Why detection failed
AI detectors work by measuring how statistically predictable a text is. Machine-generated prose tends to be smooth and probable; the theory is that human prose is burstier and stranger. The problem is that plenty of humans write predictable prose, especially students writing in a second language, neurodivergent students who learned writing as explicit pattern, and any student taught to write plainly. And plenty of AI text, lightly paraphrased, stops looking probable. The signal was always weak, and both sides of it erode as models improve.
The institutions noticed. MIT Sloan's guidance is blunt: AI detection software "is far from foolproof—in fact, it has high error rates and can lead instructors to falsely accuse students of misconduct." They note that OpenAI shut down its own AI-text classifier because of poor accuracy. WSU went further and terminated its Turnitin AI-detection contract after 33% of its review-board AI cases ended in "not responsible" findings. Even Turnitin's own documentation warns that its score should not be the sole basis for action against a student.
Process verification, the alternative
Detection asks: does this finished text look machine-made? That question gets harder every year. Process verification asks a different question: how did this text come to exist? That question gets easier, because the answer can simply be recorded.
A process record captures the writing as it happens. Keystroke timing, pauses, revisions, pastes, focus changes. Human composition has an unmistakable shape: uneven bursts, mid-sentence corrections, long thinking pauses before hard transitions. Text that arrives by paste, dictation-spoofing, or an autotyper extension looks categorically different. No model has to guess, because nothing is being inferred from the finished prose. The evidence is the writing event itself.
Detection makes the student prove a negative after the accusation. Process verification gives every student affirmative proof before any question is asked. The honest student is protected by default, not put on trial by default.
What good process evidence looks like
Not all process records are equal. Google Docs version history is the weakest form: snapshots, batched edits, time collapsed. It can suggest a document grew over time, but a determined cheat can grow a document too. The Palo Alto family learned the hard way that 1,162 pages of revision history could still be dismissed.
Strong process evidence has three properties. It is continuous (keystroke-level, not snapshots), so the rhythm of real composition is visible. It is complete (pastes, focus loss, and input anomalies are recorded, not just text changes), so the gaps tell a story too. And it is tamper-evident, so the record itself can be trusted when it matters. Here is what that looks like in practice.
The four options today
Four tools currently offer some form of writing-process visibility. They are genuinely different products, and the right one depends on what you need:
Brisk Inspect Writing. A Chrome extension that replays Google Docs revision history as a video. Free for basic replay, $99.99/yr for Pro. Easiest entry point if your class already lives in Docs. Inherits Docs' limits: snapshot-grade history, and nothing stops a paste-then-edit workflow from looking like writing.
Grammarly Authorship. Free. Tracks text provenance (typed, AI-generated, pasted) and produces a report the student can share. The catch is in that sentence: the student must install it, activate it, and choose to share. It is a tool for students to defend themselves, not a tool teachers can rely on for a whole class.
Turnitin Clarity. Turnitin's pivot in the same direction: a controlled writing environment that tracks the composition process and logs AI interactions. Named a TIME Best Invention of 2025. Sold to institutions; if your school already has a Turnitin contract and a purchase process, this is the incumbent path. Note that it still ships alongside the AI-detection scoring this page is about leaving behind.
Manupropria. A dedicated writing environment built for exactly one job: producing auditable proof of authorship. Keystroke-level capture, paste policy you control, tamper detection, and a replay you can read at a glance. No AI percentage anywhere in the product, by design. $4/month for an individual teacher, no purchase order, free tier to start.
Pick a path
If your students write in Google Docs and you want a free first step, start with Brisk's replay and know its limits. If your school is institutionally committed to Turnitin, evaluate Clarity. If you want authorship evidence that exists for every student on every assignment, with no install on the student's side and no detector anywhere in the loop, that is the product we built.