AI disclosure
I study how people disclose their use of AI. It would be a little strange to build this site with one and tell you only that I did.
“AI-assisted” is the disclosure nearly every policy I know would accept, and it is close to useless. It satisfies the letter of the rule while saying nothing about what I reasoned through and what I handed off, which is exactly the gap my research is about. So here’s the longer version.
I rebuilt this site with Claude (in Claude Code) and Codex over a few sessions in August 2026, and reworked it with Claude Code in October 2026. Below is who did what, part by part.
Who did what
Code and layout
- AI
- Wrote nearly all the code: the layout, the news box, the floating headings and story toggles, the figure lightbox, the photo easter egg behind my headshot, and the typography. It also checked its work in a browser at phone and desktop widths, including colour contrast.
- Me
- Picked the sites I wanted this to feel like, the wide two-column shape, and news running under my bio instead of in a side rail. I chose every cut, and decided when something was finished.
Research section
- AI
- Read my CV and papers and proposed ways to group them, including the three moments (asking AI for help, reviewing what it gives back, owning the result), which it borrowed from the wording of my own module proposal. It also checked which other papers on my CV fit that framing.
- Me
- Turned down grouping by field, because computing education next to research tools made my work look like two unrelated things. I chose the three moments, which papers stay and in what order, and dropped the ones that didn't fit.
Writing
- AI
- Tightened my rough notes into the bio, the project stories, and this page.
- Me
- Wrote those notes. Every claim about the research is mine, and I sent drafts back whenever they stopped sounding like me.
Look and figures
- AI
- Reworded the speech bubbles in the CHI sketch to match the paper, and built the hand-drawn headings from an open-source doodle set.
- Me
- Rejected styling that looked like a template, said what each figure had to get across, and asked for simpler versions when they got busy. I kept the real workshop photo for the FAccT paper.
What this cost
This statement took longer to write than several of the features it describes, and that’s kind of the point. Honest attribution means reconstructing decisions you made quickly and half-consciously, and working out which ones were really yours. Ticking an “AI-assisted” box takes no effort. This took real effort, and that effort falls on whoever is being the most honest.
Whether that burden is fair (between students and instructors, or between the people who write disclosure policies and the people who have to follow them) is what my collaborators and I are working on, most recently in a CRAFT session at FAccT 2026.
If you’re drafting a policy, or you’re a student trying to work out what’s fair to write down, I’d ask less about whether you used AI and more about what you decided, and how you’d know if you were wrong.
Last updated October 2026. If something here is unclear or you think it leaves something out, please tell me. That kind of critique helps the research.