Process Verification: The Detection Alternative That Actually Works

Why Process Beats Text Analysis Every Time

A detection tool analyses the finished product and asks whether it statistically resembles AI output. A process check analyses how the product was made and confirms whether the student did the work. One produces a probability estimate. The other produces observable evidence.

In any formal academic misconduct proceeding, observable evidence beats a probabilistic estimate. Turnitin's own documentation acknowledges this by stating the tool should not be used as the sole basis for punitive action. That disclaimer implicitly acknowledges that a detection score alone is not evidence. Process verification produces what the detection score cannot.

The legal exposure this avoids is documented in the pillar. Schools that take disciplinary action based solely on a detection score are exposed to procedural fairness challenges, and student defense firms are actively building caseloads around exactly those situations in 2026. A process portfolio removes that exposure by providing an evidentiary record that does not depend on a vendor's probabilistic classifier.

Google Docs Version History: What to Look For

File. Version History. See Version History. The document opens a timestamped log showing every edit from the first keystroke to the final save.

A genuine drafting process has a specific look. Text appears in sections across multiple sessions over several days. Paragraphs get restructured. Sentences get deleted and rewritten. A thesis that appears in session one gets refined by session three. The document shows a student thinking through a problem over time.

A pasted AI submission has a completely different look. A blank document. Then a single paste event containing a complete, fully formed essay. Then minor formatting changes. Then submission. The timestamp often shows the entire "drafting" process took under fifteen minutes.

That difference is unambiguous and requires no interpretation. It is timestamped, logged by Google, and cannot be retroactively altered by the student. It is the strongest single piece of process evidence available to a classroom teacher.

Make the version history link a required submission component. Same status as a bibliography. A submission without it is incomplete. The student pastes the link from File, Version History, Copy Link. It takes ten seconds. Requiring it signals that the process matters as much as the product.

The Micro-Viva

Two to three minutes. No advance notice required. Three questions, or fewer.

Explain one complex word you used in your submission. Summarise your argument in two sentences. Tell me where your primary source contradicts your thesis, or where you had to qualify a claim because the evidence was weaker than you expected.

A student who wrote the paper can answer all three without any preparation. They were in the document for five days. The vocabulary choices are theirs. The argument is something they worked through. The source is one they actually read.

A student who submitted AI output without engaging with the content typically cannot. They can often summarise vaguely. They struggle with the specific vocabulary question because the word was chosen by a language model, not by them. They cannot identify the source contradiction because they did not read the source.

The micro-viva is also virtually impossible to fake in real time with an LLM. Answering a verbal question from a standing teacher in a classroom requires immediate spoken knowledge of specific content. It cannot be delegated to a phone under the desk in the thirty seconds available.

Run micro-vivas selectively. Any submission that triggers a detection flag, teacher intuition, or portfolio inconsistency gets one. That targeting keeps the workload manageable without losing the coverage.

The Process Portfolio

Require the following alongside the final submission for major writing assignments.

The original brainstorm or notes. This can be handwritten and photographed, a bullet-point list in the document, or a mind map. It shows what the student knew and was thinking before they started writing.

A draft with visible revision marks. Not a clean second draft. A working draft with deletions, additions, and structural changes that show a student negotiating between what they wanted to say and what the evidence supports.

A source annotation: one sentence per source explaining how that specific source was used and what it contributed to the argument. A student who Googled citations after writing the paper cannot produce specific annotations. A student who read the sources as part of the research process can.

The version history link. As described above.

Students who complete this process cannot plausibly claim they did not write the work. Students who bypassed the process have nothing to submit for three of the four components. That gap is immediately visible before any detection tool is consulted.

How to Make This Fit a Class of 30

The portfolio does not require the teacher to grade every component on every submission. It requires students to produce them. That distinction matters.

Spot-check 25-30% of portfolios fully on each major assignment. Any submission that triggers a detection flag, strikes the teacher as inconsistent with the student's established writing level, or arrives without a complete portfolio gets a full review. That targeting concentrates teacher time on the submissions that actually warrant it.

Building this into a classroom AI policy makes the portfolio requirement structural rather than ad hoc. When it appears on the course outline on day one, it is not a surprise when a submission is returned for an incomplete portfolio. The expectation was set before the first assignment opened.

Teachers who have adopted process verification report a consistent outcome: the number of submissions that warrant close review drops after the first assignment. Students who might have been tempted to submit AI output without engagement make a different decision when they know the version history link is required. The portfolio requirement changes behavior before any detection tool is consulted.

What This Looks Like in Practice

Two concrete scenarios.

Scenario A: a submission gets flagged by a detector. The teacher requests the version history link and schedules a two-minute verbal check. The student produces a clear drafting history spanning six sessions over four days and explains their thesis without hesitation. The case is closed. The detection score is irrelevant. The process evidence is definitive.

Scenario B: a submission arrives without a version history link. The teacher marks it incomplete and requests the link before grading proceeds. The student produces a history showing the complete essay was pasted in a single event the night before the deadline. The teacher schedules a micro-viva. The student cannot explain two of the three vocabulary choices they used. The process violation is documented with timestamps and a record of the verbal check. That documentation holds up in any formal process in a way that a detection percentage never does.

In both scenarios the detection tool is optional. The process evidence is what matters.

Is My Assignment AI-Resistant?

Answer five questions about an existing assignment to find out whether it requires process verification to be defensible, or whether the design already produces enough evidence on its own.

FAQ