Three Perspectives
AI defaults to the most statistically probable answer β which means it tends toward the mainstream view, the safe take, the one that offends nobody. This recipe forces it to argue multiple positions genuinely, not just acknowledge that βsome people think X while others think Y.β
The Recipe
I want to explore [topic / idea / decision].
Give me three distinct perspectives in transcript format β have them actually
debate, not just take turns stating positions:
1. The Skeptic β cynical, looks for flaws, asks "what could go wrong?"
and "who benefits from this framing?"
2. The Optimist β genuinely excited, identifies real upside and potential,
argues for why this matters
3. The Pragmatist β data-driven, asks "what does the evidence actually say?"
and "what would this look like in practice?"
Each voice should challenge the others, not just present their view in a vacuum.
End with a 2-sentence synthesis: what the debate reveals, not which side won.
Variations
For a decision youβre weighing:
I'm trying to decide whether to [decision].
Run this through three perspectives:
1. Someone who would strongly advise against it and why
2. Someone who would enthusiastically support it and why
3. A neutral advisor focused only on what information is missing before
a good decision can be made
After all three: what's the one question I most need to answer before deciding?
For understanding a public debate:
Explain the debate around [topic] through three genuine perspectives:
1. The strongest version of the [position A] argument β steelmanned,
not a strawman
2. The strongest version of the [position B] argument β same standard
3. A third position that both sides often ignore or dismiss
I want to understand why smart, informed people genuinely disagree here,
not just the surface-level talking points.
For evaluating your own work:
Review [this plan / idea / piece of writing] from three angles:
1. A harsh critic looking for the weakest points
2. A supportive collaborator building on the strongest elements
3. A neutral editor focused purely on clarity and structure
Have them respond to each other's points, not just give independent assessments.
Why this works
When you ask for a single answer, the model satisfices β it finds the most defensible response. When you ask for three genuinely different positions that debate each other, it has to model multiple coherent worldviews simultaneously. You get richer analysis, surfaced assumptions, and cleaner distinctions between the parts of an argument that are solid vs. contested.
π Leftover Remixes
πΆοΈ Spicy: βNow: which of those three perspectives is closest to your actual assessment, and what would it take to change that view?β
π§ Mild: βSummarize the debate in 3 bullet points β one thing each perspective got right that the others underweighted.β
π° Budget: βWhatβs the strongest objection to [position] β the one that even supporters should take seriously? Give me that argument only, made as well as possible.β
Prompt concept via Elton Jones / Tomβs Guide. Adapted and expanded for The Prompt Kitchen.