Generative AI

Everything was in place – policies, training, guardrails. And still, AI errors slipped through. This isn’t a technology failure but a human one. Perhaps the answer is simple: someone must own verification.

I've been deep in ChatGPT since 2022. Now I'm running a deliberate, staged experiment with Claude Pro – and one feature in particular, Cowork, has genuinely surprised me. It has my full attention as I continue working it out.

A New York Times review, an AI tool, and a missed check. The result wasn’t just overlap with another review – it was a lapse in process. A simple reminder: if AI touched it, you must verify it.

The question “Was this written by AI?” misses the point. What matters is whether content is accurate, useful, ethical, and purposeful. This post explores why judgement, not detection, is the responsibility communicators can’t outsource.

When a university professor lost two years of academic work after a single change to ChatGPT's settings, the lesson was stark. If the work matters, backups are your responsibility. ChatGPT is not an archive – and convenience is no substitute for protection.

Ads in ChatGPT aren’t the real story. Trust is. As OpenAI begins testing advertising, the challenge isn’t revenue or relevance, but whether users continue to believe the system works for them, not just for advertisers.

AI can invent sources, quotes, and links with total confidence. If it touched your content, verification is no longer optional. Editors must now be fact-checkers. This is no longer best practice. It is the baseline.

ChatGPT now promises to stop using em dashes when instructed, ending a surprisingly noisy debate over a tiny punctuation mark. But the real story is how AI picked up the em dash from us in the first place – and what that says about voice, trust and the way we write today.