| name | deep-research-prompt |
| description | Write structured prompts for agentic deep research tools (OpenAI Deep Research, Perplexity, Gemini, Manus AI). Use when the user wants to craft a research prompt, prepare a deep research mission, write a prompt for autonomous research agents, or optimize a research query for AI tools. |
Deep Research Prompt Writing
Write mission-briefing-style prompts that maximize output quality from autonomous deep research agents. These agents run independently for 5-30 minutes, execute hundreds of searches, and read thousands of pages. A vague prompt wastes massive compute and returns generic summaries.
The operator's mindset: you are a strategic project manager deploying a digital analyst, not a person typing a search query.
Core Formula
Every deep research prompt must contain these six components:
Effective Prompt = Role + Objective + Context + Constraints + Safeguards + Format
Step-by-Step Construction
1. Role & Audience
Define WHO the agent is and WHO reads the output. This forces the model into the right region of its latent space.
Act as a Senior Strategy Consultant reporting to a board of directors.
Tone: executive, data-driven, focused on ROI and strategic risk.
2. Mission Objective (Decision-Ready)
Frame the objective as a decision to be made, not a topic to explore. Use If-Then conditional logic to prevent open-ended drift.
Decide whether [Company] should invest in [Technology] in 2026.
If the technology is not projected to be commercially viable within 36 months, explicitly recommend "Wait."
Bad: "Tell me about solid state batteries."
Good: "Should we invest in solid state battery manufacturing? If unit costs won't drop below $X by 2028, recommend Hold."
3. Epistemic Constraints
Control source quality, time window, and search depth.
Source hierarchy (always specify):
Prioritize peer-reviewed journals, official government datasets, and audited financial disclosures.
Exclude SEO-spam blogs, promotional vendor materials, and unverified news outlets.
Temporal bounds (always specify):
Focus on data published Q3 2024 to present. Flag anything older as "historical context" and do not use it in forward projections.
Depth enforcement (always specify):
Do not stop at surface-level summaries. If Source A reports 10% growth and Source B reports 50%, investigate and explain the methodological divergence.
4. Cognitive Safeguards
Embed these three protocols to prevent hallucination, confirmation bias, and single-source risk.
Data Triangulation: "Find at least 3 independent sources for every quantitative claim. Present the range, not a single number."
Devil's Advocate: "After your main conclusion, dedicate a section to 'The Counter-Thesis.' Execute searches specifically designed to disprove your primary finding."
Recursive Self-Correction: "Before finalizing, review your own findings for source bias or logical gaps. Execute a second round of targeted searches to resolve them."
5. Output Architecture
Force rigid structure on the deliverable. Without this, agents produce impenetrable walls of text.
Always require these four sections:
- Executive Summary - Max 1 page, Bottom Line Up Front (BLUF)
- Strategic Matrix - Comparison table (options vs. cost, feasibility, risk, scalability)
- Confidence Assessment - Low/Medium/High per major claim, based on source quality and volume
- Gap Analysis - What the agent tried to find but could not verify; areas requiring human intelligence
Prompt Template
## Role
Act as [specific role]. The audience is [specific audience], so the tone should be [tone qualities].
## Objective
The mission is to determine whether [specific decision].
- If [condition A], recommend [action A].
- If [condition B], recommend [action B].
## Context
[2-3 sentences of background the agent needs to orient its search]
## Constraints
- **Sources**: Prioritize [source types]. Exclude [source types].
- **Time window**: Focus on [date range]. Flag older data as historical context only.
- **Depth**: Do not accept surface-level answers. Investigate contradictions between sources.
- **Triangulation**: Find 3+ independent sources for every quantitative claim.
## Cognitive Protocols
- After forming your primary conclusion, write a "Counter-Thesis" section with evidence against it.
- Before finalizing, self-review for bias or gaps and run a second search round.
## Output Format
1. **Executive Summary** (1 page max, BLUF methodology)
2. **Strategic Comparison Matrix** (table comparing options on [variables])
3. **Confidence Scores** (Low/Medium/High per claim)
4. **Gap Analysis** (what you could not verify)
5. **Sources** (full citations, grouped by section)
Platform Selection Guide
| Platform | Best For | Speed |
|---|
| OpenAI Deep Research | Exhaustive synthesis, complex reasoning, executive reports | 5-30 min |
| Perplexity Deep Research | Time-sensitive market intelligence, fact-checking, source transparency | ~3 min |
| Manus AI | Multi-day workflows, continuous monitoring, offline execution | Async |
| Google Gemini Deep Think | Massive data ingestion, multimodal synthesis, expert-level problems | Variable |
References