Prompt Engineering / Reasoning Systems
Generate multiple reasoning branches at each step, evaluate them, and prune poor branches before continuing.
Why This Prompt Exists
Chain-of-thought follows one path. If that path is wrong, you never recover. Tree-of-thoughts explores multiple paths and picks the best.
You get:
- getting stuck on the first reasonable-sounding path (even if wrong)
- no way to recover from early bad decisions
- missing better solutions that require different first steps
- linear thinking when branching would help
- no systematic way to evaluate alternatives
But tree-of-thoughts solves this:
- branching: generate multiple possible next steps at each decision point
- evaluation: score each branch (promising vs. dead end)
- pruning: drop low-scoring branches
- expansion: continue from promising branches
- selection: choose best final path
Without branching, you commit too early.
This prompt implements tree-of-thoughts reasoning for complex problems.
The Prompt
Assume the role of a tree-of-thoughts reasoning engine that explores multiple solution paths. Your task is to solve a problem by exploring, evaluating, and pruning multiple reasoning branches. Generate: 1. PROBLEM RESTATEMENT - Restate the problem in your own words 2. INITIAL BRANCHING (Level 1) - Generate 3-5 different approaches to the problem - For each approach, state the first step 3. BRANCH EVALUATION - For each branch, score: (1 = dead end, 5 = very promising) - Brief rationale for each score 4. PRUNING - Discard branches with score ≤ 2 - Keep branches with score ≥ 3 5. DEEPEN SELECTED BRANCHES (Level 2) - For each kept branch, generate next step possibilities (2-3 per branch) - Show the growing tree structure 6. CONTINUE UNTIL SOLUTION OR MAX DEPTH - Repeat evaluation, pruning, and deepening 7. FINAL SOLUTION - The best path through the tree - Why this path is最优 8. ALTERNATIVE PATHS (briefly) - What promising paths were pruned and why INPUTS: Problem to solve: [PASTE THE PROBLEM] Problem type: [LOGIC / PLANNING / CREATIVE / OPTIMIZATION / OTHER] Branching factor (how many alternatives per step): [3 / 5 / 7] (higher = more thorough but more expensive) Max depth (how many steps): [3 / 5 / 7 / 10] Pruning threshold (score to keep): [2/5, 3/5, etc.] Model: [GPT-4 / CLAUDE / GEMINI] RULES: - Branch widely at first (explore options), prune aggressively later (focus on promising) - Evaluation criteria should match problem type (for logic: correctness; for creative: novelty+feasibility) - Don't prune too early — some branches look weak but become strong after a few steps - Track visited states to avoid loops (same reasoning step twice) - If tree grows too large (exponential), increase pruning aggressiveness
How To Use It
- Use for complex problems with multiple viable approaches (strategy, planning, creative tasks).
- Start with moderate branching (3 options per step) to avoid exponential explosion.
- Prune aggressively (keep only top 1-2 branches per level) for large problems.
- Visualize the tree structure to understand trade-offs between paths.
- Don’t use for simple problems — overhead isn’t worth it.
Example Input
Problem to solve:
“How can a small e-commerce company increase customer retention by 20% within 6 months with a $10,000 budget?”
Problem type:
“PLANNING / OPTIMIZATION”
Branching factor:
“3”
Max depth:
“4”
Pruning threshold:
“3/5”
Why It Works
Chain-of-thought commits to one path early — which is fine for simple problems but disastrous for complex ones.
This framework improves outcomes by forcing:
- branching exploration (not just one path)
- explicit evaluation (score each branch)
- pruning (drop dead ends early)
- deepening (explore promising branches further)
- comparison (why the chosen path is best)
Great tree-of-thoughts exploration doesn’t find the first solution — it finds the best solution by exploring many paths.
Build Better AI Systems
Subscribe for advanced prompt engineering, AI coding tools, debugging frameworks, and practical strategies for developers and engineers.