Poynter: The Financial Times case opens a debate about AI in editing, not just writing

The Financial Times added an editorial disclosure note to a column by Harvard economist Ricardo Hausmann in August 2026, stating that AI had been used to condense a longer draft of the piece before he submitted it. The column, which criticized Trump tariff policy, had already been published without the disclosure when readers raised questions about the writing quality — noting em dashes, unusually short sentences, and phrasing that suggested AI processing.

The FT’s editorial code prohibits AI in the writing process. What made this case different — and why Poynter commented on it — is that the AI was not used to write the column. It was used to shorten it. Hausmann’s research, argument, and original draft were his own. An AI tool was used at the editing stage to trim the piece to a word limit before submission.

Poynter frames this as a genuinely novel ethical question. Most editorial AI policies were written to address the question of generation: did an AI produce this text? The Hausmann case asks a different question: if a human wrote the text and an AI edited it, where does that fall in existing policy language? The FT’s code says AI should not compromise the integrity of journalism, but “integrity” was not defined with AI condensing in mind.

The case does not resolve that question. Poynter’s commentary is diagnostic rather than prescriptive: it identifies a gap between the AI policy language most publications are currently using and the situations AI tools are already creating. As these tools become more capable at condensing, restructuring, and polishing text, the category of “AI in the writing process” will need more precise definition — distinguishing generation from structural editing from copyediting, and deciding which of those activities a publication’s editorial standards are intended to govern.

For editors and policy writers, the FT case is useful not because it provides an answer but because it is a worked example of the kind of incident that policy written a year ago was not designed to address.