Research with ChatGPT without confusing a confident answer with a verified fact.

Turn my topic into a research brief before doing the research.

Topic: [TOPIC]
Decision I may need to make: [DECISION]
Audience for the final report: [ME / EXECUTIVE TEAM / CLIENT / INVESTORS]
Geography: [WHERE]
Time period: [CURRENT / LAST 12 MONTHS / HISTORICAL RANGE]
Constraints: [BUDGET / COMPANY SIZE / REGULATION / TECH STACK / ETC.]

Return:
1. The primary research question
2. 5–8 sub-questions that must be answered
3. What evidence would change the decision
4. What terms need precise definitions
5. What information is likely to be time-sensitive
6. What would remain uncertain even after good research

Do not start the full research yet.

Use this source policy for the research:

- Prefer primary sources for factual claims whenever available.
- Use high-quality independent secondary sources to add context or challenge primary-source claims.
- Use vendor content for facts about the vendor's own product, but not as independent proof that the product is superior.
- Use community discussions only for experiences, complaints, edge cases and hypotheses — not as proof of market-wide facts.
- For every material claim, record the source, publication/update date and source type.
- If multiple articles repeat a claim but point back to the same original source, treat that as one evidence chain, not multiple independent confirmations.

Build an evidence table before writing the conclusion.

Columns:
- Claim
- Evidence summary
- Source
- Source type: primary / independent secondary / vendor / community
- Publication or update date
- Directly supports claim? yes / partly / no
- Confidence: high / medium / low
- What would falsify or weaken the claim

Do not hide disagreements between sources. If evidence is missing, write “not established” instead of filling the gap with inference.

Now try to disprove the emerging conclusion.

Search specifically for:
- credible sources that disagree
- newer evidence that supersedes older evidence
- different populations or markets where the result changes
- definitions that make apparently conflicting claims both technically true
- commercial incentives or methodological limitations that could bias a source

Return the strongest counter-evidence first.
Do not create artificial “balance” when one side has substantially stronger evidence; describe the evidence asymmetry clearly.

Audit the research as if you are a skeptical editor.

Select the 10 claims that matter most to the conclusion.
For each claim:
1. Show the exact conclusion being supported
2. Identify the strongest source
3. Explain why that source is appropriate
4. State whether the source directly supports the claim or only supports an inference
5. Flag any claim that is stronger than the evidence
6. Suggest safer wording where needed

Then list the 3 claims I should manually verify before acting.

Turn the verified research into a decision memo.

Structure:
1. Executive answer in 5 sentences maximum
2. What we know with high confidence
3. What we think is likely, but evidence is incomplete
4. What remains unknown
5. Strongest evidence against the preferred option
6. Decision options with trade-offs
7. Recommended next validation step — the cheapest action that would reduce the most important uncertainty
8. Source list

Do not invent a recommendation if the evidence is genuinely insufficient. Say “insufficient evidence” and specify what would resolve it.

Act as a rigorous research analyst, not a persuasive writer.

Research question:
[QUESTION]

Decision context:
[WHAT DECISION THIS RESEARCH WILL INFORM]

Scope:
- Geography: [X]
- Time period: [X]
- Population / customer / company type: [X]
- Constraints: [X]

Research protocol:
1. Clarify ambiguous terms before research.
2. Break the question into sub-questions.
3. Prefer primary and authoritative current sources.
4. Build an evidence table with claim, source, date, source type and confidence.
5. Search deliberately for contradictory evidence.
6. Separate facts, source claims, estimates and your own inferences.
7. Flag outdated evidence and population mismatches.
8. If sources disagree, explain why and which evidence is stronger — do not average the disagreement away.
9. Do not use the existence of a citation as proof that the citation supports the sentence.
10. Mark unsupported claims as “not established.”

Final output:
- executive answer
- evidence table
- strongest supporting evidence
- strongest counter-evidence
- uncertainties and missing data
- decision options and trade-offs
- 3 manual checks I should perform
- sources with dates

Important: confidence should reflect evidence quality, not how confident the prose sounds.