This build focuses on a repeatable workflow with a review point—not “autonomous AI” for its own sake. The goal is useful leverage without surrendering accountability.
Step 1 — Frame the decision
Start with the decision or question you are actually trying to resolve. A broad topic produces broad output; a decision produces useful research.
Step 2 — Define evidence standards
Decide what counts as strong evidence before searching: primary sources, official documentation, peer-reviewed work, current reporting, or domain experts.
Step 3 — Build a source ledger
For each important claim, retain the source, date, and what it actually supports. Never let a polished summary erase provenance.
Step 4 — Search for disagreement
Ask explicitly for credible evidence against the leading conclusion. Good research tries to break its own answer.
Step 5 — Separate fact from inference
Make the assistant label established facts, estimates, opinions and its own synthesis.
Step 6 — Produce a decision brief
End with what is known, what remains uncertain, the implications, and the next action—not a wall of research.
Step 7 — Human verification
Open the most consequential sources yourself. Check numbers, dates, quotations, scope and whether the source really says what the AI claims.
Copy the complete blueprint
Customize the bracketed details for your situation, then test it on a real but low-risk task before depending on it.
You are my research assistant. Your job is to investigate one clearly defined question and produce a decision-ready brief. Before researching: 1. Restate the exact question. 2. Ask for missing context that would materially change the answer. 3. Separate what needs current evidence from what can be explained from established knowledge. During the work: - Prefer primary and authoritative sources. - Record the source, publication date, and the claim it supports. - Distinguish facts, estimates, opinions, and your own inference. - Look for credible evidence that disagrees with the leading conclusion. - Never invent a citation, quote, statistic, study, or URL. Final output: A. Executive answer B. Key evidence C. Conflicting or uncertain evidence D. What this means for my situation E. Risks / unknowns F. Sources to verify G. Recommended next step If the evidence is insufficient, say so rather than filling the gap.
The Human Edge Check
Query expansion · organizing evidence · summarization · comparison · drafting the brief
Choosing evidence standards · opening critical sources · interpreting context · accepting risk · final decision