AI Process Log: how-to
Every major deliverable ships a short AI process log. You're graded on the corrections you made, not on confessing you used AI.
AI is allowed and expected here (full AI policy on the home page). Major deliverables ship with a short AI process log, and I’m not grading the confession; I’m grading the verification: the specific, checkable corrections you made to what AI told you.
The four rules (every assignment)
- Disclose and cite which AI tools you used and what for. Disclosure never lowers your grade; hiding it is the violation.
- Keep a process log as you go: the key prompts, what the AI returned, and what you changed, rejected, or corrected.
- Verify every fact about a real organization against a primary source (its website, its IRS Form 990 via ProPublica, NCCS, or a person you spoke with). If you can’t verify it, you can’t submit it.
- You own every word. “The AI said so” is not a defense for an error, a fabricated source, or a made-up statistic.
What a weak vs. strong log looks like
❌ Weak (earns nothing): “I used ChatGPT to summarize the organization and it was all correct.”
✅ Strong: “I asked Claude to summarize Middle Way House’s finances. It claimed the org took in $4.2M in revenue in 2023 and finished the year in the black. I pulled their 2023 Form 990 from ProPublica (EIN 23-7300355): total revenue was $1,996,558 and total functional expenses were $2,531,078, so revenue was less than half what the AI said, and the year closed with expenses exceeding revenue by about $535K. I corrected both figures and cited the filing.”
A strong log names the organization, quotes the AI claim, and cites the primary source (the exact 990 line/page or URL) that confirms or refutes it.
A reusable template
For each major AI-assisted claim:
| AI claim (quoted) | What I checked (source + line/page) | Verdict & what I did |
|---|---|---|
| “…” | 990 Part VIII, line 1 / org website | False, corrected to … |
Plus a one-paragraph disclosure: which tools, for what (brainstorming / outlining / drafting / editing).
What you’re allowed to paste into a chatbot
Everything published on this website: these pages, the week pages, the assignment prompts, and the rubrics. O’Neill requires every syllabus to say that putting instructor-produced course materials into AI needs the instructor’s permission, so here is that permission, in writing. You need it to do the work this course asks of you, and I’d rather grant it once than have you email me every time.
It doesn’t stretch to another student’s work, or to anything I hand out that isn’t published here. If you’re unsure, ask me before you paste. That conversation has never once gone badly for someone who asked first.
Plausible is not the same as true
A chatbot will tell you, fluently and confidently, that an org runs a program it doesn’t, or that its revenue is a number it isn’t. That’s a hallucination: normal behavior for a tool that predicts plausible text. Catching it lying about a real organization is a stated goal of this course. Verify before you cite.