Ensemble prompting

Ensemble prompting runs the same task with multiple different prompts (or models) and aggregates the responses — typically majority vote, weighted average, or a final reconciliation model.

Ensemble prompting is the prompt-level analogue of model ensembling. Use several distinct prompt strategies (CoT, few-shot, persona, constraint) on the same task, then aggregate. The technique improves reliability on hard tasks at the cost of N× tokens. Production in 2026 uses ensembling primarily for high-stakes judging (LLM jury) and for hard reasoning tasks where redundancy is cheaper than failure. Mixture-of-agents is a specific form where each branch is a different model.

When to use ensemble prompting

• High-stakes single-shot answers.
• Hard reasoning where being right matters more than cost.

Common mistakes

• Ensembling identical prompts — no diversity, no benefit.
• Forgetting to log per-branch outputs — you can't debug an ensemble that's wrong.

FAQ

Related terms

• Mixture of agents — Mixture of agents is an inference pattern where multiple specialised LLM agents run in parallel and a router aggregator combines their outputs into a single answer — higher quality than any single agent at higher cost.
• Self-consistency — Self-consistency runs the same prompt multiple times at non-zero temperature and picks the most common final answer, raising accuracy on reasoning tasks.
• LLM jury — An LLM jury is an evaluation pattern where multiple LLM judges score the same output, and their scores are aggregated to reduce single-judge bias.

Last updated: 2026-06-01. Raw markdown: https://promtable.com/glossary/ensemble-prompting.md.