Moving Beyond Probability: A Forensic Approach to Proving Human Authorship

In the current landscape of generative AI, the conversation around "authorship" is becoming increasingly fraught with errors. We’ve all seen it: a student or a freelancer submits original work, only for it to be flagged by a probabilistic AI detector.
The problem is that most detectors (like GPTZero or Turnitin) rely on linguistic pattern matching—calculating perplexity and burstiness. This is essentially an educated guess based on word frequency. It’s a method that is prone to false positives, especially for technical writers or non-native speakers who naturally use structured, "predictable" language.
I decided to take a different approach. Instead of analyzing what was written, I focused on how it was created.
The Forensic Pivot: Proof of Human
Proof of Human is a utility designed to substantiate human effort through forensic metadata analysis rather than linguistic probability. It operates as a Google Docs Add-on to provide an objective "digital heartbeat" of the creative process.
The Technical Breakdown
Rather than looking at the final strings of text, the tool audits the internal document metadata and revision history to identify three key forensic markers:
• Typing Cadence Analysis: Humans have a distinct mechanical rhythm. We pause to think, backspace to refactor sentences, and have variable delay between keystrokes. AI-generated text insertion is typically instantaneous or robotic.
• Revision History Audit: An organic document has a specific growth curve. It starts with an outline, followed by iterative drafting and refinement. By mapping these timestamped events, we can distinguish between a natural evolution of ideas and a "bulk injection" of content.
• Manual vs. Paste Ratio: The tool calculates the volume of manually typed characters against the volume of pasted text. While pasting is normal for research, a document where 95% of the content appears in a single timestamp is a clear indicator of automated generation.
Authorship as Insurance
The goal here isn't to "catch" people using AI, but to provide Authorship Insurance for those who don't.
When a freelancer or student finishes a draft, they can run a deep audit to generate a Humanity Score and a verifiable Humanity ID. This ID can be shared with clients or editors, who can then use a public lookup tool to verify the forensic logs. It settles the argument with data instead of opinions.
Implementation
Building this within the Google Workspace ecosystem allowed for the most granular access to revision logs. The tool parses the revisions resource to reconstruct the document's history and provide a transparent report on the labor spent "in the chair."
I’m curious to hear from the community: As generative tools become the norm, do you think we’re moving toward a world where forensic process-verification is a requirement for high-stakes writing?
I've made the tool available at kyrosig.com for anyone who needs to prove their process.