Tree-of-Thought
Prompt Generator.
Free builder for tree-of-thought prompts. Branch, evaluate, prune, and pick the best path through a problem.
Standard ToT, recursive multi-level, and deliberative council formats. Pick the right one for your question.
Describe what you want
3 prompt variations
Click Copy to use# PROBLEM [state the problem] # TREE OF THOUGHT 1. Generate 3 distinct candidate approaches to this problem. Label them A, B, C, ... 2. For each approach, describe the first step and why it might work. 3. Evaluate: self-score each branch 1 to 10. 4. Pick the best approach and justify the choice. 5. Develop the chosen approach to a full answer. # OUTPUT Show all branches, the evaluation, the pick, and the developed answer.
# PROBLEM [state the problem] # RECURSIVE TREE OF THOUGHT Search depth: standard (3 levels). At each node, branch into 3 children. # PROTOCOL 1. From the root problem, produce 3 candidate approaches. 2. For each approach, project the next reasoning step. 3. Evaluate nodes at each level using: self-score each branch 1 to 10. 4. Prune branches that are dominated or clearly off-track. 5. On promising branches, recurse: from that node, branch again. 6. When depth is reached, rank leaf nodes and pick the best full path. # OUTPUT Render the tree (indented), show scores at each node, mark pruned branches, and deliver the winning path as the final answer.
# ROLE You are a council of 3 experts, each defending a distinct approach. # PROBLEM [state the problem] # ROUND 1 (independent proposals) Each expert writes their approach and reasoning, without seeing the others. # ROUND 2 (evaluation) Each expert reads the others' proposals and critiques them. Strategy: self-score each branch 1 to 10. # ROUND 3 (synthesis) A moderator reviews the critiques and either picks a winner, proposes a hybrid, or asks one expert to extend their approach. # OUTPUT Show all three rounds. End with the final recommendation.
Under the hood
Why branching beats linear for hard decisions.
A single chain commits to the first plausible approach and defends it. Branching forces the model to explicitly generate alternatives, which prevents premature lock-in on a mediocre path.
Without explicit evaluation, the model picks the last branch it wrote (recency bias). Scoring, ranking, or elimination forces a comparison step that produces genuinely better selections.
A single-level tree often picks the wrong root branch because the best path only reveals itself two or three steps in. Recursive branching with pruning captures that, at the cost of more tokens.
Related free tools
Specialized generators for specific tasks.
Chain-of-Thought Prompt Generator
Linear step-by-step when branching is not needed.
Reasoning Prompt Generator
Six reasoning modes, adjustable rigor.
DeepSeek Prompt Generator
Reasoning-model tuned prompts with verification.
All Prompt Generators
Browse 35+ free generators.
FAQ
Questions about tree-of-thought prompting.
What is tree-of-thought prompting?+
Tree of Thought (ToT) is a prompting technique where the model generates several candidate approaches, evaluates them, prunes weak ones, and develops the best. It extends chain-of-thought from a single line of reasoning into a branching search. Yao et al. introduced the formal version in 2023.
When does ToT beat chain-of-thought?+
ToT wins when there are multiple plausible approaches and it is not obvious which one is right. Strategic planning, creative ideation, hard math with multiple solution paths, root cause diagnosis. For linear problems with one obvious approach, plain CoT is faster and equally accurate.
How many branches should I use?+
Three is the sweet spot. Two forces a binary that often feels contrived. Four or five wastes tokens on weak branches. Use five only when you genuinely want to survey a wide space (brainstorming). For most decisions three is enough to surface the real tradeoffs.
What is the recursive variant doing differently?+
The recursive variant branches at each level, not just the root. You get a genuine search tree with scores at each node and pruning between levels. It is slower and produces more tokens, but for hard problems where the right path is not obvious at level one it meaningfully outperforms the single-level version.
What is the deliberative council variant?+
It frames the branches as experts with distinct perspectives. Each proposes their approach independently, then they critique each other, and a moderator synthesizes. This format is strong for contested decisions where multiple valid frames exist (ethics, policy, product strategy). The social framing pushes the model to steelman each branch rather than dismissing weaker ones.
How does evaluation work?+
The model self-scores each branch using the strategy you pick. Self-scoring 1-10 is the simplest. Ranking forces relative ordering. Eliminate-dominated is stricter and often produces cleaner trees. Monte Carlo explores each branch a few steps deep before committing. Pairwise compare is slow but tends to produce the most thoughtful evaluations.
Does ToT work with any model?+
Yes, though it rewards stronger models. GPT-4o, Claude Opus, and Gemini Pro handle branching gracefully. Smaller open-weight models sometimes collapse the branches into one and just produce a CoT. If that happens, add an explicit instruction that each branch must take a distinct approach.
Is ToT overkill for most prompts?+
Often, yes. A lot of prompts do not benefit from branching because there is one clear path. Use ToT when you can name two or more plausible approaches that a smart person might genuinely disagree on. If you cannot, use chain-of-thought instead.