How Testing Our Own Writing Patterns Is Helping Us Preserve What’s Human
Earlier this month, Alex Adam, our Head of ML, asked the entire team to analyze their own native writing patterns on Google Docs. This was in collaboration with Edwin Thomas and our ML team, who is training a typing pattern algorithm for human patterns, and our Chrome Extension team, who are working on a live writing coach.
It was part of our team retreat in Nova Scotia, Canada, where we had a lot of fun brainstorming and workshopping the future of GPTZero in person, all centered around our mission of preserving what’s human.
The exercise was designed to capture what people write as well as how they think while writing. What we learned from it was so useful in our ongoing quest to preserve critical thinking in a world of AI, as it went from a technical exercise to a philosophical one.
Digital Fingerprints
We discovered that writing patterns are surprisingly distinctive. While everyone's final text might look polished, the journey to get there tells a completely different story. Some team members wrote straight through with minimal edits; others (like me) brain-dumped thoughts first, then rearranged them into order. One colleague's typing video showed clean, linear progress… while mine was a wild ride of deletions and restructuring.
It was interesting to see how we all wrote compared to each other and to notice things like the length of the paragraphs we chose. But overall, the exercise ended up being kind of revelatory. Seeing our writing samples side by side forced us to remember just how personal writing is in both style and process.
This also showed how, if writing is an externalization of thought, then typing patterns are a direct reflection of cognitive process. Each person’s rhythm of creation (how they pause, and the way they build and rebuild sentences) creates a signature as distinctive as a fingerprint. When AI can mimic our final output, these micro-patterns of creation could just be the key to preserving original authorship.
The Prompts
This got us thinking about how everyone’s writing process is almost like a form of intellectual handwriting. It might be invisible to readers, but it’s visible to the right tools.
Specifically, we did the following series of writing exercises:
- Natural writing: Personal reflections typed normally
- AI mimicry: Copying and pasting AI-generated text, then trying to make the typing patterns look human
- Hybrid editing: Manually blending human and AI content
- AI-assisted editing: Using Gemini to revise sentence by sentence.
Even when people deliberately tried to imitate human typing patterns while using AI, the underlying rhythms felt quite different. As it turns out, the natural flow of human hesitation and revision is nearly impossible to fake convincingly.
Beyond Detection
This exercise revealed something bigger than just AI detection. By capturing these patterns, we're building a data-backed map of how people actually think and write. This has implications for understanding how our millions of users, including students and writers, create their work, and how we can better support these processes.
After all, as AI becomes more and more sophisticated, the battleground for authenticity arguably shifts from analyzing final text to understanding the underlying process of creation itself.
Writing remains the first frontier of critical thinking and the most common medium where AI is used. The future we’re building toward might have started with detecting AI-generated content, but is moving towards embedding authorship analytics directly into writing tools, so users can benefit from real-time insights as they create.
The retreat reminded us that behind every piece of text is a human mind at work, which comes with its own unique patterns. Our job at GPTZero is to help protect those distinctions in a world where originality is increasingly under threat.