Skip to content
Detection Guide

Proving Your Writing Is Human

Step-by-step strategies for building a writing provenance trail

Proving Your Writing Is Human

In 2024, the burden of proof shifted. For most of the history of writing, your words were assumed to be yours unless someone proved otherwise. Today, in universities, newsrooms, and workplaces around the world, the default assumption has inverted. If a detection tool flags your work, you are guilty until you prove yourself innocent. This guide is about building the evidence you'll need.

How AI Detection Actually Works

Before you can defend against detection tools, you need to understand what they're actually measuring. AI detectors don't "read" your writing the way a human does. They don't evaluate ideas, creativity, or voice. Instead, they perform statistical analysis on your text, looking for patterns that correlate with machine-generated output.

The two most common metrics are perplexity and burstiness. Perplexity measures how predictable your word choices are - if each word in your text follows naturally from the last, perplexity is low. Burstiness measures variation in sentence complexity - humans tend to mix short and long sentences unpredictably, while AI output tends toward uniformity.

The problem with both metrics is that they describe tendencies, not rules. Some human writers naturally produce low-perplexity text - their prose is clear, precise, and predictable. Some writers maintain consistent sentence lengths. These writing styles are perfectly valid, but they happen to look statistically similar to AI output.

Why False Positives Happen

False positives aren't random. They cluster around specific writing characteristics and author demographics. You're at higher risk if you write in formal, academic English; if English is not your first language; if you have a highly structured thinking and writing style; if you write on topics heavily represented in AI training data; or if you edit extensively, producing highly polished final text.

Understanding your risk profile doesn't mean changing how you write. It means being prepared.

Building a Writing Provenance Trail

The single most effective defense against a false accusation is a documented writing process. Here is how to build one:

Keep your drafts. Write in a tool that saves version history - Google Docs, Microsoft Word with AutoSave, Notion, or any platform that timestamps your changes. If you write longhand, photograph your notebooks with a timestamp. The goal is to show the evolution of your work from rough notes to finished text.

Save your research. Bookmark the sources you consulted. Keep screenshots of articles you read. Save your database search queries. If your writing references specific sources, be able to show how you found them.

Document your process. This doesn't mean narrating every sentence. It means saving the artifacts of real work: the outline you made, the notes you took during a phone interview, the three opening paragraphs you wrote and discarded, the email to a colleague asking for feedback on a draft.

Use metadata. File creation dates, word processing metadata, and browser history all contribute to a provenance trail. Don't delete your browser history if you're researching for a piece that might be scrutinized.

The best defense is not a better argument. It is better evidence.

What to Do If You're Accused

Don't panic, and don't get defensive. An accusation based on AI detection is a statistical claim, not evidence of wrongdoing. You have the right to understand exactly what triggered the flag and to present evidence of your authorship.

Request the specific detection report. Ask which tool was used, what confidence score was generated, and what specific passages were flagged. You need this information to respond effectively.

Present your provenance trail. Show your drafts, your research notes, your version history. Walk the reviewer through your process. The goal is to demonstrate that a real, human process produced this work.

Request human review. Ask that your work be evaluated by a subject matter expert - someone who can assess the quality of your ideas, the specificity of your examples, and the characteristics of your voice - rather than relying solely on algorithmic analysis.

Know your rights. Many institutions now have formal appeal processes for AI detection disputes. Familiarize yourself with your institution's policy before you need it.

Tools That Can Help

Several tools and practices can strengthen your position. Version-controlled writing platforms create automatic provenance trails. Writing process journals (even simple dated notes) document your workflow. Some writers have begun using screen recording software during writing sessions, creating irrefutable evidence of human authorship.

The content provenance standard C2PA, developed by Adobe, Microsoft, and others, is beginning to be adopted for text as well as images. While still early, C2PA-compatible tools may eventually offer writers a way to cryptographically prove when and where their work was created.

The situation is imperfect, and the burden should not fall on writers to prove their humanity. But until institutions catch up, preparation is your best protection.


EV

Dr. Elena Vasquez

Dr. Elena Vasquez is an AI researcher specializing in natural language processing and detection systems. She consults with universities on fair use policies and has published extensively on the limitations of current detection tools.

The Sunday Letter

Every Sunday, one email. A featured essay, a case study update, a craft tip, and a writing prompt. No AI wrote this.