I use AI tools in the course of this work. I'd rather tell you exactly how than leave you to wonder, so here's the whole picture.
The work is mine
I use AI as a tool: for research, for finding and comparing sources, for sorting and tagging, for analysis, for editing, for testing a formulation against three others, for drafting. What comes out the other side is still my responsibility. I own the judgment, the argument, the sourcing standard, and the final language of anything published under my name.
That is the line. It is not a line between editing and drafting, which is a distinction that sounds principled and means very little in practice. It is the line between assistance and accountability. A tool can help me work. It cannot be the author, and it does not carry the consequences of being wrong.
Evidence is checked against the source
A claim that rests on evidence gets checked against the thing it rests on. When a model tells me what a bill says, what a report found, or what a number is, that is a lead, not a citation. I go to the statute, the study, the filing. Where a source matters, the source controls, and what I publish reflects the source rather than a summary of it.
Model output is not self-validating. Nothing here goes live that I have not read against what it claims to be reporting.
What I don't put into these tools
I point these tools at editorial material: public sources, my own drafts, my own notes. Sensitive, confidential, proprietary, and personally identifying information gets handled deliberately, not pasted into an external system because it happened to be open in another window. I do not run subscriber or visitor information through them. How I handle your personal data is covered in the Privacy Policy.
When I get it wrong
If something I publish is wrong, it is my error, whatever role a tool played in producing it, and the correction is mine to make. "The model said so" is not an explanation, and I will not offer it as one.