As generative AI becomes increasingly integrated into college writing, the critical question is no longer whether students are utilizing it, but rather how they choose to use it, where its influence appears, and what implications this has for their learning experience.
A new tool, DraftMarks, aims to illuminate this often-invisible process by tracking changes in a document over time and highlighting the points at which AI contributes to the writing.
DraftMarks was created by researchers at Georgia Tech and Stanford University.
The introduction of this tool coincides with a time when students are increasingly integrating AI into their academic work. A 2025 AI in Education trend report revealed that 90 percent of college students use AI for their schoolwork, with nearly half relying on it during the drafting phase.
This trend has created challenges for many educators. Traditional methods of evaluating writing are becoming inadequate, and simply detecting AI use falls short of providing meaningful insights.
Moving Beyond Basic AI Detection
Unlike traditional tools that merely provide a percentage of AI-generated content or speculate on its contributions, DraftMarks centers on the writing process itself.
It reveals when a student revises using AI, when they accept AI-generated content, when they choose to discard it, and when they completely reject AI input.
This approach reflects the understanding that AI integration in writing involves a series of decisions. One student may use AI to develop an argument, outline a paragraph, enhance a sentence, or adjust their tone, while another might depend on it more heavily, letting it do much of the work. These are distinctly different approaches, and DraftMarks is designed to acknowledge these differences.
Instead of condensing all this complexity into a mere suspicion score, the tool endeavors to narrate how a piece of writing has evolved.
This shift is significant because AI has become so commonplace that merely asking if it was used offers little value. If most student writers are engaging with AI in some form, the pertinent question shifts to understanding the role it plays in their actual work.
Tracking Students’ Writing
DraftMarks functions as an enhanced reading layer applied to a document. As readers engage with the text, they are presented with visual indicators reflecting various levels of AI involvement.
Eraser crumbs highlight sections that underwent extensive revisions, while smudges indicate AI-generated alterations that modified an argument’s strength without completely changing its content.
Additionally, masking tape marks passages originally written by AI, and glue residue indicates areas where AI-generated text was later removed. Ghost text signals that a writer prompted AI but chose not to incorporate the results, while different fonts differentiate between human-written lines and AI-produced content. The overall effect resembles a dynamic record rather than a polished final draft.
Instead of presenting a neat and finalized document, the writing reveals traces of its development.
“By making the typically invisible parts of the process tangible, it compels writers to confront whether they are genuinely engaging with AI or merely accepting it without thought,” said lead author Momin Siddiqui, a master’s student at Georgia Tech. “Ultimately, it encourages writers to make more intentional decisions about their future collaboration with AI.”
Designed for Real Educational Settings
The researchers began by focusing on educators rather than building a detector and then seeking feedback from teachers. During initial studies involving 21 participants, the team observed how instructors evaluated student writing and the types of cues they naturally sought when assessing revision, originality, and evidence of learning.
As a result, DraftMarks incorporates physical cues commonly associated with the writing process: eraser residue, tape, smudges, and removal traces.
“These indicators are designed to mirror aspects of the writing process we are already familiar with,” explained Adam Coscia, a Ph.D. student at Georgia Tech. “They help both students and teachers appreciate the effort behind a piece of writing and whether students have achieved the intended learning objectives.”
In the background, DraftMarks tracks the history of drafts and categorizes edits and AI interactions in real time, allowing those visual cues to emerge nearly instantaneously.
Testing in Real-World Contexts
To evaluate the tool’s effectiveness outside controlled environments, the researchers tested it with a broader group of 70 participants, including students, teachers, journalists, and general readers.
Educators expressed a keen interest in understanding the progression of the writing, seeking insights into how ideas evolved, the extent of AI’s influence, and where students were making independent decisions.
Conversely, other readers frequently concentrated on a more abstract quality: trust.
For them, DraftMarks provided insights into authorial intent, helping them assess whether a writer utilized AI thoughtfully or haphazardly.
As AI becomes a fundamental aspect of journalism, public discourse, and daily communication, people may desire more nuanced answers regarding its role in writing. They may look for clarity on how AI has shaped the voice and decisions underlying the final output.
Transitioning from Detection to Judgment
What distinguishes DraftMarks is its emphasis on shifting the focus from detection to judgment.
Traditional AI detection tools often imply that the primary goal is identifying misuse. In contrast, DraftMarks posits that the more enriching task may involve helping both writers and readers reflect more clearly on collaboration.
“DraftMarks entirely transformed my perspective on my own writing,” Coscia remarked. “I was astonished by how much I cared about authorial intent once I could visibly see how AI influenced my tone. It made me recognize how minor AI choices can subtly alter my intended message.”
The researchers aspire for tools like DraftMarks to contribute to more meaningful discussions about AI’s role in education. Rather than solely asking if students utilized AI, such tools may encourage both teachers and students to reflect more honestly on the nature of learning when humans and AI collaborate in writing.
This project was presented at the Association for Computing Machinery Conference on Human Factors in Computing Systems in Barcelona.
—–
If you found this article insightful, consider subscribing to our newsletter for more engaging articles, exclusive content, and the latest updates.
Explore more on EarthSnap, a complimentary app developed by Eric Ralls and Earth.com.
—–
