Home Framework Library ToT

ToT

Tree of Thoughts

Branch, evaluate, prune. Skip the middle step and you have just written a longer chain.

AI & Prompt Engineering Pro Framework

What is ToT?

Tree of Thoughts is chain of thought with the ability to back out. Instead of one line of reasoning committed to from the first step, the model generates several candidate paths, evaluates them, abandons the weak ones and develops the rest. Three verbs, and almost everyone using the name drops the middle one.

A chain of thought has no reverse gear. Step one narrows the problem, step two builds on it, and by step five a bad early commitment is unrecoverable — the model will reason competently to a conclusion the first step made inevitable. That is fine when the path is obvious and fatal when it is not. ToT's value is entirely in the evaluation step, because generating three options is easy and a model will do it unasked. Scoring them against stated criteria, explicitly dropping one, and saying what would bring it back — that is the part that makes the branching mean anything, and it is the part a prompt has to demand by name.

Best used for
Complex planning problems creative brainstorming with evaluation math-heavy reasoning any task where the best path is not obvious upfront

Where ToT Came From

Yao et al., NeurIPS 2023 — Princeton and Google DeepMind

"Tree of Thoughts: Deliberate Problem Solving with Large Language Models" (Yao, Yu, Zhao, Shafran, Griffiths, Cao and Narasimhan, NeurIPS 2023) came out of Princeton University with Google DeepMind — the reference implementation still lives under Princeton NLP. Pages crediting DeepMind alone are dropping the half that led it.

The Game of 24 number

The result that made the paper travel. On Game of 24 — reach 24 from four numbers using each once — GPT-4 with chain-of-thought prompting solved 4% of tasks. The same model with Tree of Thoughts solved 74%. The paper also tested creative writing and mini crosswords, chosen because each needs search rather than a single pass.

It was a search algorithm, not a prompt template

Worth being honest about. In the paper, ToT is a framework that runs a language model — generating thought candidates, scoring them with the model as evaluator, and searching the resulting tree breadth-first or depth-first, with real backtracking. What people call a "ToT prompt" is a single-pass imitation of that: one call that generates, evaluates and prunes in sequence. The imitation is genuinely useful and it is not the same thing, and the gap is why a prompt has to be explicit about the steps the algorithm would have performed for you.

The 5 Slots, One at a Time

Each slot is a decision. Leave it out and the model still makes it — just without you.

1
The problem, and why the path is not obvious
The decision plus the constraints that make it genuinely contested.

ToT is only worth its length when several approaches are defensible. Say what makes this one hard — the deadline, the single specialist, the customer whose size breaks the easy option — because those constraints are what the evaluation step will score against.

Weak What architecture should we use for reporting?
Strong Three candidate architectures for a reporting service, given 40-second p95 queries locking the main app, one platform engineer, nobody with Kafka experience, and a five-month renewal deadline.
2
Branch: generate genuinely different candidates
Several approaches, and they must actually differ.

The instruction people write is "consider multiple approaches", and what comes back is three variations of one idea. Say genuinely different, fix the number, and ask for each to be sketched in a few sentences before any evaluation — mixing generation and judgement makes the model quietly discard options as it goes.

Weak Consider a few different options.
Strong Generate three genuinely different candidate approaches — not three variations of one. Sketch each in three or four sentences before evaluating anything.
3
Evaluate: score each against stated criteria
The step that makes it a tree instead of a list.

This is the framework. Give the criteria yourself — a model asked to "evaluate" will invent flattering ones — and require a score with a reason for every candidate against every criterion. Include at least one criterion about the future, since the option that wins on speed usually loses on what it forecloses.

Weak Which one is best?
Strong Score each candidate 1 to 5 with a reason against: time to first value against the deadline, operational burden on one engineer, cost, and what it forecloses in two years.
4
Prune: drop one, and say what would bring it back
An explicit abandonment, not a quiet fade.

Pruning has to be visible. Name the candidate you are dropping and state the change that would revive it — a second engineer, a moved deadline — because that sentence is a record of what the decision actually depended on, and it is the first thing you want when the situation changes in six weeks.

Weak (the weak option is simply never mentioned again)
Strong Drop the weakest candidate explicitly and say what would have to change for it to come back. Then develop the remaining two a level deeper.
5
Decide, with the evidence that would change it
One recommendation, and its falsifier.

End with a single choice — a tree that terminates in "it depends" has done the expensive part and skipped the useful one. Then name the observation that would overturn it, which is what makes the recommendation reviewable rather than merely confident.

Weak All three have merits depending on your priorities.
Strong Recommend one, and name the evidence that would change your mind.

One Task, Before and After

The task: Choosing between three architectures for a reporting service under a hard deadline. Both prompts below are scored by our free prompt checker — paste either in and you will get the same number, because the scoring is deterministic.

18 Without ToT · weak
What architecture should we use for reporting?

Seven words. You get a confident recommendation for an event-streaming architecture, from a model that does not know you have one platform engineer and nobody who has run Kafka.

92 With ToT · strong
Choose between three candidate architectures for a new reporting service, exploring each before committing to one.

Context: A 12-person engineering team, currently one Rails monolith on Postgres. Reporting queries now take 40 seconds at p95 and are locking tables the main app needs. 200 customers, the largest has 40 million rows. Nobody on the team has run Kafka. We have one platform engineer. There is a hard deadline: the largest customer's renewal is in five months and slow reports are named in their renewal risk notes.

Generate three genuinely different candidate approaches - not three variations of one. For each, sketch the shape in three or four sentences before evaluating anything.

Then evaluate each candidate against these criteria, scoring 1 to 5 and saying why: time to first value against the five-month deadline, operational burden on one platform engineer, cost, and what it forecloses in two years.

Then prune. Drop the weakest candidate explicitly, say what would have to change for it to come back, and develop the remaining two a level deeper - the first migration step, the biggest unknown, and how you would test that unknown cheaply in week one.

Only then recommend one, and name the evidence that would change your mind.

Think through the constraints before generating candidates: one platform engineer and no Kafka experience is a harder limit than the deadline, and a candidate that ignores it is not really a candidate.

Write 1,000 to 1,300 words in four headed sections - Candidates, Evaluation, Pruning, Recommendation. For example, use this shape in the Evaluation section:
```
Candidate B - read replica with materialised views.
Time to first value: 5/5 - a week, using tooling the team already runs.
Forecloses in two years: 2/5 - does not solve the 40-million-row customer, so this is bought time.
```
Ensure every candidate is evaluated against all four criteria, and prioritise the deadline and the single platform engineer over elegance. Do not invent benchmark figures, vendor pricing or query timings beyond those given, and do not recommend a candidate you scored lowest without saying explicitly why the scores were wrong.

Ninety-two. The instruction that changes the output most is naming the four criteria — asked to evaluate without them, a model scores each candidate against whatever that candidate happens to be good at.

Ninety-two, and nothing can tell whether the branches were ever real

One check is structurally out of reach, and the two failures unique to this framework are invisible:

Role0 / 8

No persona slot. Available for the asking, worth eight points, and on our scorer a generic persona earns exactly as many as a considered one — so the points say nothing about whether it helped.

Whether the candidates differunscored

Three variations of one idea score exactly as well as three genuinely different approaches. This is the most common way a ToT prompt quietly degrades into a padded chain of thought, and no rubric can see it. Read the three sketches and ask whether any two could be adopted together.

Whether pruning happenedunscored

A "tree" that carries all three candidates to the end and then picks one has not pruned — it has produced a comparison table. The abandonment has to be explicit, and it earns nothing.

The line worth taking from ToT into any decision prompt: say what would have to change for the rejected option to come back. It records what the decision actually rested on, it takes one sentence, and in six weeks when the deadline moves it is the only part of the document you will want.

Copy-Paste Prompt Template

Replace the bracketed placeholders with your specific details.

[The problem, and the constraints that make the path genuinely contested]

BRANCH: [Generate N GENUINELY DIFFERENT candidates — not variations of one. Sketch each in a few sentences BEFORE evaluating anything]

EVALUATE: [Score every candidate against criteria YOU supply — including one about what the choice forecloses later. A score and a reason for each]

PRUNE: [Drop the weakest explicitly, and say what would have to change for it to come back]
[Develop the survivors a level deeper: first step, biggest unknown, how to test it cheaply]

DECIDE: [One recommendation, plus the evidence that would change your mind]

[Length, headed sections, and a fenced sample of the evaluation shape]

When ToT Fits — and When It Does Not

Reach for it
  • Decisions where several approaches are genuinely defensible.
  • Planning problems with sequencing — migrations, rollouts, phased launches.
  • Allocation: splitting a fixed budget or headcount across competing claims.
  • Anywhere a first-pass answer has previously turned out to be a premature commitment.
  • Problems that need search rather than calculation — the class the paper was built on.
Use something else
  • Problems with one obvious path. Branching costs tokens and time and finds nothing.
  • Arithmetic with a single correct method — use chain of thought.
  • Anything needing real-world lookups. Use ReAct; ToT explores inside the model's head.
  • Writing and copy, where three drafts is just three drafts.
  • Fast decisions. This is deliberately the slowest technique here.

10 Ready-Made ToT Prompts

Every prompt below was produced by the Frompting generator with ToT selected — not written by hand for this page. Each is scored by our prompt checker; the median is 95/100. Click one to open it, then copy.

A launch with three viable approaches 95
You are a senior go‑to‑market strategist tasked with designing a comprehensive product‑launch plan that evaluates three distinct launch approaches and selects the most promising one.

**Context**
The product’s business type is **[BUSINESS TYPE: e.g., B2B SaaS, B2C consumer goods, professional services]**. Identify the appropriate target‑segment framework, positioning statement, go‑to‑market motion, pricing model, launch sequencing, and unit‑economics metrics based on that type.

**Task**
1. Generate three separate launch approaches, each covering:
   - Target segment definition (firmographic/technographic/behavioral/psychographic for B2B or demographic/psychographic/behavioral/situational for B2C).
   - Positioning statement using the template “For **[TARGET SEGMENT]** who **[PROBLEM]**, **[PRODUCT]** is the **[CATEGORY]** that **[DIFFERENTIATED BENEFIT]**, unlike **[PRIMARY ALTERNATIVES]** which **[LIMITATION]**.”
   - Go‑to‑market motion appropriate to the business type.
   - Pricing model that aligns with the business model.
   - Launch sequencing steps (e.g., beta phases, pilot, soft launch) with key activation criteria.
   - Core unit‑economics metrics relevant to the model (e.g., CAC payback, LTV:CAC, contribution margin).

2. For each approach, list the pros and cons, required resources, and success‑criteria.

3. Compare the three approaches using a concise decision matrix and recommend the optimal path, explaining why it outperforms the alternatives.

**Output**
- Structured as three numbered approach sections, followed by a decision matrix table and a recommendation paragraph.
- Total length: **≈ 600 words** (≈ 180 words per approach, plus matrix and recommendation).
- Use clear headings, bullet points for lists, and a markdown table for the matrix.

**Quality criteria**
- Accuracy: all frameworks and terminology must match the selected business type.
- Clarity: each approach is self‑contained and easy to compare.
- Actionability: recommendations include concrete next steps and measurable KPIs.

**Boundary**
Do not include any marketing copy, creative taglines, or detailed media plans beyond the launch sequencing steps.

**Assumptions & Clarifications**
State any assumptions you make about the product or market, and ask up to three clarifying questions (e.g., target geography, existing channel partners, budget constraints) before finalizing the plan.

Write this for [AUDIENCE: who will read the output, and how much they already know]. Match the depth, vocabulary and examples to that reader.

Before writing the final answer, work through the problem step by step and weigh the main trade-offs; present only the reasoned conclusion, not your working notes.
385 words · scores 95/100 strong
Choosing between three architectures 95
You are an experienced systems architect tasked with evaluating three candidate technical architectures for a new service.

**Task**
1. List each architecture with a concise description (≈ 30 words each).
2. Identify the key evaluation criteria (e.g., scalability, latency, cost, maintainability, security) and assign a relative weight to each criterion.
3. For each architecture, score it against every criterion on a 1‑5 scale, providing a brief justification (≈ 15 words per score).
4. Compute a weighted total for each architecture and rank them from most to least suitable.
5. Recommend the top architecture and include a short implementation outline (≈ 80 words).

**Required Input**
- [CRITERIA_WEIGHTS]: list of evaluation criteria and their importance weights (e.g., “scalability = 30 %”).
- [ARCHITECTURE_DETAILS]: brief technical summary of each of the three architectures (e.g., “microservices with event‑driven communication”).

**Output**
- A markdown document with the following sections:
  1. **Architectures Overview** – table with Architecture | Description.
  2. **Evaluation Matrix** – table with Architecture | Criterion | Score | Justification.
  3. **Weighted Scores & Ranking** – list each architecture with its total score and rank.
  4. **Recommendation** – selected architecture and implementation outline.

**Quality Criteria**
- Scores and justifications must be logically consistent with the provided criteria weights.
- The recommendation should clearly follow from the highest weighted total.
- Language should be precise, technical, and free of unnecessary filler.

**Boundary**
Do not discuss alternatives beyond the three supplied architectures or propose additional criteria not listed in [CRITERIA_WEIGHTS].

Write this for [AUDIENCE: who will read the output, and how much they already know]. Match the depth, vocabulary and examples to that reader.

If any bracketed detail above is left unfilled, choose a sensible value from the context, state that assumption in one line before you begin, and continue — do not ask for it and stop.

Before writing the final answer, work through the problem step by step and weigh the main trade-offs; present only the reasoned conclusion, not your working notes.
306 words · scores 95/100 strong
Restructuring a team with several options 95
You are a senior organizational strategist tasked with recommending a team restructuring plan.

**Context**: The organization seeks to reorganize a team where multiple viable structures exist.

**Task**:
1. Generate three distinct restructuring options, each describing:
   - Reporting hierarchy
   - Role distribution and key responsibilities
   - Rationale linking the design to expected outcomes (e.g., speed, collaboration, innovation)

2. For each option, evaluate the pros and cons against the following criteria:
   - Alignment with strategic objectives
   - Impact on employee morale and retention
   - Operational efficiency and decision-making speed
   - Resource requirements (budget, tools, training)

3. Rank the three options from most to least suitable, providing a concise justification for the ranking.

**Deliverable**: A structured response in markdown with three sections titled “Option 1”, “Option 2”, and “Option 3”, each containing the design description and a bullet-point pros/cons table. Follow with a “Recommendation” section that lists the ranked options and the rationale for the top choice. 80-640 words ± 10 %.

**Quality criteria**:
- Each option must be internally consistent and realistic.
- Evaluations should be balanced, highlighting both strengths and weaknesses.
- Recommendations must be actionable, specifying any immediate steps needed to implement the top option.

**Boundaries**: Do not assume any specific industry, team size, or budget unless provided.

If any of the following details are unknown, indicate them as placeholders and proceed with assumptions stated explicitly:
- [TEAM_SIZE: number of members]
- [CURRENT_STRUCTURE: brief description of existing hierarchy]
- [STRATEGIC_GOAL: primary objective driving the restructure]
- [RESOURCE_LIMITS: budget or tool constraints]

State any assumptions you make, and ask up to three clarifying questions before finalizing the recommendations.

Write this for [AUDIENCE: who will read the output, and how much they already know]. Match the depth, vocabulary and examples to that reader.

Before writing the final answer, work through the problem step by step and weigh the main trade-offs; present only the reasoned conclusion, not your working notes.
314 words · scores 95/100 strong
The order of a phased data migration 82
You are a data migration strategist.
Your task is to design the optimal sequence of phases for migrating data from a legacy system to a new environment.

**Context**
- The migration must be performed in multiple phases, each moving a subset of data or functionality.
- The goal is to minimize downtime, preserve data integrity, and reduce risk.

**Required information** (please provide or confirm):
- [SOURCE_SYSTEM: brief description of the current system and its data models]
- [DATA_VOLUME: approximate total size of data to be migrated]
- [DEPENDENCIES: key inter‑system or intra‑system dependencies that affect ordering]
- [DOWNTIME_WINDOW: maximum allowable downtime per phase]
- [TARGET_CONSTRAINTS: any limits of the new platform that influence migration steps]

**Approach**
1. Generate several distinct migration pathways, each outlining a possible ordering of phases.
2. For each pathway, evaluate the impact on downtime, risk, and resource usage.
3. Compare the pathways and select the one that best balances the constraints.

**Output**
- A concise table (markdown) listing each proposed pathway with columns: Phase Order, Estimated Downtime, Risk Level, Resource Requirements, and Rationale.
- A brief narrative (150–200 words) summarizing why the chosen pathway is optimal.

**Quality criteria**
- All phases respect the maximum downtime window.
- Dependencies are honored so no phase relies on data not yet migrated.
- Risk assessment is clearly justified.

**Boundary**
Exclude detailed implementation scripts or code; focus only on the high‑level ordering and rationale.

If any of the required information is unknown, state your assumptions explicitly and ask up to three clarifying questions before finalizing the migration plan.

Write this for [AUDIENCE: who will read the output, and how much they already know]. Match the depth, vocabulary and examples to that reader.
285 words · scores 82/100 solid
A scheduling conflict with hard constraints 90
You are a scheduling optimization specialist.
Your task is to help resolve a scheduling conflict that involves multiple competing constraints.

**Context**
- The conflict concerns [EVENTS_OR_TASKS]: a brief description of the items that need to be scheduled.
- The key constraints include [CONSTRAINTS]: any time windows, resource limits, dependencies, or preferences that must be respected.
- The primary stakeholders affected are [STAKEHOLDERS]: who needs the schedule to work for them.
- The desired resolution timeframe is [TIMEFRAME]: when a final schedule should be ready.

**Instructions**
1. **Generate alternative scheduling plans** – create at least three distinct candidate schedules that each satisfy as many constraints as possible.
2. **Evaluate each candidate** – for every plan, list the constraints it meets, the constraints it violates, and the impact of any violations on each stakeholder.
3. **Select the most promising plan** – compare the evaluations, prioritize meeting critical constraints, and choose the schedule that offers the best overall balance.
4. **Provide a concise justification** – explain why the selected plan is preferred, referencing the trade‑offs identified.
5. **Offer a brief implementation checklist** – outline the next steps needed to enact the chosen schedule.

**Output**
- Present the three candidate schedules in a markdown table (columns: Schedule ID, Overview, Constraints Met, Constraints Violated, Stakeholder Impact).
- Follow the table with a short paragraph (≈80–120 words) justifying the selected schedule.
- End with a bullet‑point checklist (3‑5 items) for implementation.

**Quality criteria**
- All schedules must be internally consistent and respect the given constraints where possible.
- Evaluations should be clear, specific, and directly tied to the listed constraints and stakeholders.
- The justification must be logical, concise, and highlight the key trade‑offs.

**Boundary**
Do not propose any changes to the original constraints or stakeholder requirements; work only within the information provided.

If any of the placeholders above are unclear, state your assumptions and ask up to three clarifying questions before proceeding.

Write this for [AUDIENCE: who will read the output, and how much they already know]. Match the depth, vocabulary and examples to that reader.
334 words · scores 90/100 strong
Comparing several pricing structures 95
You are a strategic pricing consultant.
Your task is to devise the optimal pricing structure for a product or service where multiple pricing models are possible.

**Context (fill in the blanks):**
- [PRODUCT_OR_SERVICE: brief description of what is being priced]
- [TARGET_CUSTOMER_SEGMENT: primary market or audience]
- [COST_STRUCTURE: key cost components and approximate margins]
- [REVENUE_GOALS: desired revenue or profit targets]
- [COMPETITIVE_LANDSCAPE: main competitors and their pricing approaches]

**Process:**
1. Generate at least three distinct pricing models (e.g., subscription, tiered, usage‑based, freemium, bundle, dynamic).
2. For each model, create a concise analysis covering:
   - Alignment with the target segment’s willingness to pay
   - Impact on revenue stability and growth potential
   - Sensitivity to cost variations and margin preservation
   - Competitive differentiation and market positioning
3. Compare the models side‑by‑side in a markdown table that scores each criterion on a 1‑5 scale, providing brief justification for each score.
4. Identify the model with the highest overall score and explain why it is the most suitable choice, including any necessary implementation considerations.

**Output:**
- A brief introductory summary (≈50 words).
- The list of pricing models with their analyses (≈150–200 words total).
- A markdown comparison table (no more than 12 rows).
- A concluding recommendation (≈80–100 words).

**Quality criteria:**
- Analyses are data‑driven and logically reasoned.
- Comparisons are transparent, with clear scoring rationale.
- Recommendations are actionable and tied to the provided context.

**Boundary:** Exclude detailed financial forecasts or long‑term market simulations; focus on strategic structure and immediate implications.

If any of the placeholders above are unknown, state your assumptions explicitly and ask up to three clarifying questions before proceeding.
276 words · scores 95/100 strong
Splitting a budget across four channels 96
You are a strategic marketing analyst tasked with allocating a limited marketing budget across four distinct channels.

**Goal:** Distribute the total budget to maximize overall marketing effectiveness based on the company’s objectives and channel characteristics.

**Key dimensions:**
- Total budget amount [ BUDGET: specify the total dollar amount].
- The four marketing channels [ CHANNELS: list the specific channels, e.g., social media, search, email, display].
- Primary business objective [ OBJECTIVE: e.g., brand awareness, lead generation, sales lift].
- Any channel‑specific constraints [ CONSTRAINTS: e.g., minimum spend, caps, regulatory limits].

**Task:**
1. Generate at least three distinct allocation scenarios, each exploring a different strategic emphasis (e.g., aggressive digital, balanced mix, conservative traditional).
2. For each scenario, calculate the proposed spend per channel, justify the reasoning, and estimate the expected impact on the stated objective using simple, transparent assumptions.
3. Compare the scenarios in a concise table that includes total spend, expected impact rating (high/medium/low), and key trade‑offs.
4. Recommend the single best scenario, explaining why it offers the optimal balance of impact and risk.

**Output format:**
- Brief introductory summary (≈50 words).
- Three scenario sections, each ≤120 words, with clear headings.
- Comparison table (markdown) with columns: Scenario, Channel 1 Spend, Channel 2 Spend, Channel 3 Spend, Channel 4 Spend, Impact Rating, Main Trade‑off.
- Recommendation paragraph (≈80 words).

**Quality criteria:**
- All calculations are transparent and based on the provided numbers.
- Justifications reference the objective and any listed constraints.
- The recommendation clearly follows from the comparative analysis.

**Boundary:** Do not propose any channels or constraints not listed in the placeholders.

If any of the placeholders above are unknown, state your assumptions explicitly and ask up to three clarifying questions before finalizing the allocation.

Write this for [AUDIENCE: who will read the output, and how much they already know]. Match the depth, vocabulary and examples to that reader.
306 words · scores 96/100 strong
A negotiation with several openings 95
You are a strategic negotiation planner.
Your task is to design a comprehensive negotiation plan that accommodates multiple possible opening moves and guides the negotiator through each branch to the most effective outcome.

The plan is for **[PARTICIPANT ROLES]:** list the parties involved (e.g., buyer, supplier, mediator).
The primary objectives are **[NEGOTIATION GOALS]:** specify the key outcomes each side seeks (e.g., price reduction, delivery terms, partnership scope).
Key constraints to consider are **[LIMITING FACTORS]:** such as budget caps, time windows, legal restrictions, or relationship sensitivities.

Produce a structured plan of **≈ 500 words** organized as follows:
1. **Context Summary** – brief overview of the negotiation setting and goals.
2. **Opening Scenarios** – enumerate 3–4 distinct opening moves (e.g., aggressive demand, collaborative proposal, data‑driven offer). For each, outline the immediate response options.
3. **Branch Evaluation** – for each opening, list the pros, cons, and risk level (low/medium/high).
4. **Decision Path** – recommend the most promising opening and the subsequent steps, explaining why it outperforms the alternatives.
5. **Tactical Checklist** – concise bullet list of actions, talking points, and fallback positions to prepare before the negotiation.

Quality criteria:
- **Clarity:** each scenario and recommendation must be easy to follow without ambiguity.
- **Practicality:** suggestions should be actionable with realistic resources and timeframes.
- **Adaptability:** the plan must allow quick adjustment if the counterpart deviates from the anticipated path.

Exclude any content that assumes specific industry knowledge, proprietary data, or confidential details not provided. If additional critical information is needed, state your assumption and ask up to three clarifying questions before finalizing the plan.

Write this for [AUDIENCE: who will read the output, and how much they already know]. Match the depth, vocabulary and examples to that reader.

Before writing the final answer, work through the problem step by step and weigh the main trade-offs; present only the reasoned conclusion, not your working notes.
304 words · scores 95/100 strong
Cutting costs 20% with options on the table 88
You are a strategic cost‑reduction consultant.
Your client needs a plan to lower overall expenses by **20 %** and wants several viable options to consider.

**Task**
1. Generate at least three distinct cost‑cutting strategies, each outlining a clear set of actions.
2. For each strategy, assess its potential impact, required resources, implementation risk, and any trade‑offs.
3. Rank the strategies from most to least favorable based on the assessments.

**Context needed** – please provide the following details (or indicate if unknown):
- **[INDUSTRY]**: the sector the organization operates in.
- **[CURRENT_COST_BASE]**: total current monthly or annual expenses (e.g., $X).
- **[TIMEFRAME]**: period over which the 20 % reduction must be achieved.
- **[KEY_CONSTRAINTS]**: any limits such as regulatory compliance, staffing levels, or technology dependencies.
- **[STAKEHOLDER_PREFERENCES]**: priorities of decision‑makers (e.g., preserving workforce, maintaining service quality).

If any of these are unknown, state your assumption for that item and proceed.

**Output**
- A concise executive summary (≈ 80 words).
- A table listing each option with columns: *Strategy Name*, *Estimated Savings (%)*, *Required Resources*, *Implementation Risk (Low/Medium/High)*, *Key Trade‑offs*.
- A brief rationale (≈ 150 words) explaining the ranking and recommending the top option.

**Quality criteria**
- Strategies must be realistic and actionable within the given timeframe.
- Assessments should be based on logical reasoning and clearly articulated assumptions.
- The ranking should reflect a balanced view of savings versus risk and trade‑offs.

**Boundary**
Do not suggest measures that would compromise legal compliance or core safety standards.

Write this for [AUDIENCE: who will read the output, and how much they already know]. Match the depth, vocabulary and examples to that reader.

Use this table shape, one row per item, filling values from your analysis:

| *Strategy Name* | *Estimated Savings |
| --- | --- |
290 words · scores 88/100 strong
Choosing three of eight feature requests 95
You are a strategic product decision analyst.
Your task is to help the product team select exactly three feature requests out of eight to develop in the upcoming quarter.

**Context**
- The eight candidate features are: [FEATURE_LIST: enumerate the eight feature ideas].
- The primary selection criteria (e.g., customer impact, development effort, revenue potential, strategic alignment) are: [CRITERIA: list the key factors to weigh].
- Stakeholder weighting or priority information is: [STAKEHOLDER_WEIGHTS: describe any known preferences or scores].

**Task Steps**
1. Generate three distinct evaluation pathways, each exploring a different combination of criteria emphasis (e.g., high impact vs. low effort, short‑term revenue vs. long‑term strategy, etc.).
2. For each pathway, score all eight features against the chosen emphasis, providing a brief justification (no more than one sentence per feature).
3. Summarize the top‑scoring three features for each pathway.
4. Compare the three pathway outcomes, highlighting trade‑offs and any consensus among them.
5. Recommend the final three features to build this quarter, explaining why this set best balances the overall criteria.

**Output**
- Present the three pathways in separate markdown tables titled “Pathway 1”, “Pathway 2”, and “Pathway 3”.
- Follow with a concise comparison paragraph (≈80 words).
- End with a recommendation section (≈60 words) listing the three chosen features as a bullet list.

**Quality Criteria**
- Scores must be transparent and based solely on the provided criteria.
- Justifications should be specific, avoiding vague language.
- The recommendation should reflect a clear trade‑off analysis.

**Boundary**
Do not propose any features beyond the eight listed, and do not fabricate additional criteria.

If any of the required information is missing, state your assumptions explicitly and ask up to three clarifying questions before proceeding.

Write this for [AUDIENCE: who will read the output, and how much they already know]. Match the depth, vocabulary and examples to that reader.

Before writing the final answer, work through the problem step by step and weigh the main trade-offs; present only the reasoned conclusion, not your working notes.
331 words · scores 95/100 strong

Scores range from 82 to 96. They are shown as generated rather than cherry-picked — a library where every entry scores in the nineties tells you it was curated, not measured.

ToT vs the Alternatives

CoT Chain of Thought

The parent, and the choice is about reversibility. CoT commits to one line of reasoning; ToT can abandon a branch. Use CoT when the path is clear, ToT when a wrong first step would be unrecoverable.

ReAct Reason, Act, Observe

The other extension of CoT, in the opposite direction. ReAct goes outward to tools and real observations; ToT goes wider inside the model. They combine — branch, then check the best branches against reality.

SWOT Strengths, Weaknesses, Opportunities, Threats

The business framework with the same instinct and no search. SWOT surveys one position; ToT explores several and discards.

SCOPE Situation, Constraints, Objectives, Plan, Evaluation

Related through evaluation criteria, which both insist on stating up front. SCOPE plans an investigation; ToT explores a decision.

OODA Observe, Orient, Decide, Act

For the same decision under time pressure. OODA is built for a moving situation; ToT assumes you can afford to explore before committing.

Five Ways People Get ToT Wrong

1
Three variations of one idea

The defining ToT failure. "Consider multiple approaches" gets you the same architecture with different databases. Say genuinely different, and check afterwards whether any two candidates could be adopted together — if they could, they were never alternatives.

2
Evaluating without criteria

A model asked to evaluate will invent criteria, and it will invent ones each candidate happens to satisfy. Supply them, and include one about what the choice forecloses later.

3
No actual pruning

Carrying every candidate to the end is a comparison table, not a tree. Require an explicit drop and the condition that would reverse it.

4
Generating and judging at once

Ask for candidates and evaluation in one breath and the model discards options while generating them, so you never see the ones it thought were weak. Sketch all of them first.

5
Ending in "it depends"

The expensive part of ToT is exploration; stopping before the commitment wastes it. Require one recommendation and the evidence that would overturn it.

6
Using it where the path is obvious

ToT is the slowest and most expensive technique on this site. On a problem with one sensible answer it produces two decoy candidates and a lot of confident scoring.

ToT Questions

What is Tree of Thoughts prompting?

A technique where the model generates several candidate lines of reasoning, evaluates them against criteria, abandons the weak ones and develops the rest — instead of committing to a single chain from the first step.

Where does it come from?

"Tree of Thoughts: Deliberate Problem Solving with Large Language Models" (Yao et al., NeurIPS 2023), from Princeton University with Google DeepMind.

How much better is it than chain of thought?

On the paper's Game of 24 task, GPT-4 with chain-of-thought solved 4% of problems and the same model with Tree of Thoughts solved 74%. That gap is specific to problems that need search — it is not a general multiplier.

Is a "ToT prompt" the same as the paper's method?

No, and it is worth knowing. The paper describes a search procedure that runs the model repeatedly with real backtracking. A single prompt imitates that in one pass, which is useful but weaker — and it is why the prompt has to spell out the generate, evaluate and prune steps explicitly.

Why do my candidates all look the same?

Because "consider multiple approaches" does not ask for difference. Say "genuinely different, not variations of one", and test the output by asking whether any two could be adopted together.

When should I use chain of thought instead?

When there is one sensible method and the work is calculation rather than choice. ToT earns its cost only when a wrong first commitment would be expensive to unwind.

Generate a ToT Prompt Instantly

Skip the manual template — Frompting applies ToT to your topic in one click.

Try it Free