Expert Assistant Prompt for ChatGPT (COT-Safe Template)

Turn any session into a world-class expert assistant. This prompt gets you clear answers, concise reasoning summaries, viable alternatives, and a step-by-step action plan—without exposing chain-of-thought.

What This Prompt Does

This world-class expert assistant prompt makes models respond like a top-tier specialist: clear conclusions first, a concise reasoning summary (no private chain-of-thought), alternatives you might not consider, and an actionable plan. It helps founders, analysts, students, and teams get reliable, structured outputs that are easy to execute immediately.

Quick Start: Copy-Paste Prompt

Paste the following into your AI tool.

[SYSTEM ROLE / INSTRUCTIONS]
You are a domain-expert assistant. Always provide:

a clear, direct answer;
a concise reasoning summary with key steps/assumptions (no hidden chain-of-thought or internal scratch work);
2–3 alternative approaches or perspectives;
a practical, prioritized action plan.
If the question is broad, break it into parts. Adopt the relevant professional role when helpful. Ask only for essential missing info.

[USER PROMPT]
“From now on, act as my expert assistant. For every request:

Give the direct answer first.
Then summarize how you got there (key steps, assumptions, and—when applicable—sources or calculations) without revealing private chain-of-thought.
Offer alternative solutions or perspectives I might have missed.
End with a concrete, prioritized action plan I can apply immediately.
If the topic is broad, break it into parts. Push analytical rigor while staying precise and verifiable.”

[OPTIONAL: NEGATIVE PROMPT / DO-NOTS]

Do not expose internal chain-of-thought, hidden scratch work, or proprietary training data.
Avoid vague generalities; quantify or exemplify when possible.
Do not fabricate sources; cite only if you are confident.
Avoid sensitive personal data unless explicitly provided and necessary.

Model-Specific Settings (Recommended)

ChatGPT (OpenAI): temperature 0.2–0.4; max tokens: as needed (e.g., 1200–2000); JSON mode/tools: off unless a schema is requested; reasoning: allow brief rationale only; safety: default.
Claude: temperature 0.2–0.4; max tokens generous (e.g., 3000–4000); encourage “concise reasoning summary” not full chain-of-thought; use tool use when citations/data extraction help.
Gemini: style “precise”; safety: default; long-context: summarize sources; emphasize bullet-pointed reasoning summaries and concrete action steps.
Image models (if relevant): steps 30–50; CFG 5–7; seed unset (for variety); aspect 16:9 for hero images.

Step-by-Step (How to Use)

Open your AI tool and start a new chat.
Paste the prompt block above and add your task/question.
Apply the recommended model settings for concise, reliable outputs.
Review with the checklist below (clarity, assumptions, alternatives, actionability).
Refine with follow-up mini-prompts (see “Fast refiners”).

Variations & Personas

Variant A — Analyst/Consultant: “Adopt a management consultant lens; quantify assumptions and show a brief sensitivity analysis.”
Variant B — Engineer/Builder: “Adopt a senior engineer lens; propose architecture, trade-offs, and risks; include a rollout checklist.”
Variant C — Teacher/Coach: “Adopt a master teacher lens; use plain language, analogies, and 3 practice tasks with solutions.”

Example Inputs & Expected Outputs

Example 1 — Input: “Evaluate three pricing models for a $49 online course targeting HR managers; recommend one.”
Expected Output (abridged):

Direct answer: Tiered value-based pricing with a $49 anchor + bundle upsell.
Reasoning summary: Audience price sensitivity; CAC; conversion benchmarks; LTV.
Alternatives: Freemium lead magnet; cohort-based premium; annual license to teams.
Action plan: A/B test 2 price pages; measure CVR, ARPU, refund rate; 14-day timeline.

Example 2 — Input: “Create a 5-point plan to reduce churn for a SaaS with 3% monthly churn.”
Expected Output (abridged):

Direct answer: Onboarding fixes + lifecycle emails + in-app nudges + save offers + win-backs.
Reasoning summary: Churn sources: activation debt, low feature adoption.
Alternatives: Annual prepay incentives; usage-based pricing pilot.
Action plan: Instrument activation funnel; run 2 retention experiments/2 weeks.

Example 3 — Input: “Draft a study plan to learn SQL in 30 days.”
Expected Output (abridged):

Direct answer: Daily 45-min schedule with topic ladder from SELECT→JOIN→WINDOW→CTE.
Reasoning summary: Skill gradient; spaced repetition boosts retention.
Alternatives: Weekend bootcamps; project-first approach.
Action plan: 4 projects (analytics report, cohort retention, funnel chart, KPI dashboard).

Quality Checklist & Iteration Tips

Covers answer → reasoning summary → alternatives → action plan.
Specific numbers/examples where possible; assumptions are explicit.
Constraints acknowledged; risks and trade-offs listed.
Formatting: bullets, short paragraphs, bold section headers.
Accuracy: avoid speculation; cite or mark as assumption.

Fast refiners:

“Tighten to 150 words.”
“List 3 quantifiable KPIs to track.”
“Give me a one-page executive brief.”
“Translate this plan into a 10-step checklist.”
“Propose a 2-week experiment roadmap.”

Troubleshooting

Issue: Output too generic → Fix: Add constraints (audience, budget, timeline) and ask for a KPI.
Issue: Hallucinations → Fix: Request sources or say “If unknown, say so.”
Issue: Length or format drift → Fix: Re-paste the prompt and demand the 4-part structure.
Issue: Missing data → Fix: Ask for one clarifying question only; proceed with stated assumptions.

Safety & Compliance

No personal/sensitive data unless the user provides it and it’s necessary. Prefer original wording; avoid copyrighted text. Request brief reasoning summaries—do not expose hidden chain-of-thought or private scratch work. Note potential bias and include trade-offs where relevant.

TL;DR

Use a single prompt to force a 4-part expert response.
Get direct answers, concise reasoning summaries, alternatives, and an action plan.
Stay chain-of-thought safe while maximizing rigor.