Tool Lessons4 min read

Use Claude Research to Build a Contradiction Map

Turn Claude Research into a decision aid by separating source agreement, disagreement and open questions before you write a recommendation.

Bokili EditorialPublished September 14, 2026· Verified September 14, 2026

Several sources flow into a map of agreements, contradictions and open questions before becoming a cited decision brief.

Research becomes risky when a polished summary hides where sources disagree. Claude Research can gather widely and cite its findings, but the useful output for a real decision is not a longer answer. It is a contradiction map: a compact view of what the evidence supports, where it conflicts and what remains unresolved.

This lesson is for people who compare public evidence before writing a recommendation: procurement analysts, project leads, consultants and researchers. The thesis is simple: ask Claude to separate agreement, disagreement and uncertainty before it drafts a conclusion. That structure makes verification visible and stops a confident synthesis from flattening an important conflict.

Use Claude Research when disagreement could change the decision

Anthropic’s current help describes Research as a paid-plan feature on the web, desktop and mobile. It conducts multiple searches that build on one another and returns citations. Web search must be enabled. Anthropic also distinguishes Research from ordinary web search: use Research for a question that needs broad, multi-source investigation rather than one or two factual look-ups.

Citations are a route back to evidence

A citation makes a claim easier to inspect. It does not make the claim correct. Open the source, check the passage and record what the source actually supports.

The three-lane contradiction map

Agreement

Claims supported in materially the same way by two or more independent sources. Record scope and date, not only the shared conclusion.

Disagreement

Claims that conflict because of different data, definitions, time periods or interests. State the reason for the conflict when you can verify it.

Open question

A decision-relevant point that the available sources do not answer. Give it an owner or a next source rather than filling the gap with inference.

A repeatable Claude Research workflow

From broad question to checked map

1
1. State the decision
Name who will decide, the date, the options and the evidence that could change the choice.
2
2. Set source boundaries
Ask for primary or authoritative sources first, a defined date range and explicit publication dates. Exclude sources that cannot support the claim.
3
3. Request three lanes
Require separate sections for agreement, disagreement and open questions. Ask for a citation beside every material claim.
4
4. Inspect the decisive claims
Open the sources behind the few claims that would change the recommendation. Check wording, scope, date and whether Claude inferred more than the source says.
5
5. Draft after verification
Only then ask for a decision brief that distinguishes verified facts, reasoned interpretation and unresolved risk.

Prompt Claude for a contradiction map

Use Research to help me prepare a decision, not a general summary.

Decision: [state the decision, owner and deadline].
Scope: [countries, products, time period and exclusions].
Prioritise primary and authoritative sources.

Build a table with: claim; agreement; disagreement; likely reason for the difference; source and publication date; what remains unknown; decision impact.

Put a citation beside every material claim. Do not resolve a conflict unless the evidence supports the resolution. Finish with the five claims I should verify manually before drafting a recommendation.

A useful result keeps conflicts visible. It might show that two current policy sources agree on a requirement, one vendor page uses a broader definition, and an implementation deadline is still unclear. The open point remains open until a source resolves it.

Replace the bracketed fields with a real, non-confidential decision. Follow your organisation’s rules before connecting or searching internal sources.

Reading is a start. Practice makes it stick.

Start learning

Worked example: compare supplier claims without inventing a winner

Imagine a fictional procurement lead comparing three software suppliers’ public claims about data retention. The decision is not “Which supplier sounds safest?” It is “Which offer meets the company’s twelve-month retention limit and deletion requirements?” The lead asks Claude Research to collect current product documentation, contract terms and relevant regulator guidance.

A normal summary could produce three tidy profiles. The contradiction map does something more useful. It records that two supplier pages describe customer-controlled deletion, while one contract defines a longer backup period. It flags that “deletion” and “inaccessible to users” may not mean the same thing. It also leaves one question open: whether the stated period covers support logs. No winner is declared until those decisive claims are checked in the original documents and confirmed through the procurement process.

	Fast summary	Contradiction map
Structure	One narrative per option	Agreement, disagreement and open questions
Uncertainty	Often absorbed into smooth prose	Kept visible and linked to the decision
Verification	Many claims appear equally important	The few decision-changing claims are prioritised
Next action	Draft the recommendation	Resolve or accept each material gap

Avoid three common mistakes

Before you use the map

Do not ask for an exhaustive web review when the decision only turns on three or four criteria.
Do not treat two pages repeating the same underlying announcement as independent agreement.
Do not let a newer source automatically outrank a primary source; compare authority, scope and date.
Do not paste confidential material or enable internal connectors without the permissions and data controls your organisation requires.
Do not ask Claude to choose before you have checked the claims that could reverse the choice.

Build a ten-minute contradiction map

Choose a low-risk public question with two plausible answers.
Write the decision, deadline, scope and three criteria.
Enable Research and run the contradiction-map prompt.
Open the citations for the two most decision-relevant claims.
Mark each as supported, overstated or still open.
Write one sentence explaining what evidence would change your current view.

Keep the conflict visible until the evidence earns closure

Claude Research is valuable because it can investigate a broad question across several sources in minutes. Its speed is most useful when your structure resists premature certainty. Build the contradiction map first, verify the decisive claims and draft the recommendation last.

For a related research setup, use the Perplexity Projects research brief. Then test what a summary may have missed with the omission check and expose hidden premises with the assumption log. Bokili’s tool-specific learning approach turns workflows like these into short, role-aware practice.