AI detectors ranked by minimum text length per check
Phrasly states the smallest minimum text length of any detector in the index, 30 words "for accurate results" on its free web tool, about 165 characters at 5.5 characters a word, per the detector page on phrasly.ai, read 8 November 2026. AI or Not states 250 characters per text API request and Copyleaks 255 characters on its web platform, read 9 and 8 November 2026; at the other end Turnitin requires "at least 300 words of prose text in a long-form writing format", about 1,650 characters, per its AI writing detection FAQ on guides.turnitin.com, read 8 November 2026. Twelve of the 72 detectors state a minimum as a requirement (Wordtune at 250 characters joined on 10 November 2026, from the withdrawn widget's validation message, then AI Writing Check at 550 characters from its archived counter); nine more state a recommended length or a code guard and sit in a second table (StealthWriter and Binoculars added the same day, then GPT Radar's advice and Ghostbuster's refusal message), and the rest state no floor.
Updated Limits checked 8 November 2026 for the first sixteen cards and 9 November 2026 for the twenty-six added that day
Short answer
Phrasly states 30 words, about 165 characters, read 8 November 2026; AI or Not 250 characters per API request, read 9 November 2026; Copyleaks 255 characters on the web platform and 350 in the browser extension; JustDone 350 characters; QuillBot 80 words, about 440 characters; Originality.ai and Surfer 100 words each, about 550 characters; Smodin 1,000 characters; Turnitin and Turnitin Clarity 300 words each, about 1,650 characters. Word figures are converted to characters at 5.5 characters a word for ordering only; the vendor's own unit is printed in every row. The withdrawn OpenAI classifier of 2023 required 1,000 characters, a figure that lives in its study record, not on a card. Sources read 8 and 9 November 2026. No detector score should be the only evidence in a misconduct or hiring decision.
Which AI detectors state the smallest minimum text length for a check?
| Rank | Detector | Smallest stated minimum, in characters | As the vendor states it | Source and read date | Best for | Main limitation |
|---|---|---|---|---|---|---|
| 1 | Phrasly | 165 characters | "text at least 30 words long for accurate results" (web tool); at least 50 words per Business API request | phrasly.ai, read 8 November 2026 | Short posts and messages checked free | The same company sells a humanizer; no indexed study tested Phrasly |
| 2 | AI or Not | 250 characters | "text from 250 characters (about 64 words)" per API request | docs.aiornot.com, read 9 November 2026 | Developers scoring short text through an API | The web tool states no text minimum; the one study record scored images, not text |
| 3 | Wordtune | 250 characters | "Not enough text to scan. The minimum text length is 250 characters" (validation message in the widget bundle) | archived widget bundle, 26 November 2024 (read 10 November 2026) | Readers tracing what the withdrawn widget required | The tool is withdrawn (404 since 26 July 2025); the floor is in the shipped JavaScript, not on a page |
| 4 | Copyleaks | 255 characters | "minimum 255 characters" on the web platform; 350 characters in the browser extension | copyleaks.com, read 8 November 2026 | Schools checking a paragraph inside an LMS | 5 of 10 human samples falsely accused in Perkins 2024 |
| 5 | JustDone | 350 characters | "minimum text length for reliable results: 350 characters" | justdone.com, read 8 November 2026 | Students checking a short answer | No study record; plan prices absent from the detector page read |
| 6 | QuillBot | 440 characters | "minimum 80 words per scan" | quillbot.com, read 8 November 2026 | A free check with no account at 80 words | Two 2024 and 2025 records on single texts and mean scores, no rate |
| 7 | Originality.ai | 550 characters | "100-word minimum" | originality.ai/pricing, read 8 November 2026 | Publishers checking short web copy | 19 of 495 pre-2022 human passages flagged in Epoch AI 2026 |
| 8 | AI Writing Check | 550 characters | "100 Minimum words required" (counter on the archived page); 400 words maximum | web.archive.org, 17 July 2023 (read 10 November 2026) | Readers tracing what the withdrawn tool required | The tool has been withdrawn since August 2023; one record with an averaged rate |
| 9 | Surfer | 550 characters | "text must be at least 100 words long for a scan to run" | docs.surferseo.com, read 9 November 2026 | SEO teams checking a short section free | No study record; the same page sells a humanizer |
| 10 | Smodin | 1,000 characters | "the input box requires a 1,000 character minimum" | smodin.io, read 9 November 2026 | A long paragraph checked free without an account | No study record; the highest character floor in the index |
| 11 | Turnitin | 1,650 characters | "at least 300 words of prose text in a long-form writing format" | guides.turnitin.com, read 8 November 2026 | Institutions scoring essays, not answers | Nothing under 300 words is processed; lists, tables and code are excluded |
| 12 | Turnitin Clarity | 1,650 characters | "pasted blocks of 300 or more words receive AI writing results" with the Originality add-on | guides.turnitin.com, read 9 November 2026 | Instructors reading a writing process, not a score | A process record; the AI writing result applies only to pasted blocks |
Stated as a recommendation, not a minimum
| Detector | Vendor wording | Source, read date |
|---|---|---|
| Sapling | "more accurate after about 50 words"; API: "at least 300 characters recommended", results under 500 characters flagged low_confidence | sapling.ai and sapling.ai/docs, read 8 November 2026 |
| ZeroGPT | "at least 150 to 200 words" as guidance | zerogpt.com/pricing, read 8 November 2026 |
| TruthScan | "at least 200 words recommended for best accuracy" (text API) | truthscan.com, read 9 November 2026 |
| Undetectable AI | "at least 200 words recommended for best accuracy" (API) | help.undetectable.ai, read 9 November 2026 |
| BrandWell | a tip to try more than 250 characters, on the archived page of 9 August 2026 | web.archive.org, read 9 November 2026 |
| StealthWriter | "at least 100 words and complete sentences" recommended; results for shorter passages "may be less accurate" | stealthwriter.ai/ai-detector, read 10 November 2026 |
| Binoculars | "We recommend using 200-300 words of text at a time" (Space disclaimer); the Space refuses text under 64 tokens, a code guard rather than a stated minimum | huggingface.co/spaces/tomg-group-umd/Binoculars, read 10 November 2026 |
| GPT Radar | "At least 400 words are needed for reliable results": advice on the archived home page, not a floor the tool enforced (the GPTKit reading); withdrawn in 2026 | web.archive.org, 12 July 2026, read 10 November 2026 |
| Ghostbuster | "Ghostbuster is not reliable for texts less than 100 words. Please try again with a longer text": the demo's refusal message, a code guard rather than a stated requirement (the Binoculars reading); a warning follows between 100 and 250 words | ghostbuster.app, read 10 November 2026 |
Cards that state no minimum
Twenty-seven cards state no minimum text length at index version 2026.11.13. Seven are text detectors whose pages state only a maximum or nothing at all: Grammarly and Scribbr (no input limits field), Winston AI (200,000 characters per scan, no floor), GPTZero (10,000 and 150,000-character ceilings, no floor), Pangram ("minimum and maximum words per text check not published"), Proofademic and Walter Writes (per-request maximums only), all read 8 November 2026; and GLTR, the GPT-2 Output Detector, DetectGPT, AIDetectPlus, Quetext, Writer and Compilatio, read 9 November 2026, where Compilatio's 2023 free tool took 200 to 2,000 characters per Weber-Wulff 2023 Table 2, a historical figure the card records but the ranking does not sort. The remainder detect images, video, audio or code, or record a writing process, so a text minimum does not apply: Hive, WasItAI, Reality Defender, Grammarly Authorship, SynthID, Sightengine, Deepware, Sensity, Codequiry, Codeleaks, Illuminarty, Pindrop Pulse and Resemble Detect.
One of the plan rankings in the rankings hub, the other end of the field the input limits ranking sorts. The home board answers which AI detector checks the shortest text? and the question page on AI detectors for short texts walks the same floors for one reader. The minimum word count question, added 9 November 2026, reads the same field and separates the ten hard minimums from the five recommendations.
Which AI detectors should you check first?
- Phrasly, rank 1Best for: Short posts and messages checked free. Main limitation: The same company sells a humanizer; no indexed study tested Phrasly.
- AI or Not, rank 2Best for: Developers scoring short text through an API. Main limitation: The web tool states no text minimum; the one study record scored images, not text.
- Wordtune, rank 3Best for: Readers tracing what the withdrawn widget required. Main limitation: The tool is withdrawn (404 since 26 July 2025); the floor is in the shipped JavaScript, not on a page.
- Copyleaks, rank 4Best for: Schools checking a paragraph inside an LMS. Main limitation: 5 of 10 human samples falsely accused in Perkins 2024.
- JustDone, rank 5Best for: Students checking a short answer. Main limitation: No study record; plan prices absent from the detector page read.
What does each stated minimum mean in practice?

Phrasly
- Verdict
- Fits readers who want to check a social post or a short message without paying.
- Value
- 165 characters, the index's conversion of "text at least 30 words long for accurate results" at 5.5 characters a word; the Business API documentation states at least 50 words per request (phrasly.ai, business.phrasly.ai, read 8 November 2026).
- Best for
- Moderators scoring short posts at volume through the API, read beside the marketplaces view.
- Main limitation
- Phrasly sells a humanizer on the same site and no record in the studies index tested the detector.

AI or Not
- Verdict
- Fits developers who need a stated floor in characters for an API pipeline.
- Value
- 250 characters, "about 64 words", the stated lower bound of the text endpoint whose upper bound is 500,000 characters (docs.aiornot.com, read 9 November 2026).
- Best for
- Teams scoring reviews or comments in bulk.
- Main limitation
- The web tool states no text minimum, and the one indexed record, Ha and Passananti 2024, scored images under the Optic name (arXiv 2402.03214, read 9 November 2026).

Copyleaks
- Verdict
- Fits schools that check a paragraph rather than a paper.
- Value
- 255 characters on the web platform and 350 in the browser extension, with the web platform taking up to 2,000 pages (copyleaks.com, read 8 November 2026).
- Best for
- LMS users who paste a short discussion-board answer.
- Main limitation
- Perkins 2024 recorded 5 false accusations in 10 human samples, and Alshammari and Rao 2025 saw the mean AI score on DeepSeek text fall from 99.7% to 71% after a humanized paraphrase (Perkins 2024, Alshammari and Rao 2025).

JustDone
- Verdict
- Fits a short student answer checked on a paid plan.
- Value
- 350 characters, stated as the "minimum text length for reliable results" (justdone.com, read 8 November 2026).
- Best for
- Students checking one answer at a time.
- Main limitation
- No indexed study tested JustDone, and the detector page read on 8 November 2026 carries no plan prices.

QuillBot
- Verdict
- Fits a free check on a short passage with no account.
- Value
- 440 characters, the conversion of "minimum 80 words per scan"; the free tier takes up to 1,200 words a scan (quillbot.com, read 8 November 2026).
- Best for
- Students inside the QuillBot suite.
- Main limitation
- The two records, Kar 2024 and Alshammari and Rao 2025, print single-text scores and mean scores, so no false positive or false negative rate exists for QuillBot (Kar 2024, Alshammari and Rao 2025).

Originality.ai
- Verdict
- Fits publishers who check short web copy and meta text.
- Value
- 550 characters, the conversion of the "100-word minimum" stated on the pricing page, where 1 credit buys 100 words (originality.ai/pricing, read 8 November 2026).
- Best for
- Editors scanning product descriptions.
- Main limitation
- Epoch AI 2026 flagged 19 of 495 pre-2022 human passages (3.84%), 16 of them fiction (Epoch AI 2026).

Surfer
- Verdict
- Fits SEO teams checking a section of a draft free.
- Value
- 550 characters, the conversion of "text must be at least 100 words long for a scan to run" (docs.surferseo.com, read 9 November 2026).
- Best for
- Content teams already in Surfer.
- Main limitation
- No indexed study tested Surfer, and the Humanizer on the same page is listed as a DAMAGE 2025 L2 humanizer (DAMAGE 2025).

Smodin
- Verdict
- Fits a long paragraph checked free without an account.
- Value
- 1,000 characters, the input box's stated minimum, under a 50,000-character ceiling (smodin.io, read 9 November 2026).
- Best for
- Bloggers checking a full section at once.
- Main limitation
- The highest character floor in the index, so a short answer cannot be checked; no study record.

Turnitin
- Verdict
- Fits institutions that score essays and not short answers.
- Value
- 1,650 characters, the conversion of "at least 300 words of prose text in a long-form writing format"; the minimum was raised from 150 to 300 words in May 2023 per Turnitin's own post (guides.turnitin.com, read 8 November 2026; turnitin.com, read 9 November 2026).
- Best for
- Universities with a Turnitin Originality licence.
- Main limitation
- A submission under 300 words is not processed, and lists, headers, references, tables, formulas and code are excluded from scoring.

Turnitin Clarity
- Verdict
- Fits instructors who want a writing process record with an AI result on long pasted blocks only.
- Value
- 1,650 characters, the conversion of "pasted blocks of 300 or more words receive AI writing results when the Originality add-on is licensed" (guides.turnitin.com, read 9 November 2026).
- Best for
- Courses writing inside the Clarity space in English, Japanese or Spanish.
- Main limitation
- Clarity records a process rather than returning a score, so the 300-word figure governs one feature, not the product.

Wordtune
- Verdict
- Fits readers who want to know the floor a withdrawn tool enforced.
- Value
- 250 characters: the widget's validation message read "Not enough text to scan. The minimum text length is 250 characters" and its ceiling 5,000 characters (archived widget bundle, 26 November 2024, read 10 November 2026).
- Best for
- Readers reading a 2023 or 2024 Wordtune result on a short text.
- Main limitation
- The page returned 404 on 10 November 2026 and in Wayback captures since 26 July 2025, so the floor is a record, not a live requirement (web.archive.org, 30 March 2025, read 10 November 2026).

AI Writing Check
- Verdict
- Fits readers who want the stated floor of a tool that no longer runs.
- Value
- 100 words, 550 characters at 5.5 characters a word: the paste box counters read "100 Minimum words required", "Current words" and "400 Max words" (web.archive.org, 17 July 2023, read 10 November 2026).
- Best for
- Readers reading a 2023 verdict on a short student response.
- Main limitation
- The notice has read "AI Writing Check is no longer available" since August 2023 (aiwritingcheck.org, read 10 November 2026); the one record, Liang 2023, prints an averaged rate across seven tools.
How we ranked
- Field sorted
- Smallest text length the vendor states as a requirement for a check ("minimum", "at least", "must", or a result returned only above the figure), normalised to characters at 5.5 characters a word for ordering only, the vendor's unit printed beside it; a figure the vendor calls recommended, guidance or a tip is tabled apart
- Direction
- Lower first
- Tie rule
- More study records first, then name
- Qualification
- A vendor page that states a numeric minimum as a requirement; a card that states only a recommendation or no floor is listed apart
- Qualified
- 12 of 72 detectors
Field sorted: the minimum input field on each card, written on 9 November 2026 from the input limits text that was read on 8 November 2026 for the first sixteen cards and 9 November 2026 for the twenty-six added that day. A figure qualifies when the vendor states it as a requirement ("minimum", "at least", "must", "requires", or a result returned only above the figure); a figure the vendor calls recommended, guidance or a tip does not qualify and sits in the second table. Where a vendor states two floors, the smaller one is sorted and both are printed. Word figures are converted to characters at 5.5 characters a word for ordering only, so a 30-word floor sorts as 165 characters, and the vendor's unit is printed beside it. Direction: smaller ranks higher, because the page answers which detector will check the shortest text. Tie rule: more study records first, then product name, so Originality.ai (7 records) sits above Surfer (0) at 550 characters and Turnitin (5) above Turnitin Clarity (0) at 1,650. 12 of 72 qualify; 9 state a recommendation or a code guard; 51 state no minimum. Full rules: methodology page. No record in the studies index measures a detector's behaviour at its minimum length; the OpenAI classifier record notes that OpenAI called its 2023 classifier "very unreliable" below 1,000 characters, the only vendor statement of reliability by length in the index (OpenAI classifier 2023).
Why numbers differ between sources
Vendors state floors in different units on different pages, and some state two. Copyleaks states 255 characters for the web platform and 350 for the browser extension on one page (copyleaks.com, read 8 November 2026). Phrasly states 30 words on the detector page and 50 words in the Business API documentation (phrasly.ai, business.phrasly.ai, read 8 November 2026). Sapling states "about 50 words" on the web page and "at least 300 characters recommended" in the API documentation, both as recommendations (sapling.ai, sapling.ai/docs, read 8 November 2026). The 5.5 characters a word conversion is the rate every page of the index uses since 9 November 2026; the input limits ranking converted at 6 characters a word until that date and now prints the same rate, so the two pages give one character figure for the same word count. All read 8 and 9 November 2026; the Updated line is the date the page last changed.
The newsrooms view, added 9 November 2026, filters these floors by a second field, an API or browser extension, and finds 5 cards under 500 characters that state both.
Questions about AI detector minimum text lengths
Which AI detector checks the shortest text?
Phrasly, at 30 words "for accurate results" on its web tool, about 165 characters (phrasly.ai, read 8 November 2026). In characters, AI or Not's 250 per API request and Copyleaks' 255 on the web platform are the next two floors (docs.aiornot.com, read 9 November 2026; copyleaks.com, read 8 November 2026).
Can an AI detector check a tweet?
A 280-character post clears three stated floors, Phrasly's 165 characters, AI or Not's 250 and Copyleaks' 255, and falls under the other seven, from JustDone's 350 characters to Turnitin's 300 words (justdone.com, read 8 November 2026). OpenAI called its own 2023 classifier "very unreliable" under 1,000 characters (OpenAI classifier 2023), and no indexed study scored any detector on texts that short.
Why is a 30-word minimum ranked as 165 characters?
Because the ranking sorts one unit. Word floors are multiplied by 5.5, an average English word length with its following space, to place them on the same line as character floors; the vendor's words and characters are printed in every row, and the conversion is used for nothing else (methodology).
Which detectors state a recommended length instead of a minimum?
Five: Sapling (about 50 words; 300 characters recommended for the API), ZeroGPT (150 to 200 words as guidance), TruthScan and Undetectable AI (200 words recommended for the API) and BrandWell (a tip to try more than 250 characters), per pages read 8 and 9 November 2026. A recommendation describes where the vendor trusts its own result; it does not stop a shorter check running. GPT Radar's "At least 400 words are needed for reliable results" and Ghostbuster's refusal message under 100 words joined the second table on 10 November 2026, the first as advice and the second as a code guard.
Does Turnitin check anything under 300 words?
No. Turnitin's FAQ requires "at least 300 words of prose text in a long-form writing format", and its May 2023 update records the floor rising from 150 to 300 words (guides.turnitin.com, read 8 November 2026; turnitin.com, read 9 November 2026).
Is a lower minimum better?
A lower minimum means the tool will return a result on shorter text, not that the result is sound. The false positive rate ranking and the false negative rate ranking carry the measured rates, all on texts of paragraph length or longer, so no indexed figure describes accuracy at 165 or 250 characters.
Which cards state no minimum at all?
Twenty-seven at index version 2026.11.13: Grammarly, Scribbr, Winston AI, GPTZero, Pangram, Proofademic, Walter Writes, GLTR, the GPT-2 Output Detector, DetectGPT, AIDetectPlus, Quetext, Writer and Compilatio among text tools, and thirteen media, code and process-record cards where a text floor does not apply, read 8 and 9 November 2026 (AI detector register).
How often is this ranking updated?
The ranking was created on 9 November 2026 from input limits read on 8 and 9 November 2026 and changes whenever a vendor's minimum wording changes on the page recorded on the card. The next scheduled recheck of every card is 8 December 2026 (methodology).
Sources and revisions
Sources
- Phrasly detector pagephrasly.ai/ai-detector
- Phrasly Business API documentationbusiness.phrasly.ai/docs/ai-detection-api
- AI or Not text API documentationdocs.aiornot.com/api-reference/reports-by...
- Copyleaks detector pagecopyleaks.com/ai-content-detector
- JustDone detector pagejustdone.com/ai-detector
- QuillBot detector pagequillbot.com/ai-content-detector
- Originality.ai pricingoriginality.ai/pricing
- Surfer help centre, humanizer and AI detectordocs.surferseo.com/en/articles/9396434-su...
- Smodin detector pagesmodin.io/ai-content-detector
- Turnitin AI writing detection FAQsguides.turnitin.com/hc/en-us/articles/284...
- Turnitin, AI writing detection update, 23 May 2023turnitin.com/blog/ai-writing-detection-up...
- Turnitin Clarity instructor FAQsguides.turnitin.com/hc/en-us/articles/371...
- Sapling detector page and API docssapling.ai/ai-content-detector
- ZeroGPT pricingzerogpt.com/pricing
- TruthScan text API documentationtruthscan.com/truthscan-ai-text-detection...
- Undetectable AI detector API articlehelp.undetectable.ai/en/article/detector-...
- BrandWell detector page, archived 9 August 2026web.archive.org/web/20260809041054/https:...
- OpenAI, new AI classifier for indicating AI-written textopenai.com/index/new-ai-classifier-for-in...
- Ha and Passananti 2024arxiv.org/abs/2402.03214
Revision log
- Ranking created from the minimum input field written at index version 2026
- The one-rate note updated: the input limits ranking now converts at 5
- Eighth-wave pages linked
- Twelve wave 10 cards added
- Nine wave 11 cards added
- Sentence-level highlighting question linked (eleventh wave)
). 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.: register plate](/img/independently-tested-480.webp)


