Questions about AI detectors, answered from the index

A question page answers one literal question about AI detectors in its first sentence, with a number, the source of the number and the date the source was read. Twenty-seven questions are answered as of 9 November 2026, sixteen of them before the seventh wave of that day: whether a 100% accurate AI detector exists, whether a legit one exists, whether a free one with no limit exists, whether JustDone's detector is legit or accurate, how many AI detectors exist, which AI detectors share an owner, which were shut down, whether detectors store the text you paste, whether they work on machine-translated text, which vendors claim under 1% false positives and who checked, and whether an AI detector is itself an AI. Four more were added with the fourth wave of cards later on 9 November 2026: whether an AI detector works on short texts, which detectors claim to detect ChatGPT, what AI detector College Board uses, and which detectors support Spanish. Each answer is built from the same two places as every other page on this site, the 72 cards in the detector register and the 35 records in the studies index, and nothing on a question page is measured here or estimated here. Where the index has no figure, the page says what the vendor states and stops. Eight more were added in the eighth wave on 9 November 2026, bringing the hub to thirty-five: which detectors accept Google Docs, which appear on a public leaderboard, how detectors handle bullet points, tables and code, which were in Weber-Wulff 2023 and where they are now, which cover voice and video together, which publish a model card, what languages Turnitin's detection covers, by date and which let you delete your data.

The shape is fixed so that an answer can be quoted whole. The first sentence carries the answer and the number; a table from the index follows before the second heading, with the same fields for every tool or study on the page; three or more headings are phrased as questions; three to five short questions close the page, each answered in its first clause; and a Sources list gives every URL used with its read date. One caveat appears on every page: no detector score should be the only evidence in a misconduct, hiring or publishing decision. A revision log records the month the page was created and every change after.

The answers lean on the study records, because the records are where the field's ceiling and floor are printed. Walters 2023 reports that Copyleaks and Turnitin classified all 126 documents correctly, the only perfect score in the index, and notes that Turnitin may have seen the human essays in training (DOI 10.1515/opis-2022-0158). Weber-Wulff 2023 reports that all 14 detectors scored under 80% on averaged accuracy, with Turnitin first at 41 of 54 documents (DOI 10.1007/s40979-023-00146-z). A question about accuracy is answered with both figures and their bases, never with one of them alone.

Vendor statements appear in the vendor's own words, attributed and dated, and are kept apart from study results. GPTZero states an accuracy rate of 99.9% on its FAQ; the RAID paper's Table 5 reports 98.8% and 9.4% for GPTZero in two settings; Weber-Wulff 2023 reports 29 of 54. A question page prints all three with their sources, and the false positive rate ranking and the ranking by independent testing hold the sorted versions of the same fields.

Questions are chosen from the register the editors keep of what readers ask search engines and AI assistants about detectors, and each page carries a register ID in the AIDI-Q series. A question is added when the index can answer it from published fields and no page on this site already does. The methodology page holds the source rules, and the rankings hub holds the computed pages the answers cite.

On 10 November 2026 the tenth wave of cards changed four answers without adding a question: the Weber-Wulff roster now links 12 of the 14 tools to a card, with Check For AI and Writefull recorded as withdrawn; the shut-down question gained four rows (Wordtune, Check For AI, Writefull, Passed.ai); the shared-owner question gained twelve owners and three lineage rows; and the voice and video question gained Deepfake-O-Meter in its main table and the ElevenLabs classifier and FakeCatcher in its second.

The eleventh wave, later the same day, added one question and changed three answers: Which AI detectors offer sentence-level highlighting instead of a single score? reads 17 cards whose pages state a sentence, passage or word-level mark, from GLTR's four-colour word map to Turnitin's retained segments; the shut-down question gained GPT Radar, Grammica and AI Writing Check; the voice and video question gained Hiya and McAfee in its voice-only table; and the offline question gained RADAR as a local run and McAfee as the index's first on-device tool. The launch year ranking, added the same day, dates each withdrawal the shut-down question quotes.

The twelfth wave, later the same day, added four questions: the student plan question (4 of 72 price a student or academic tier for one person), the scanned PDF question (0 of 72 state OCR of a scanned PDF; 18 accept a PDF), the music detectors question (5 cards state music; no record tests one) and the nonprofit and education question (0 state a discount; 12 sell an institutional programme), each answered from the fields read 8 to 10 November 2026.