Alternatives to AI detectors for teachers

The alternative is to record the process

An AI detector reads a finished paper and returns a number: the probability, by its model, that the text was machine-generated. A process-based tool ignores the finished text and records how it was produced: what was typed, what was pasted, when, in what order, with what pauses and rewrites. The first approach produces a score nobody can cross-examine. The second produces a record a person can watch and a student can explain.

That is the whole idea. The rest of this page covers why teachers are leaving detectors, which tools record process, how they differ, and what a record can and cannot tell you.

Why teachers are looking for an alternative

The false-positive research is the reason. Four findings come up in nearly every faculty discussion.

• Liang, Yuksekgonul, Mao, Wu and Zou at Stanford, writing in Patterns in 2023, found that GPT detectors flagged 61.3 percent of TOEFL essays by non-native English writers as AI-generated.
• Common Sense Media reported in 2024 that 20 percent of Black teens said they had been falsely accused of using AI, against 7 percent of white teens.
• The University of Waterloo turned off Turnitin’s AI indicator in September 2025 after an internal test flagged human-written text as 100 percent AI.
• Vanderbilt turned the indicator off in August 2023 with a public statement, and Northwestern, Yale and Johns Hopkins followed across 2023 and 2024. Public trackers now count more than 60 institutions.

Source: Liang et al., Patterns, 2023; Common Sense Media, 2024; University of Waterloo announcement, September 2025; Vanderbilt Brightspace statement, August 2023; GradPilot tracker; all checked 2026-09-08. The false-positive rates page collects the numbers, and the list of institutions is kept current.

Turnitin, for its part, states a document-level false positive rate under 1 percent for documents with more than 20 percent AI writing, and publishes a page on understanding false positives. Both things can be true: a low rate across millions of submissions, and a real number of students wrongly flagged. The flags are also rising. Turnitin’s February 2026 data put 15 percent of submissions at more than 80 percent AI for October 2025 to February 2026, against 3 percent at launch in 2023. Whatever the true rate of AI use, the number of contested cases grows with it, and a contested case needs evidence that is not a score.

Source: Turnitin blog and press materials, checked 2026-09-08.

The courts have started to say the same. In Newby v. Adelphi University (2026 NY Slip Op 26021), a Turnitin flag of 100 percent was contradicted by Grammarly and ZeroGPT, and the court found the finding “without valid basis” and ordered the record expunged. The lawsuits list tracks the others.

The tools that record process

Eight tools do some version of this. They are described here from the vendors’ own materials as checked on September 8, 2026, with what each records and what it costs. None of them is a detector, though two are sold by companies that also sell one.

Source: processfeedback.org; draftback.com; revisionhistory.com; briskteaching.com; turnitin.com; grammarly.com/authorship; cursivetechnology.com; plagiarismtoday.com (November 6, 2025); phrasly.ai/human-auto-typer; all checked 2026-09-08. Cursive was acquired by Blackboard (Anthology), announced September 2, 2026.

How to read the differences

Three lines divide the table. The first is when the record is made. Draftback, Revision History and Brisk read a log Google kept, after the fact. That log records document states rather than keystrokes, the student can turn it off, and auto-typer tools such as Phrasly Human Auto Typer, Undetectable.ai, TypeX and naturaltypist.com can fabricate it by replaying pasted text as keystrokes. Clarity, Authorship, Cursive, Process Feedback and EssayGuard record while the student writes.

The second is who holds the record. Authorship’s report belongs to the student. Clarity’s belongs to the institution’s Turnitin account. EssayGuard’s belongs to the instructor who created the assignment, under a retention window they chose. Which of those is right depends on whether you want a student’s defense, a campus system, or an instructor’s record.

The third is the purchase path. Clarity requires an institutional Turnitin contract with no individual purchase option. Cursive’s paid tiers are quote-based and per-student. The Docs replay tools, Process Feedback, Authorship and EssayGuard can all be started by one teacher.

What a record shows that a score cannot

A keystroke-level record shows whether a paragraph was typed or arrived in one paste, and keeps the paste marked through later edits. It shows the pause before a hard sentence, the thesis attempted three times, the paragraph moved from the end to the opening, and the order in which the argument formed. That is what a person writing looks like. A detector score describes the finished text and says nothing about any of it.

What a record cannot show

A record shows how text entered the editor. It does not show whether the ideas were the student’s own, whether a draft was dictated from notes written elsewhere, or what happened in another window. No process tool should hand you a verdict, and you should distrust one that does. The record is evidence for a conversation between an instructor and a student, and the finding belongs to the person, not the software.

Where EssayGuard sits

EssayGuard is one entry in the category above, built for the instructor’s side of the desk. The instructor shares one link; students write in a browser editor with nothing to install; every keystroke, revision, pause and paste is recorded with a server-side timestamp, and periodic snapshots are checked against the event stream so that a fabricated session reads as unrecorded. The instructor gets a replay and a printable evidence report. There is no stylistic analysis, no camera, no tab monitoring, and no verdict. The rest of this semester is free for every course you teach with unlimited students; after that the Educator plan is $99 per year. Details at /how-it-works and /pricing.

What to do this term

1. Put the process requirement in the syllabus before the first paper. The syllabus language page has wording.
2. Pick the tool that matches where your students write and who needs to hold the record.
3. Tell students the record exists and what it captures. A record only defends a student who knows about it.
4. When a paper raises a question, open the record and have the conversation. Do not open a detector.