Critique as a named skeptic

Why it works

Models are sycophantic by default - they'll happily polish a draft that has a fundamental problem. This prompt flips that: it tells the model to assume the work is flawed and find what's broken. The 'pretend you're [skeptical role]' framing is doing the heavy lifting. Without a named skeptic, you get bland feedback ('consider adding more detail'). With one, you get the specific objections that role would actually raise. Use this on any draft you're about to ship: a doc, an email to a stakeholder, a PR description, a launch announcement. Run it twice with two different skeptical roles and you'll catch nearly every weakness.

Example output

Reading this as a skeptical CFO, here's what I'd push back on:

1. Slide 4 claims '40% efficiency gain' with no source or methodology. I'd assume it's made up until proven otherwise. Either cite the study or remove the number.
2. The roadmap on slide 7 has no costs attached. A roadmap without a budget is a wish-list.
3. The risks section lists three risks but no mitigations. That tells me you haven't actually thought about the risks - you've just noted them.
4. The 'next steps' slide has six bullets and no owner on any of them. By next week, none of these will have moved.

Common mistakes

Don't use a generic skeptic ('a skeptical reader'). The named role is what produces specific feedback. Common useful skeptics: 'skeptical CFO' for anything with numbers, 'skeptical engineer' for anything claiming feasibility, 'skeptical journalist' for anything claiming impact, 'tired senior PM who has seen this fail before' for anything ambitious. Also: do not argue with the critique in the chat - just take the notes and revise the draft yourself.