AI Usage Disclosure v1.0

This page describes how I have used AI tools in my writing in 2026. It is intended to help readers, editors, and publishers evaluate my work knowing how it was produced. It applies to my essays and articles on Ice Frontiers.

My poems are written without AI.

Guest contributors write their own pieces, and I have no control over how they use AI.

The models I currently use are ChatGPT, Claude, and Gemini.

My writing begins with an idea or a question I want to explore. I research the subject using traditional search engines, reading relevant material, and discussing ideas with AI. I then develop my argument through notes, rough paragraphs, and conversations. I may describe an experience, introduce an analogy, propose a conclusion, or raise an objection. The result doesn’t look like a traditional manuscript. It is closer to the collection of post-it notes on a whiteboard from which a manuscript emerges, an earlier stage of what is usually considered a first draft.

AI helps me organize the material, develop polished paragraphs, and assemble a preliminary version of a complete text. What follows is an iterative process of editing, sometimes changing a few words or sentences, sometimes rewriting or reorganizing entire sections. I may reject a suggestion, develop an alternative, or realize that the argument needs to go in a different direction.

The pace changes as the text matures. Early iterations are quick. As the text gets closer to publication, I switch to more capable models and higher effort settings in Claude, and I submit the manuscript to several models and other writing tools for longer reviews. This is much like circulating a scientific manuscript among colleagues or reviewers.

I often take the revised text back to a model that contributed to an earlier version and ask for feedback. These exchanges can be surprisingly entertaining. I have seen a model readily acknowledge that another tool’s revision was better than its own. At other times, models argue vigorously against each other’s suggestions.

These disagreements are particularly useful because they force me to examine the argument more carefully and decide which interpretation I actually support. Of course, disagreement between models is not proof that either interpretation is correct. I still need to evaluate the evidence and exercise my own judgment.

The process continues until the text reaches a relatively stable form, when the major objections have been addressed and further revisions no longer seem to improve it. That does not mean the result is perfect, any more than a scientific paper is perfect when all its coauthors agree to submit it. It means that I am satisfied with the argument and prepared to put my name on it.

I claim authorship of the texts I write using this process, and responsibility for them.