A structured JSON prompt enhanced with chain-of-thought reasoning and markdown output formatting to retrieve and summarize recent literature on AI’s impact on productivity. The model is guided to think through its process step-by-step and then deliver the summaries in clear, human-readable markdown (headings, bullet points, etc.).

structured-ai-productivity-lit-scan-post2025.json

{
  "instructions": "You are performing a reproducible, controlled literature scan. Do not speculate. Follow these steps strictly and explain each one before proceeding:\n\n1. **Sources**: Query *only* from these two sources: `https://nber.org` and `https://arxiv.org`. Do not include any other source. Clearly state which source produced each result.\n\n2. **Time Filter**: Only include papers published after **January 1, 2025**. If none are available from a given source, explicitly say so and retrieve the **most recent** paper *after January 1, 2024*, clearly labeling it as an exception.\n\n3. **Topic Filter**: Use this exact query: `\"AI\" AND \"Labor Productivity\"`. Search titles, abstracts, and keywords only. Do not substitute synonyms or related concepts.\n\n4. **Ranking and Selection Criteria**:\n   - Prefer empirical or theoretical papers with clearly stated methods over speculative or opinion-based content.\n   - Select papers that provide either quantitative findings or methodological contributions directly related to labor productivity.\n   - If more than three papers qualify, rank by *recency*, then *relevance to firm-level or economy-wide productivity outcomes*. Return the top 3 only.\n\n5. **Summary Structure (Markdown Output)**: For each paper, output in **markdown** using the following consistent format:\n   - `## Title`\n   - `**Publication Date:** [YYYY-MM-DD]`\n   - `**Source:** [NBER or arXiv]`\n   - `**Summary:**`\n     - `Research question or hypothesis`\n     - `Data or method used`\n     - `Key finding or insight`\n   - `**Relevance Comment:** How this connects to organizational or policy planning`\n\n6. **Reproducibility**: Include a section at the end titled `### Query Log` that lists:\n   - Exact query terms used per source\n   - Date and time of query (UTC)\n   - Number of papers returned from each source\n   - URLs (if available) for retrieved papers\n\n7. **Additional Sections**:\n   - `### Deep Research Request Template`: Generate a reusable research prompt based *strictly* on the topics, methods, or gaps in the three summaries above.\n   - `### Email Summary`: Write a short, professional email summarizing the findings. Use a neutral tone. Be concrete and avoid speculation or interpretation beyond the paper’s own claims.\n\n8. **Failure Mode**:\n   - If no papers are found post-Jan 2025, clearly state: “No eligible papers found after January 1, 2025.” Then proceed to retrieve the 3 most recent papers post-Jan 2024. Label each as: `NOTE: PRE-2025 PAPER USED`.\n   - Do not return fewer than 3 papers unless explicitly none exist.\n\nYou must return all required sections. Do not add extra commentary outside the format. Do not generate fake or imaginary papers.",
  
  "task": "Controlled Literature Scan on AI and Labor Productivity (Post-2025 Only; Strict Summary Format)",

  "context": {
    "user_goal": "Preparing a policy briefing or strategic memo on the labor productivity effects of AI. Needs rigorously selected, replicable evidence from top sources."
  },

  "constraints": {
    "sources": ["https://nber.org", "https://arxiv.org"],
    "time_window": "Only papers published after January 1, 2025. Fallback to post-Jan 2024 if none found.",
    "topic_filter": [
      "AI AND Labor Productivity"
    ],
    "max_results": 3,
    "output_format": "markdown with headers per section"
  }
}