AI detectors for marketplaces and review platforms
An AI detector for a marketplace or a review platform is judged on three fields before any other, because a product review is short, arrives in volume and is checked by a machine rather than a person: an API in the api field, a stated minimum input under 300 characters or an explicit statement of how text that short is handled in input_limits, and a bulk or batch route in the same field or in the plans. Of the 72 cards, 39 state an API, 15 state a minimum text length of any kind, and 2 state all three of the deciding fields: Phrasly and Sapling, as read on 8 November 2026; none of the nine cards added on 10 November 2026 states a minimum text length as a requirement, so none joins either table. A second table below lists the six cards that state two of the three, because a platform that can live without one of them has more to choose from. This page is one of the views under AI detectors by use case; the agencies view reads input limits and bulk checking for long copy, the business view adds retention and team seats, and you can start from the full table to see all 72 cards. The short texts question lists every stated minimum in the index.
- Field 1Stated minimum or short-text handling
- Field 2API and per-request limit
- Field 3Bulk or batch route
- Field 4Source, read date
The minimum comes first because a typical review sits between 100 and 600 characters, and a detector that refuses or discounts anything under 300 characters cannot score most of a platform's reviews. The table quotes each vendor's minimum or short-text statement in its own units, because Phrasly states words and Sapling states characters. The API and the bulk route are quoted as stated, with the API's own per-request limits where the vendor publishes them. The false positive evidence on each card appears in the notes, because a flagged review is a seller's complaint; the false positive rate ranking orders every card on that field. Field rules are in the methodology. Of the twelve cards added on 10 November 2026 with the tenth wave, none qualifies: Trinka publishes an API and no minimum or bulk route and joins the second table; Ircam Amplify publishes an API and a 20-file batch for music, not text; Wordtune's 250-character minimum belongs to a withdrawn widget with no API; the other nine state no API (trinka.ai, docs.ircamamplify.io, read 10 November 2026).
Which fields decide fit for a review platform?
| Detector | Stated minimum or short-text handling | API and per-request limit | Bulk or batch route | Source, read date |
|---|---|---|---|---|
 Phrasly | Web tool: "at least 30 words long for accurate results"; Business API: at least 50 words | Yes, business.phrasly.ai; up to 15,000 characters per request; $0.02 per 1,000 words scanned | Bulk scans from a CSV of up to 100 documents (Business API) | phrasly.ai/ai-detector, Business API docs, read 8 November 2026 |
 Sapling | API: at least 300 characters recommended; results on text under 500 characters returned with a low_confidence flag; web tool "more accurate after about 50 words" | Yes, sapling.ai/docs; up to 200,000 characters per request; from $0.005 per 1,000 characters | 5 files per batch (Free), 10 (Pro and Enterprise); API zero_retention: true option | sapling.ai/ai-content-detector, API docs, read 8 November 2026 |
Values as each vendor states them, read 8 November 2026. Phrasly's minimum is stated in words, and 30 words is below 300 characters at the average English word length of about five characters plus a space; the index records the vendor's unit and does not convert it. Sapling states no refusal threshold: text under 500 characters is scored and flagged rather than rejected, which is the short-text handling the view reads.
The six cards below state two of the three fields. Each lacks one in the vendor's own words: Copyleaks states a 255-character minimum and an API but no bulk route on the pages read; ZeroGPT, GPTZero, Quetext and Smodin state an API and batch files but a minimum above 300 characters or none at all; TruthScan states an API and batch uploads with a 200-word recommendation.
Values as each vendor states them on copyleaks.com, zerogpt.com/pricing, gptzero.me/faq (read 8 November 2026), quetext.com/integrations, smodin.io and the TruthScan text API documentation (read 9 November 2026).
Why each card is in this view
- AIDI-D-0012Phrasly
the detector page states "at least 30 words long for accurate results" and "no character limits on detection", and the Business API page states a 50-word minimum, a 15,000-character ceiling per request, $0.02 per 1,000 words scanned and bulk scans from a CSV of up to 100 documents (phrasly.ai/ai-detector, Business API docs, read 8 November 2026). No indexed study tested Phrasly, so the card carries no false positive figure; the detector page says analysed text is "deleted immediately after analysis", and the privacy policy of 29 June 2023 does not mention submitted text.
- AIDI-D-0010Sapling
the API documentation states at least 300 characters recommended and a
low_confidenceflag on results for text under 500 characters, up to 200,000 characters per request and azero_retention: trueoption; the product page says results are more accurate after about 50 words and the free check covers 2,000 characters with 5 files per batch (sapling.ai/ai-content-detector, API docs, read 8 November 2026). Walters 2023 recorded 0 of 42 human papers flagged with 31% returned uncertain, and Sapling's own caveat reads "The shorter the text is, the more general it is, and the more essay-like it is, the more likely it is to result in a false positive" (De Gruyter 10.1515/opis-2022-0158, Table 4).
What the evidence says for short reviews
No study in the index scored product reviews as a text class, so the closest evidence is the record that used reviews among its human texts. Jabarian and Imas 2025 drew its 1,992 human passages from six genres including Amazon reviews and restaurant reviews, all written before 2020, and reported false positive rates at a 0.5 threshold of 0.0000 for Pangram, 0.0011 for Originality.ai and 0.0071 for GPTZero on the whole set, with per-genre rates at Youden-optimised thresholds of at most 0.0075, 0.0218 and 0.0238 (NBER w34223, Tables 3 and 6, read 9 November 2026; Jabarian and Imas 2025 record). None of the three is in the first table: Pangram and GPTZero state no minimum, and Originality.ai's minimum is 100 words.
The second body of evidence is about length itself. Pegoraro 2023, on question-and-answer responses rather than essays, found no online tool with a true positive rate above 45.7% and most tools "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). Short responses are the text form closest to a review, and the result points the same way as the vendors' own minimums: on short text a detector is more likely to miss AI than to flag a person, and a platform that reads a low score as a clean review is reading noise. OpenAI said the same of its own classifier in 2023, that it was "very unreliable" on texts under 1,000 characters (OpenAI classifier record).
The vendors' own caveats finish the point. Sapling says no current detector should be used as a standalone check; ZeroGPT's terms bar using an output as the sole basis for a decision affecting another person's rights; Copyleaks states a 255-character floor because results below it are not meaningful (sapling.ai, zerogpt.com/terms-of-use, copyleaks.com, read 8 November 2026). No detector score should be the only evidence in a misconduct, hiring or publishing decision, and a removed review or a suspended seller is such a decision.
A newsroom checking short texts has a sibling view: the newsrooms view, added 9 November 2026, filters on an API or browser extension and a minimum under 500 characters and finds 5 cards. The bulk and batch view, added the same day, lists every bulk route in the index, 15 cards.
A sibling view adds the retention test to the same short-text fields: the customer support QA view lists the two cards, Phrasly and Sapling, that state an API, a short-text statement, a bulk route and a retention policy together, 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://phrasly.ai/pricing?plan=monthly (read 8 November 2026)
- https://sapling.ai/ai-content-detector (read 8 November 2026)
- https://sapling.ai/docs/api/detector/ (read 8 November 2026)
- https://sapling.ai/pricing (read 8 November 2026)
- https://copyleaks.com/ai-content-detector (read 8 November 2026)
- https://www.zerogpt.com/pricing (read 8 November 2026)
- https://www.zerogpt.com/terms-of-use (read 8 November 2026)
- https://gptzero.me/faq (read 8 November 2026)
- https://www.quetext.com/integrations (read 9 November 2026)
- https://www.quetext.com/developers-api (read 9 November 2026)
- https://smodin.io/ai-content-detector (read 9 November 2026)
- https://truthscan.com/truthscan-ai-text-detection-api-documentation (read 9 November 2026)
- https://www.nber.org/papers/w34223 (read 9 November 2026)
- https://arxiv.org/pdf/2304.01487 (read 9 November 2026)
- https://www.degruyterbrill.com/document/doi/10.1515/opis-2022-0158/pdf (read 8 November 2026)
Revision log: 2026-11-09: view created at index version 2026.11.62026-11-09: minimum word count question linked (seventh wave)2026-11-09: eighth-wave pages linked2026-11-10: nine ninth-wave cards checked; Isgen states an API and a bulk route but no minimum, so no card joins2026-11-10: twelve tenth-wave cards checked, none qualifies; Trinka added to the second table2026-11-10: twelfth-wave pages linked
FAQ: AI detectors for marketplaces and review platforms
Which AI detectors fit checking product reviews at volume?
Two cards state all three deciding fields, an API, a minimum under 300 characters or explicit short-text handling, and a bulk or batch route: Phrasly (30 to 50 words, CSV bulk of 100) and Sapling (text under 500 characters scored with a low-confidence flag, 5 to 10 files per batch), as read on 8 November 2026. Six more state two of the three.
Can an AI detector score a 200-character review?
Sapling scores it and flags it low_confidence; Copyleaks refuses anything under 255 characters; Phrasly asks for 30 words; ZeroGPT recommends 150 to 200 words and Smodin requires 1,000 characters, all as stated on 8 and 9 November 2026. A 200-character review clears Phrasly's floor and Sapling's scoring threshold and no other stated minimum in the index. The minimum word count question, added 9 November 2026, lists every hard minimum and recommendation in the index with the characters converted at 5.5 a word.
What do the stated API rates cost per review batch?
The index does not rank price. The stated API rates are Sapling from $0.005 per 1,000 characters, Phrasly $0.02 per 1,000 words, ZeroGPT $0.042 per 1,000 words and Quetext $0.10 per 1,000 words, each metered in its own unit and read on 8 and 9 November 2026 (sapling.ai/pricing, phrasly.ai/pricing, zerogpt.com/pricing, quetext.com/developers-api).
Has any study tested AI detectors on reviews?
Jabarian and Imas 2025 included Amazon reviews and restaurant reviews among its 1,992 human passages and reported false positive rates of 0.0000 to 0.0071 at a 0.5 threshold for Pangram, Originality.ai and GPTZero (NBER w34223, Table 3, read 9 November 2026). No indexed study scored a review-only set.
Do these detectors keep the reviews you send them?
Phrasly's detector page says text is deleted immediately after analysis; Sapling's API offers zero_retention: true and its policy says content is stored to improve the service otherwise; Copyleaks' policy allows training on data unless a direct customer opts out (phrasly.ai/ai-detector, sapling.ai/docs/api/detector, copyleaks.com, read 8 November 2026). The data retention ranking sorts every stated period.
Should a low score remove a review?
The vendors say no: Sapling and ZeroGPT both bar standalone use in their own words, and Pegoraro 2023 found short text more often missed than flagged. A score is a reason to look at the account's other signals, not a result.
