Expert panel prompt builder
Some questions have no single right answer — they have tradeoffs that look different depending on who you ask. The classic way to surface those tradeoffs is to put several experts in a room and let them argue. This builder produces a prompt that makes an LLM play multiple expert personas at once, each reasoning from their own discipline and incentives, so you see the full landscape of a question instead of one flattened answer.
The core insight is simple: a single LLM query collapses the full problem space into a single perspective. Even when you ask for “multiple viewpoints,” the model tends to smooth over real disagreement and produce a polite consensus. Forcing it to inhabit several named roles with explicitly different incentives breaks that tendency — the roles argue because the prompt tells them they must.
How it works
You define the question, the roster of experts (each with a title and specialty), the number of exchange turns, and the moderation style. The tool assembles a prompt that assigns the model each persona explicitly, tells it to keep their voices and priorities genuinely distinct, and instructs them to push back on each other where they truly disagree. With a synthesizing moderator, the prompt adds a final role that weighs the arguments and delivers a reasoned conclusion. The result reads like a transcript of a real panel, with the disagreements left intact rather than smoothed over.
The three moderation styles
- None — experts speak in turn without steering. Good when you want to see where the disagreement lands naturally.
- Light — a facilitator keeps the conversation on topic and may prompt quieter experts to respond. Useful for questions that would otherwise let one perspective dominate.
- Synthesizing — the moderator delivers a final summary that names the key tensions, weighs them, and suggests a direction. This is the most useful output for real decisions because it does the integration work for you.
Worked example
Question: “Should we expand our SaaS into the enterprise tier before we have product-market fit in the SMB segment?”
Experts:
- A VP of Sales who cares about near-term revenue and close rates
- A Head of Engineering who cares about scope, support burden, and roadmap debt
- A CFO who cares about burn rate, deal size, and customer concentration risk
Three turns, synthesizing moderation.
What you get back: the VP argues enterprise deals fund the runway you need to find PMF; the engineer warns that enterprise support contracts will eat the entire team; the CFO notes that a single large customer at pre-PMF creates dangerous concentration risk but also gives you faster feedback cycles. The moderator’s synthesis surfaces the real decision — it is not “enterprise or not” but “what contract protections and resource commitments can contain the support burden?” — which is the insight you actually need.
Tips for better results
- Conflict is the feature. Roles with naturally opposing incentives — a security architect and a growth marketer, a lawyer and an engineer, a founder and a CFO — produce the most useful tension. Three agreeable product managers produce repetition.
- Name the role precisely. “A senior infrastructure engineer at a scale-up” elicits more specific pushback than “a technical person.”
- Two to three turns is usually enough. Beyond three turns the experts tend to soften their positions and the discussion loops.
- Treat the output as structured brainstorming, not an oracle. The experts are simulated, so validate any claim that influences a real decision against real sources or real humans.
- Use synthesizing moderation for decisions, none for exploration. When you want to understand a problem space, let the conversation run freely. When you need to act, the synthesizer gives you a conclusion you can pressure-test.