Deep research

Deep Research

Use this skill when the user asks for a deep dive, research report, source-backed answer, literature scan, competitive review, policy summary, market landscape, technical investigation, trend analysis, or evidence comparison on any topic.

Core Workflow

1. Scope the research question:
   • Topic, decision or purpose, audience, geography, timeframe, depth, and preferred output format.
   • Whether the user wants a quick brief, detailed report, comparison table, annotated bibliography, timeline, or recommendation.
2. Ask only the clarifying questions needed to avoid wasted research. If the request is clear enough, state assumptions and proceed.
3. Build a source plan:
   • Prefer primary sources, official documents, standards, datasets, peer-reviewed research, reputable expert bodies, and high-quality reporting.
   • Use secondary sources to discover leads, terminology, and opposing views, but verify important claims against stronger sources.
   • For topics that may have changed recently, browse for current sources and check publication or update dates.
4. Gather and evaluate evidence:
   • Track what each source actually supports.
   • Compare agreement, disagreement, methodology, incentives, and recency.
   • Separate established facts from inference, speculation, and opinion.
5. Synthesize the answer:
   • Start with the answer or thesis, then provide the supporting analysis.
   • Explain the strength of evidence and confidence level.
   • Highlight open questions, missing data, and material caveats.
6. Cite sources clearly:
   • Include links for every non-obvious factual claim that depends on research.
   • Preserve source dates when recency matters.
   • Avoid long quotations; paraphrase unless a short quote is necessary.

For source-quality rules, report templates, and citation patterns, read research framework when the task needs a structured research deliverable.

Source Standards

Use stronger sources for stronger claims:

• Best for facts and current rules: official documentation, laws, regulations, standards, public datasets, company filings, and primary announcements.
• Best for scientific claims: systematic reviews, clinical guidelines, peer-reviewed studies, preprints with caveats, and recognized expert bodies.
• Best for markets and organizations: annual reports, investor materials, regulatory filings, reputable industry research, pricing pages, and direct product documentation.
• Best for events: multiple reputable news reports, direct statements, public records, and dated primary material.

Avoid relying on unsourced summaries, SEO content, anonymous reposts, AI-generated pages, or single-source claims when the conclusion matters.

Safety and Quality Boundaries

• Do not fabricate sources, citations, quotes, page numbers, statistics, or certainty.
• Do not treat one source as conclusive when the topic is disputed or evidence is thin.
• Do not provide individualized medical, legal, financial, or safety-critical advice as a substitute for a qualified professional.
• When research affects consequential decisions, make the decision criteria, assumptions, risks, and confidence explicit.
• For controversial topics, represent the strongest credible positions fairly and distinguish evidence quality from popularity.

Output Guidance

• Lead with a concise answer, then provide the research detail.
• Use headings, bullets, tables, or timelines when they improve scanability.
• Include "what we know", "what is uncertain", and "sources" when useful.
• State confidence as high, medium, or low, with a short reason.
• For recommendations, explain the decision logic and tradeoffs.
• End with the most useful next research steps only when gaps remain.