Ep 39 "Trendslop": Why AI Keeps Giving Strategists the Same Answer
A chief strategy officer asks ChatGPT, Claude, and Gemini the same strategic question before his board retreat. Different phrasing, identical conclusions — and no way to tell if any of it is actually good advice. Harvard researchers Romasanta, Thomas and Levina tested seven LLMs across thousands of trials on seven strategic tensions — exploit versus explore, differentiate versus commoditize — and got the same answer every time, regardless of company or industry. Reordering options shifted responses by 22%; adding rich context moved the needle only 11%, because models default to whatever appears most often as "correct" online. Given the option to hedge, LLMs picked both sides 25% of the time — Porter's stuck-in-the-middle, dressed up as sophistication. The researchers' fix — expand options, don't converge — assumes a clearly articulated strategy process to compare answers against. Most companies have a five-box diagram instead. Process clarity, not better prompting, is the real prerequisite. Timestamps: 00:01:52 Episode built around Harvard's April 2026 trendslop research documenting how AI converges on identical strategic advice 00:19:23 Reordering the same two options in a prompt shifted AI recommendations by 22 percent, more than added context did 00:30:10 At AT&T, legacy systems no one fully understood made management afraid to modify them, a metaphor for opaque strategy processes 00:45:11 AI's real strategic value confines to one stage: building a shared factual snapshot, where speed and coverage genuinely beat teams alone 01:04:31 The real prerequisite isn't better prompting, it's process clarity: an articulated strategy process to compare AI output against
- English (US)
Not Rated Yet