Writing with AI, honestly

It would not be hard to build a tool that scrubs the traces of AI out of a draft. I decided against it, for a practical reason as much as a principled one. A small but growing number of funders now ask directly. An application that quietly answers “no AI” when AI was used is a misrepresentation — and it lands on the nonprofit, not on the software that helped write it. A plain sentence describing how the work was made is both safer and easier to defend.

It rarely comes from people cutting corners. It comes from fundraisers writing in their second or third language, and from people who think far better out loud than on the page, who use AI to be understood. Their ideas are their own. They should not be judged on how those ideas reached the page — and saying so directly protects them better than staying quiet does.

Two honest limits. The marks that identify machine-written text are being introduced across the industry on different timelines, so nobody — including me — can promise you what is or isn't detectable in a given piece of writing. And the separate tools that claim to spot AI by how the prose sounds are unreliable: they are documented to flag people writing in a second language far more often than native speakers. Neither is a reason to conceal how you work. Both are reasons to describe it plainly.

• Read every draft before it goes out, and check the numbers, names and dates against your own records.
• Put back the specifics only you know.
• If you're asked, disclose it in a sentence.