AI detectors compared on the same fields

A comparison page sets two AI detectors side by side on the same fields their cards carry, and runs only where both tools have a card in the detector register and at least one of the two has a record in the studies index. Ten comparisons are published as of 9 November 2026: Pangram against GPTZero, Pangram against Turnitin, ZeroGPT against Turnitin, the three-way GPTZero, Pangram and Originality.ai page, and six added in the fifth wave of the roadmap, Winston AI against Turnitin, GPTZero against Grammarly, the three-way ZeroGPT, Scribbr and QuillBot page, Pangram against Originality.ai, Winston AI against Originality.ai, and Scribbr against Turnitin. Each criterion on a comparison page opens with which tool the published field favours and why, each number names its source and date in the same sentence, and a 12-month cost table lists plan, monthly price, annual price, inclusions and the date checked for both tools. No page on this hub names a winner or passes a price verdict; the fields are printed and the reader decides.

The evidence section is where the pairs differ most. Where both tools appear in the same study, the comparison prints the two results from that study in one table with their bases. ZeroGPT and Turnitin share three records: Weber-Wulff 2023, where Turnitin classified 41 of 54 documents correctly and ZeroGPT 32 of 54, with 0 false accusations in 18 human documents for both (DOI 10.1007/s40979-023-00146-z); Walters 2023, where Turnitin scored 100% of 126 documents and ZeroGPT 87% (DOI 10.1515/opis-2022-0158); and Perkins 2024, where Turnitin left 88 of 104 AI samples undetected and ZeroGPT 77. Three shared records give that pair the fullest evidence table on the hub.

Pangram has no study record in the index, so the two Pangram pages carry the other tool's records and Pangram's vendor statements, attributed and dated: an accuracy of over 99% and a false positive rate of 1 in 10,000 on an aggregate of public datasets, read 8 November 2026 on the vendor's site. GPTZero brings seven records to its pair, from 9 false accusations in 18 human documents in Weber-Wulff 2023 to a 0.00% false positive rate at threshold 0.5 in RAID 2024, and Turnitin brings four. A vendor figure and a study figure never share a table on a comparison page, and the page says in plain words which tool has a record and which has a claim.

Two pairs added on 10 November 2026 in the ninth wave test the format where the evidence is thinnest. Scribbr vs QuillBot sets two brands of the Learneo group side by side with no record that ran both, so its evidence table carries each tool's own records and QuillBot's vendor-submitted RAID leaderboard entry in labelled rows; Reality Defender vs Sensity sets two deepfake vendors with no study record at all against each other on the six fields a deepfake buyer asks about first, and says once where each vendor publishes nothing.

Two more pairs added later on 10 November 2026 in the tenth wave take the two shapes the hub now holds. Pangram vs Copyleaks sets the two API-first text detectors that each carry a study-measured zero on human text against each other, with Van Vlasselaer 2026 as the one record that ran both (97.5% inclusive accuracy against 75% of the same 40 papers scored under 20%); QuillBot vs Turnitin sets a free consumer check against an institutional licence with no record that ran both, Turnitin's four measured zeros and QuillBot's vendor-submitted leaderboard entry in labelled rows, and a cost table that says Turnitin publishes no price.

Pairs are chosen from the register the editors keep of comparison queries readers ask, in order of demand, and a pair is skipped when another site already compares the two tools. Sixteen pairs cleared that check for the year ahead; the ten below are the first. Each page carries a register ID in the AIDI-C series, a visible Updated date that moves only when a figure changes, and a Sources list with every vendor page and DOI used. Prices on every comparison are rechecked on 8 December 2026 with the rest of the index.

The same fields appear in the same order on every pair: what each tool detects, languages, input limits, free tier, paid plans with the 12-month table, API and integrations, vendor claims, study records, retention, and the rankings each card appears in. The methodology page sets the source rules, the rankings hub holds the sorted fields the comparisons cite, and the questions hub answers the single-sentence questions behind many comparison searches.

Two pairs closed the roadmap on 10 November 2026: the Sapling vs GPTZero comparison, two text detectors with three shared records (Liang 2023, Walters 2023, Kar 2024), and the Pindrop vs Resemble Detect comparison, two voice vendors with no record between them, on the 14 fields and the voice fields.