How I write here
Some posts here carry a one-line note: "Drafted with AI; researched, edited, and fact-checked by me." This page is what stands behind that line. A label is a promise, and you should be able to see the promise.
The short version
I use AI to draft. I do the research, the editing, and the fact-checking, and I am the one accountable for every published word. If something here is wrong, that is on me — not the model.
Why I say it at all
"Made with AI" has turned into a warning sign, often fairly, because a lot of AI output is generic and unchecked. I would rather tell you how a post was made than let you assume. The useful question was never whether a tool was involved — it is what the person did with it. So here is what I do.
How a post is made
- AI drafts. I work with a model to get a first draft down and to pressure-test the structure and the explanations.
- I research and verify. What "research" means depends on the kind of post (below).
- I edit. I cut the generic AI texture and rewrite until it sounds like me and earns its length. Generic explanation is cheap now; the specific example, the correction, the failure mode is the part worth your time.
- I fact-check. Every checkable claim — numbers, quotes, attributions, citations, technical assertions — gets traced to a primary source. Quotes have to match the original, contested attributions are marked as contested, and every link has to actually say what I claim it says. A second pass — often a different model than the one that drafted it — tries to break each claim. Anything I can't stand behind gets cut or softened.
- I stand behind the result. On quality and on truth.
Two kinds of post
- Field notes. Most of what I write. These come out of building real software (mostly Narration Room, the Mac app I work on). The "research" here is the work itself — the bugs I hit, the trade-offs I made, the numbers from my own system. I check the draft against what I actually shipped, because the real risk in an AI-drafted field note isn't an invented web fact, it's the model confidently misdescribing my own work — and I'm the only one who can catch that. The codebase is closed, so I can't link you the commits; the specificity is the evidence.
- Essays. Less often, I write about a broader topic that leans on outside sources. There I verify every citation against its primary source, and I lean on what I've actually done rather than recycling a generic take.
What I won't claim
I don't say "verified" — that sounds like a third party signed off, and no one did. I don't hide behind "AI-assisted" when AI wrote the first draft — that undersells the tool. The label names the machine's job and my responsibility, and nothing more.
How this lines up with the standards
None of the formal rules bind a personal blog, but they all point the same way, and so does this page:
- The principle behind the EU AI Act's text-transparency rule is that AI-generated text is fine when a human has had editorial control and holds responsibility for it.
- Academic publishers (COPE, ICMJE, Nature) hold that an AI can't be an author and a human stays accountable.
- Google asks creators to tell readers how content was made — this page is that.
- Content Credentials (C2PA) now define a provenance format for text; when that's practical for a plain HTML blog I'll adopt it. For now, machine provenance for prose is still immature, so the contract is human: my name and my accountability.
Found an error? Tell me and I'll fix it.