Critique loop output — three rounds on the same paragraph

The critique loop is one of the most consistently useful prompts in the stack. The logic is simple: tell the AI what you wrote, then ask it to find the weakest part and explain why. Then fix it. Repeat.

Round 0 — original paragraph:

“AI tools are increasingly being used by consulting professionals to improve the efficiency of their workflows, enabling them to process more information in less time and deliver better outputs to clients.”

This is the kind of paragraph that gets written on a deadline. It says nothing. Every word is defensible; the paragraph contains no claim.

Round 1 critique:

“The weakest element is ‘increasingly being used’ — this is a hedge with no anchor. It could describe ten users or ten million. The phrase ‘better outputs’ is also unsupported and vague. A reader can’t disagree with this paragraph, which means they can’t learn from it.”

Round 1 revision:

“Senior consultants who’ve tested AI tools for more than six months report one consistent finding: the tools are fastest at breadth, slowest at depth. They’ll map a literature review in forty minutes. They won’t tell you which finding matters.”

Round 2 critique:

“‘Senior consultants who’ve tested’ is now doing the credibility work but it’s anonymous. The paragraph is better, but the hedging has moved rather than been resolved. Also: ‘slowest at depth’ is asserted but not shown.”

Round 2 revision (the published version):

“The machine is fast at ‘here are ten things.’ You are paid for ‘here is the one that matters.’”

What the loop can’t fix: The critique loop identifies weakness in the argument structure. It can’t tell you whether the underlying claim is true. Round 0 was weak because it said nothing. Rounds 1 and 2 were stronger because they made a specific, testable claim. Whether that claim is accurate is a human judgment.