November 2026); no study record, and a vendor claim of 80% overall accuracy with a 10.3% error rate published on its own page.
- AIDI-D-0007QuillBot
80 words per scan, six free scans a day without sign-up under 1,200 words, and extensions for Chrome and Edge (quillbot.com/ai-content-detector, read 8 November 2026); 2 study records (Kar 2024, Alshammari and Rao 2025), a RAID leaderboard entry of 30 January 2025 recorded on the leaderboard question, and no retention terms on the detector pages read.
What the evidence says for short newsroom texts
The human baseline a desk is replacing is weaker than its confidence. Fleckenstein 2024 asked 89 pre-service teachers and 200 experienced teachers to tell student essays from ChatGPT essays: the novices identified 45.1% of the AI essays and the 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; Fleckenstein 2024 record). Gao 2023's blinded reviewers caught 68% of ChatGPT-written medical abstracts and called 14% of the real ones generated (DOI 10.1038/s41746-023-00819-6, read 9 November 2026). An editor's reading is not the control a detector is measured against; the studies measured both and found both wanting.
The detector record on question-and-answer text, the closest class in the index to a quote or a reader comment, is the Pegoraro 2023 sweep. Twenty tools and methods were run against about 10% of 131,512 responses (58,546 human, 72,966 ChatGPT) to 24,322 questions from medicine, finance, open-domain question answering and social platforms, and no method identified more than 47.3% of the ChatGPT text; ZeroGPT was the best online tool at a 45.7% true positive rate with a 92.2% true negative rate, and the authors write that most detectors "are prone to classifying any text as human-written, with a general high TNR of 90% and low TPR" (arXiv 2304.01487, Table I and Conclusion, read 9 November 2026; Pegoraro 2023 record). None of the five cards in the table above was scored in that study. The one indexed test under 50 words is in Jabarian and Imas 2025, whose stub set of 160 human and 160 AI passages per model is read on the short texts question; on full passages the same study found 0.0000 of 1,992 human texts flagged by Pangram and 0.9503 by the 2019 RoBERTa detector (NBER w34223, read 9 November 2026).
Vendors set their own floors for a reason they state. Turnitin's FAQ says that "in shorter documents where there are only a few hundred words, the prediction will be mostly all or nothing"; Sapling's API marks anything under 500 characters low_confidence; Quetext's FAQ says detectors "may struggle with short phrases" (guides.turnitin.com, sapling.ai/docs, quetext.com, read 8 and 9 November 2026). A desk that checks a 60-word quote with a tool whose floor is 165 characters is inside the vendor's stated range and outside every indexed study's. The voice and video question lists the 9 cards that cover both media for a desk receiving clips, and the bulk and batch view the 15 that take many files at once. No detector score should be the only evidence in a publishing decision, and the vendors' own caveats on the cards say so in their words.
The customer support QA view reads the same minimum-input field with a bulk route and a retention policy for agent replies, and the mobile apps view lists the five cards whose integrations name a phone app, both read 10 November 2026.
Sources
- https://phrasly.ai/ai-detector (read 8 November 2026)
- https://business.phrasly.ai/docs/ai-detection-api (read 8 November 2026)
- https://docs.aiornot.com/api-reference/reports-by-modality/image (read 9 November 2026)
- https://copyleaks.com/ai-content-detector (read 8 November 2026)
- https://justdone.com/ai-detector (read 8 November 2026)
- https://quillbot.com/ai-content-detector (read 8 November 2026)
- https://sapling.ai/docs/api/detector/ (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.quetext.com/ai-content-detector (read 9 November 2026)
- https://doi.org/10.1016/j.caeai.2024.100209 (read 9 November 2026)
- https://doi.org/10.1038/s41746-023-00819-6 (read 9 November 2026)
- https://arxiv.org/pdf/2304.01487 (read 9 November 2026)
- https://www.nber.org/papers/w34223 (read 9 November 2026)
Revision log: 2026-11-09: view created at index version 2026.11.92026-11-10: nine ninth-wave cards checked; none states a minimum requirement, so none joins; the no-floor count is 312026-11-10: twelve tenth-wave cards checked; none states both a route and a minimum under 500 characters, so none joins; the no-floor count is 352026-11-10: twelfth-wave pages linked
FAQ: AI detectors for newsrooms
Which AI detectors check a short text through an API or extension?
Five of 72 cards state both an API or browser extension and a minimum under 500 characters, read 8 and 9 November 2026: Phrasly (30 words), AI or Not (250 characters), Copyleaks (255), JustDone (350) and QuillBot (80 words). Twenty-three more state a route and no minimum.
What is the shortest text an indexed detector will score?
Phrasly's web tool states 30 words, about 165 characters, as the floor for accurate results, and its Business API 50 words (phrasly.ai/ai-detector, read 8 November 2026). The minimum input ranking orders all 10 stated floors.
How well do editors tell AI text from human text without a tool?
In Fleckenstein 2024, 89 pre-service teachers identified 45.1% of ChatGPT essays and 200 experienced teachers 37.8%, with confidence near 80% (DOI 10.1016/j.caeai.2024.100209, read 9 November 2026). Gao 2023's blinded reviewers caught 68% of generated abstracts.
Has a study tested detectors on question-and-answer text?
Pegoraro 2023 ran 20 tools and methods on ChatGPT and human answers to 24,322 questions; no method caught more than 47.3% of the ChatGPT text and ZeroGPT led the online tools at 45.7% (arXiv 2304.01487, read 9 November 2026). None of the five cards above was in that study.
Which of the five also checks images or video?
Four: AI or Not (image, video, voice, music and text in one API), Copyleaks (text, code, image, video), QuillBot (text, image) and Phrasly (text, image), as read on 8 and 9 November 2026; JustDone states text only. The text and images question lists all 15 cards covering both.
