Is there an AI detector for short texts? Every stated minimum length in the index
Phrasly states the lowest minimum text length of any AI detector in the index, 30 words "for accurate results" on its free web tool (phrasly.ai, read 8 November 2026), and 15 of the 72 detectors state a minimum or a recommended floor at all, by the input limits field of every card read on 8 and 9 November 2026. The highest floors are Turnitin's 300 words of prose and Turnitin Clarity's 300-word pasted block (guides.turnitin.com, read 8 November 2026), and the one vendor to publish a reliability figure by length, OpenAI, called its own 2023 classifier "very unreliable" on texts under 1,000 characters before withdrawing it (openai.com, read 8 November 2026). One indexed study, Jabarian and Imas 2025, scored stubs under 50 words (Table B.1, read 9 November 2026), and no other record scored texts under 250 words, so a short-text result rests on vendor floors and that one test; the minimum word count question, added 9 November 2026, separates the hard minimums from the recommendations.
A minimum is the vendor's own floor for a check to run or to be trusted, as the card's input limits field quotes it from the vendor page, with the URL and the date read. 10 of the 15 are stated requirements (the tool refuses or warns below the figure) and 5 are recommendations ("for best accuracy", "more accurate after"). Character figures are converted to words only as "about N words" at 5.5 characters a word, the one rate every page of the index uses since 9 November 2026 (the page converted at 6 before that date). The table lists the 15 cards with a floor, shortest first; a second table gives the 23 cards that state none, in the vendor's words. This page is one of the questions answered from the index and reads the input limits field of every card on the AI detector register; since 9 November 2026 the ten stated requirements are also sorted on the minimum input ranking, smallest first, with word floors converted to characters at 5.5 characters a word; it reads the same field as the input limits ranking, which sorts the other end of the field, the largest text a vendor accepts, and the free tiers ranking says which of these checks cost nothing.
| Detector | Stated minimum text length | Kind | Source | Read date |
|---|---|---|---|---|
| Turnitin Clarity | 300 or more words in a pasted block, for AI writing results with the Originality add-on | Stated requirement | guides.turnitin.com/hc/en-us/articles/371... | 9 November 2026 |
| Phrasly | 30 words for accurate results on the web tool; 50 words on the Business API | Stated requirement | phrasly.ai/ai-detector | 8 November 2026 |
| BrandWell | Tip to try more than 250 characters (archived page, 9 August 2026) | Recommendation | moxby.com/marketplace/ai-content-detector | 9 November 2026 |
| Copyleaks | 255 characters on the web platform; 350 characters in the browser extension | Stated requirement | copyleaks.com/ai-content-detector | 8 November 2026 |
| Sapling | More accurate after about 50 words; API: at least 300 characters recommended, under 500 characters flagged low_confidence | Recommendation | sapling.ai/ai-content-detector | 8 November 2026 |
| JustDone | 350 characters for reliable results | Stated requirement | justdone.com/ai-detector | 8 November 2026 |
| AI or Not | 250 characters, about 64 words (text API) | Stated requirement | docs.aiornot.com/api-reference/reports-by... | 9 November 2026 |
| QuillBot | 80 words per scan | Stated requirement | quillbot.com/ai-content-detector | 8 November 2026 |
| Originality.ai | 100 words | Stated requirement | originality.ai/pricing | 8 November 2026 |
| Surfer | 100 words for a scan to run | Stated requirement | docs.surferseo.com/en/articles/9396434-su... | 9 November 2026 |
| Smodin | 1,000 characters in the input box | Stated requirement | smodin.io/ai-content-detector | 9 November 2026 |
| TruthScan | At least 200 words recommended (text API) | Recommendation | truthscan.com/truthscan-ai-text-detection... | 9 November 2026 |
| Undetectable AI | At least 200 words recommended (API) | Recommendation | help.undetectable.ai/en/article/detector-... | 9 November 2026 |
| ZeroGPT | At least 150 to 200 words (guidance) | Recommendation | zerogpt.com/pricing | 8 November 2026 |
| Turnitin | 300 words of prose text | Stated requirement | guides.turnitin.com/hc/en-us/articles/284... | 8 November 2026 |
Which AI detectors state the lowest minimum text length?
Phrasly, at 30 words on the web tool and 50 on its Business API (phrasly.ai, read 8 November 2026); Sapling, which says results are "more accurate after about 50 words" and recommends at least 300 characters on its API, flagging anything under 500 characters low_confidence (sapling.ai, read 8 November 2026); AI or Not, whose text API takes text from 250 characters, about 64 words (docs.aiornot.com, read 9 November 2026); and QuillBot at 80 words per scan (quillbot.com, read 8 November 2026). Copyleaks states 255 characters on its web platform, about 46 words, and 350 in its browser extension (copyleaks.com, read 8 November 2026). A floor is where the vendor lets the check run, not where the vendor says it is accurate: Phrasly's 30 words carry the qualifier "for accurate results", and the Phrasly card has no study record at any length.
The marketplaces and review platforms view, added 9 November 2026, reads these minimums for one reader: a platform scoring product reviews at volume, which needs an API and a bulk route beside a minimum under 300 characters, a combination only Phrasly and Sapling state.
Which AI detectors refuse short texts outright?
Turnitin requires at least 300 words of prose in a long-form writing format and does not process a submission below it (guides.turnitin.com, read 8 November 2026); Turnitin Clarity shows AI writing results only on a pasted block of 300 or more words when the Originality add-on is licensed (guides.turnitin.com, read 9 November 2026). Smodin's input box requires 1,000 characters, about 182 words (smodin.io, read 9 November 2026); Surfer's help centre says text "must be at least 100 words long for a scan to run" (surfer.ai, read 9 November 2026); Originality.ai sets a 100-word minimum (originality.ai, read 8 November 2026); JustDone names 350 characters for reliable results (justdone.com, read 8 November 2026). A tweet, a text message or a one-line email answer falls under every one of these floors.
What do the vendors and the studies say about short text?
OpenAI's announcement of its 2023 classifier is the one vendor statement in the index that puts a number on length: the classifier was "very unreliable" on texts under 1,000 characters, reliability "typically improved as input length increased", and the tool identified 26% of AI-written text while mislabelling human text 9% of the time before it was withdrawn in July 2023 (openai.com, read 8 November 2026; OpenAI classifier record). Smodin's detector page says a passage that looks "brief, standardized, and mechanically polished" can be labelled machine-written (smodin.io, read 9 November 2026), and the Moxby listing for BrandWell's detector says human text can be flagged "especially when it is short, technical, templated, highly edited" (moxby.com, read 9 November 2026).
The indexed studies tested long text. Weber-Wulff 2023 wrote its human documents at about 10,000 characters and set a 2,000-character minimum for its ChatGPT documents (arXiv 2306.15666, read 8 November 2026; Weber-Wulff 2023 record); Perkins 2024 started from 15 AI texts of about 500 words each (arXiv 2403.19148, read 8 November 2026; Perkins 2024 record); Chaka 2023 ran 21 whole chatbot responses and 30 translations and prints no word counts (DOI 10.37074/jalt.2023.6.2.12, read 9 November 2026; Chaka 2023 record). One of the 35 records scored detectors on texts under 250 words: Jabarian and Imas 2025 recorded false positive rates of at most 0.0250 for Pangram and 0.0333 for GPTZero on 160 human stubs under 50 words per model, and excluded Originality.ai from that test because its API returned a length filter error (NBER w34223, Table B.1, read 9 November 2026; Jabarian and Imas 2025 record). No other record scored texts under 250 words, so the index holds no short-text false positive or false negative rate for any other card, and the ranking by independent testing says nothing about length.
Cards that state no minimum
| Detector | What the input limits field says instead | Source | Read date |
|---|---|---|---|
| Grammarly | No input limits field; the detector page states no minimum | grammarly.com/ai-detector | 8 November 2026 |
| Scribbr | No input limits field; the detector page states no minimum | scribbr.com/ai-detector | 8 November 2026 |
| Winston AI | States a 200,000-character maximum and no minimum | gowinston.ai/pricing | 8 November 2026 |
| Pangram | "Minimum and maximum words per text check not published" | pangram.com | 8 November 2026 |
| GPTZero | States 10,000 and 150,000-character maximums and no minimum | gptzero.me/faq | 8 November 2026 |
| Proofademic | States per-request maximums (3,500 to 25,000 words) and no minimum | proofademic.ai/ai-text-detector | 8 November 2026 |
| Walter Writes | States per-request maximums (750 to 2,000 words) and no minimum | walterwrites.ai/pricing | 8 November 2026 |
| Hive | Text processed in 2,048-character chunks; no minimum | docs.thehive.ai/docs/supported-file-formats | 8 November 2026 |
| GLTR | Any pasted text; no minimum | gltr.io | 9 November 2026 |
| AIDetectPlus | A 3,000-character widget box; no minimum | aidetectplus.com | 9 November 2026 |
| GPT-2 Output Detector | Evaluated on 510-token examples; the web UI states no minimum | huggingface.co/openai-community/roberta-b... | 9 November 2026 |
| Codequiry | Source code per submission; no minimum | codequiry.com/detect-ai-written-code | 9 November 2026 |
| WasItAI | Images only | wasitai.com | 9 November 2026 |
| Reality Defender | Media and .txt files by size; no text minimum | docs.realitydefender.com/sdks/quickstart | 9 November 2026 |
| Writer | Up to 5,000 words per check on the archived page; no minimum | web.archive.org/web/20251209115654/https:... | 9 November 2026 |
| Grammarly Authorship | A writing-process record; no text is submitted | support.grammarly.com/hc/en-us/articles/2... | 9 November 2026 |
| SynthID | Media files only | synthid.com | 9 November 2026 |
| Compilatio | Credits of 250 words; no minimum published (the 2023 free tool took 200 to 2,000 characters, Weber-Wulff 2023 Table 2) | compilatio.net/en/product/studium | 9 November 2026 |
| Quetext | 1,000 words per free scan; the FAQ says detectors "may struggle with short phrases" but states no minimum | quetext.com/integrations | 9 November 2026 |
| DetectGPT | No product limits; passages in the paper are roughly paragraph length | arxiv.org/pdf/2301.11305 | 9 November 2026 |
| Sightengine | Images, video and audio only | sightengine.com/pricing | 9 November 2026 |
| Deepware | Video only | deepware.ai/us/faq/index.html | 9 November 2026 |
| Sensity | Video, image and audio only | docs.sensity.ai/resources/formats-limits | 9 November 2026 |
A fact desk checking a tip or a quote reads the same floors with a second condition, an API or browser extension: the newsrooms view, added 9 November 2026, finds 5 cards stating both.
Questions about AI detectors and short texts
What is the shortest text an AI detector will check?
Phrasly states 30 words "for accurate results" and Sapling about 50 words, per their pages read 8 November 2026 (phrasly.ai, sapling.ai); Copyleaks takes 255 characters on the web, about 46 words (copyleaks.com, read 8 November 2026). 23 cards state no floor at all and will accept a sentence, which is not a statement about accuracy.
Can an AI detector check a tweet or a text message?
Most will run it and none has a published result for it. A 280-character tweet is about 47 words, under Phrasly's 30-word floor only; QuillBot (80 words), Originality.ai and Surfer (100), Smodin (1,000 characters) and Turnitin (300 words) would refuse or warn, per the vendor pages read 8 and 9 November 2026. OpenAI's 2023 classifier was "very unreliable" under 1,000 characters (openai.com).
Why do AI detectors need a minimum length?
The vendors that explain it say the same thing: Sapling flags results on under 500 characters low_confidence and says the tool is "more accurate after about 50 words" (sapling.ai, read 8 November 2026); QuillBot's page says "the longer the text the better the results" (quillbot.com, read 8 November 2026). The index records the statements and tests none of them.
Does any study test AI detectors on short text?
No indexed study does. The shortest texts in any record are Perkins 2024's 15 AI texts of about 500 words (arXiv 2403.19148, read 8 November 2026); Weber-Wulff 2023 used documents of 2,000 characters and up (arXiv 2306.15666, read 8 November 2026). The 35 records in the studies index all score passages longer than a paragraph.
Which free AI detector has the lowest minimum?
Phrasly, free with no limit on checks and a 30-word floor (phrasly.ai, read 8 November 2026), and Sapling's free check, which covers the first 2,000 characters and is "more accurate after about 50 words" (sapling.ai, read 8 November 2026). The free tiers ranking sets out what each free plan allows.
Sources
All read 8 November 2026 unless a later date is shown; the input limits source of every card is linked in the tables.
- Phrasly AI detector: https://phrasly.ai/ai-detector
- Sapling AI detector: https://sapling.ai/ai-content-detector
- QuillBot AI detector: https://quillbot.com/ai-content-detector
- Copyleaks AI detector: https://copyleaks.com/ai-content-detector
- Originality.ai pricing: https://originality.ai/pricing
- JustDone AI detector: https://justdone.com/ai-detector
- Turnitin AI writing detection FAQs: https://guides.turnitin.com/hc/en-us/articles/28477544839821
- Turnitin Clarity instructor FAQs: https://guides.turnitin.com/hc/en-us/articles/37198804513165-FAQs-for-Instructors-using-Turnitin-Clarity (read 9 November 2026)
- Smodin AI detector: https://smodin.io/ai-content-detector (read 9 November 2026)
- Surfer AI detector: https://surfer.ai/ai-content-detector/ (read 9 November 2026)
- AI or Not text API: https://docs.aiornot.com/api-reference/reports-by-modality/image (read 9 November 2026)
- Moxby AI Content Detector Mod (BrandWell): https://moxby.com/marketplace/ai-content-detector/ (read 9 November 2026)
- OpenAI, New AI classifier for indicating AI-written text (31 January 2023): https://openai.com/index/new-ai-classifier-for-indicating-ai-written-text/
- Weber-Wulff et al. 2023: https://arxiv.org/pdf/2306.15666
- Perkins et al. 2024: https://arxiv.org/pdf/2403.19148
- Chaka 2023: https://doi.org/10.37074/jalt.2023.6.2.12 (read 9 November 2026)
Revision log: 2026-11-09: page created from the input limits field of 38 cards at index version 2026.11.5; 2026-11-09: marketplaces and review platforms view linked (fifth wave); 2026-11-09: link to the minimum input ranking added; 2026-11-09: character conversions moved from 6 to 5.5 characters a word (Copyleaks 255 characters about 46 words, Smodin 1,000 about 182) under the one-rate rule, and the Jabarian and Imas 2025 stub test under 50 words added to the studies paragraph and the lead (seventh wave); 2026-11-09: eighth-wave pages linked
Related pages: Which AI detectors claim to detect ChatGPT? The generators field of all 38 cards, What AI detector does College Board use? What the AP and SAT policy pages state, AI detectors ranked by independent testing Two records added in the fifth wave on 9 November 2026, Pegoraro 2023 and Fleckenstein 2024, changed nine counts and none: Pegoraro 2023 tested ZeroGPT, GPTZero, Copyleaks, Originality.ai, Writer, Content at Scale (BrandWell), DetectGPT, the GPT-2 Output Detector and GLTR, lifting GPTZero to 12, the GPT-2 Output Detector to 10, Copyleaks to 7 and Originality.ai and ZeroGPT to 6; Fleckenstein 2024 tested 289 teachers and no software, so it counts for no card ([Fleckenstein 2024 record](/studies/fleckenstein-2024-teachers/)). Illuminarty, one of four cards added in the same wave, joined with its one image record from Ha and Passananti 2024, so 18 of 72 cards now qualify. and Questions about AI detectors, answered from the index.