Educator's Guide

How to AI-proof your essay assignments

Thanks to generative AI, submitted essays are weak evidence of learning. AI detection is insufficiently reliable to be useful. The solution is an assessment method that evaluates the entire process of essay writing, including in particular reading and researching.

The challenge from AI

A student can now generate a plausible essay in seconds. AI detectors cannot sufficiently reliably distinguish generated passages from human writing for practical enforcement. This has forced most instructors to abandon at-home writing assignments, even though such assignments have long been central to training in the arts and humanities.

Thankfully, there is now a better response.

The solution: make the process part of the assignment

In grade school, you had to show your work when solving a math word problem. There was not always just one acceptable method, but you had to show how you arrived at the final number.

Essays can work the same way: we can treat the whole process as part of the deliverable, including a record of reading , note-taking , drafting, revising, and editing . AI-generated (or otherwise plagiarized) essays don't come with any such record. So, if an essay's entire history is part of the deliverable, there is no way to cheat (at least not without incurring a grade penalty).

Not only does this way of structuring assignments make it hard to cheat with impunity, but it allows instructors to provide feedback on a student's entire process and reward effort that would otherwise be invisible. It is better pedagogy.

But how?

There are many tools for tracking a student's editing work on a document (most analyzing edits in Google Docs), but how can the reading and researching be assessed?

This is where MATCHA comes in. When students use MATCHA 's Complete Work History feature, all their reading in its built-in browser is recorded and analyzed. This provides an extensive record of their research and reading. MATCHA 's built-in editor also tracks and summarizes editing. As a result, MATCHA submissions come with a complete, certified record of a student's process.

Instructors can replay the work when a close look is useful, but they do not need to replay every submission in detail: concise analyses summarize reading, writing, editing, pasted text, and other key features of the process. The result is a practical way to sidestep the AI problem while improving learning at classroom scale.

Can't all of this be faked by AI?

MATCHA prevents automation tools from inserting text, controlling the mouse, or otherwise simulating human engagement. Unlike most edit trackers, which are web-based, MATCHA is an installed desktop application written partly in low-level, OS-specific code. Its access to the operating system allows it to defeat all automation tools we have tested. It does this without requiring intrusive permissions (it requires no special privileges) or modifying the operating system. Students appreciate its non-intrusive, friendly approach to integrity.