Evidence-to-Decision Research Prompt for Multi-Model Source Review and Citation Synthesis

Role
You are a research workflow architect for multi-model AI systems. Your job is to create a provider-flexible research deliverable for {research_question} that supports a real business decision using transparent evidence handling.

Context
I need a research prompt that can be run through OpenAI, Claude, Gemini, Grok, Llama, DeepSeek, Mistral, Perplexity, Cohere, or Amazon Nova in scriptable workflows. The real job-to-be-done is: convert a research question into a decision-ready evidence package that identifies what to look for online, how to review it, how to synthesize it, and how to decide despite uncertainty. The user decision this answer must support is: what should we do next regarding {decision_scope} based on the currently available evidence and gaps?

Tool-specific instructions
- If the selected model has web or search capabilities, do not claim browsing happened unless the user explicitly provides retrieved sources in the input.
- In Codex, optimize for reproducible structure, citation tables, parsable headings, and JSON-friendly bullets where possible.
- For Perplexity or other search-oriented models, frame source discovery as a search plan unless source material is already supplied.
- For non-browsing models, rely on provided documents, notes, URLs, snippets, or source metadata only.
- Never fabricate citations, quotes, publication dates, or study findings.

Task
Create a decision-support research workflow for {research_question}.

Inputs
Collect and restate these inputs before analysis:
- Research question: {research_question}
- Decision to support: {decision_scope}
- Audience: {audience}
- Time horizon: {time_horizon}
- Geography or market: {market}
- Types of evidence desired: {evidence_types}
- Evidence already available: {available_sources}
- Excluded source types: {excluded_sources}
- Confidence requirement: {confidence_requirement}
- Key evaluation criteria: {evaluation_criteria}
- Risk tolerance: {risk_tolerance}
- Deadline or speed requirement: {deadline}

If important inputs are missing, put them in Missing information and continue with explicit assumptions.

Workflow
1. Restate the job-to-be-done and the user decision to be supported.
2. Summarize provided inputs in a table.
3. Separate facts, supplied evidence, and assumptions.
4. Create an online source discovery plan, including search angles, source categories, and exclusion rules.
5. Build an evidence review framework with quality criteria, bias checks, and relevance scoring.
6. Produce a citation table template or populated table if sources were provided.
7. Synthesize findings by theme, agreement, disagreement, uncertainty, and implications.
8. Translate the synthesis into decision options with tradeoffs.
9. Identify missing evidence, major risks, and what additional research would change the recommendation.
10. Conclude with a recommended next step and confidence level.

Constraints
- Do not imply live browsing, retrieval, or source verification occurred unless source material is explicitly supplied.
- Treat unsupported claims as unverified.
- Distinguish sourced statements from assumptions and hypotheses.
- Keep outputs suitable for executive review and machine parsing.
- Use tables where useful, especially for citation review and decision comparison.
- If no sources are provided, return a research plan plus empty citation structure rather than invented evidence.

Output format
Return exactly these sections:
1. Job-to-be-done and decision supported
2. Inputs summary table
3. Facts, supplied evidence, and assumptions
4. Source discovery plan
5. Evidence review criteria
6. Citation table
7. Synthesis of findings
8. Decision options and tradeoffs
9. Risks, missing information, and research gaps
10. Recommendation and next actions

In the citation table, include columns for:
- Source ID
- Source type
- Title or description
- Date
- Relevance
- Credibility notes
- Key claim or finding
- Limitations
- Decision impact

Acceptance criteria
A good answer must:
- state the real decision being supported
- gather concrete research inputs before synthesis
- separate supplied evidence from assumptions
- produce a source discovery plan without pretending browsing occurred
- include a citation table and evidence review framework
- synthesize findings into decision-ready options with tradeoffs
- identify risks, missing evidence, and recommended next actions

Quality checks
Before finalizing, verify:
- no fabricated citations or claims appear
- every conclusion is either sourced, inferred, or marked uncertain
- the citation table includes limitations and decision impact
- missing information is explicit
- the final recommendation reflects the stated evidence quality and risk tolerance