Deep Research Methods

researched 2026-05-25· 20 sources · 9 concepts · skill deep-research-methods

Methodology reference for rigorous AI-agent research. Complements deep-research (tool usage for firecrawl/exa) with research thinking — how to decompose questions, evaluate sources, synthesize finding

Deep Research Methods

Methodology reference for rigorous AI-agent research. Complements deep-research (tool usage for firecrawl/exa) with research thinking - how to decompose questions, evaluate sources, synthesize findings, and avoid bias. [source]

When to Use

Planning a research strategy or approach [source]
Decomposing a broad question into researchable sub-questions [source]
Evaluating source credibility or handling conflicting sources [source]
Synthesizing findings from multiple sources [source]
Avoiding confirmation bias or echo chambers [source]
Setting up multi-agent research fan-outs [source]
Deciding when to stop researching [source]

When NOT to Use

Conducting actual research on a topic → use deep-research [source]
Writing or editing a finished document → use a writing skill [source]
Quick factual lookups where methodology guidance isn't needed [source]

Source credibility hierarchy

Peer-reviewed papers (highest) [source]
Official documentation [source]
Practitioner post-mortems / production case studies [source]
Technical blog posts by domain experts [source]
General blog posts [source]
Forum discussions [source]
AI-generated content (lowest - treat as hypotheses to verify) [source]

Stopping criteria

Stop when: (1) new sources repeat known information, (2) every sub-question has 2–3+ independent sources, (3) contradictions are identified even if unresolved, (4) the next search is unlikely to change conclusions. [source]

Anti-pattern checklist

[ ] Sources from multiple ecosystems (not just one vendor's blog network)? [source]
[ ] Disconfirming evidence actively sought? [source]
[ ] Claims evaluated on evidence quality, not source prestige? [source]
[ ] Citation chains checked (do 5 "sources" trace to 1 original study)? [source]
[ ] Token budget allocated across sub-questions, not all on the first one? [source]

Three planning approaches

Planning-only: Generate research tasks directly from the user question. Fast but brittle with ambiguous queries. [source]
Intent-to-planning: Clarify user intent before generating queries. Ask: "What decision will this research inform?" to bound scope. [source]
Unified intent-planning: Generate preliminary plans while engaging the user. Surfaces assumptions early. [source]

Decomposition recipe

Identify the DECISION the research will inform (not just the topic) [source]
Extract key concepts that need operationalization [source]
Map sub-questions with dependency ordering: [source]
Independent sub-questions → fan-out in parallel [source]
Dependent sub-questions → sequence (answer A informs query B) [source]
For each sub-question, identify: [source]
What source type would authoritatively answer this? [source]
Is this a fast-moving or stable domain? [source]
What would a DISCONFIRMING answer look like? [source]
Assign token/time budget proportional to sub-question importance [source]

The search-read-infer loop

The agent does NOT search once and reason once. The canonical loop: [source]
Each iteration: (1) Act - produce search actions, (2) Observe - capture outcomes and assess gaps, (3) Optimize - update strategy, (4) Remember - persist key findings. [source]

Cross-referencing protocol

Minimum 5–8 authoritative sources before finalizing any conclusion. [source]
Citation chain detection: When 5 "different sources" all cite the same study, you have 1 evidentiary point, not 5. [source]
Ecosystem diversity: Seek sources from competing vendors, different regions, opposing viewpoints. [source]
Temporal clustering: If all evidence comes from one time period, the conclusion may reflect a trend, not a durable truth. [source]
Methodology check: A rigorous study by unknowns outweighs an unsupported claim by a famous researcher. [source]

Source quality checklist

For each source, assess: [source]
[ ] Does it cite its own sources? [source]
[ ] Is the author identifiable with relevant credentials? [source]
[ ] Is the publication venue reputable for this domain? [source]
[ ] Does it acknowledge limitations or counterarguments? [source]
[ ] Is it selling something? (vendor content requires extra skepticism) [source]
[ ] When was it published relative to domain rate of change? [source]
[ ] Has it been cited by other credible sources? [source]
[ ] Does it provide reproducible methodology or just conclusions? [source]

Handling conflicting sources

Check methodology - which source has better evidence backing? [source]
Check recency - in fast-moving domains, newer may reflect evolved understanding [source]
Check scope - are they actually answering the same question? [source]
Preserve the contradiction - report both positions rather than forcing resolution [source]
Flag confidence impact - contradictions lower overall claim confidence [source]

The four analytical passes

Consensus detection - which claims appear in 3+ independent sources? These form high-confidence findings. [source]
Contradiction mapping - where do sources directly disagree? Map: Source A claims X because [evidence]. Source B claims not-X because [different evidence]. [source]
Gap identification - what questions are implied but not addressed? Often the most valuable output. [source]
Cross-source narrative - organize thematically with confidence levels attached to each claim. [source]

Dual-perspective retrieval

Retrieve evidence using BOTH the original claim AND its negation. This captures supporting and contradicting evidence simultaneously. [source]

Confidence levels

High (3+ independent quality sources agree, no contradictions) - state as finding [source]
Medium (2 sources agree OR quality sources with minor caveats) - state with qualifier [source]
Low (single source OR contradicted) - flag as tentative/contested [source]
Speculative (no direct evidence, inferred from adjacent findings) - label explicitly [source]

Thematic organization

Wrong: "Smith found X. Jones found Y. Chen found Z." [source]
Right: "The evidence shows X [Smith 2024, Jones 2023]. However, this may not hold at scale [Chen 2022], and one study found the opposite under condition W [Park 2025]." [source]

Fan-Out (production standard)

One lead agent spawns 3–5 subagents in parallel. Each receives a structured brief: [source]
Critical rule: each subagent gets a bounded, purposeful brief - NOT a dump of the orchestrator's full history. [source]

Adversarial / Debate

Multiple agents reason independently, then argue toward convergence. Use architecturally diverse models - homogeneous agents become polarized rather than converging on truth. [source]

Council Mode

Three phases: (1) classify research question by complexity, (2) 3+ diverse models generate independent assessments, (3) structured consensus synthesis identifying agreement, disagreement, and unique findings. Reduces hallucination rates by ~36%. [source]

Practical ceiling

5–10 parallel agents before communication overhead exceeds value. Under fixed budgets, single-agent deep reading often beats multi-agent shallow reading. [source]

Confirmation bias

Detection: all found sources agree with the initial hypothesis; no disconfirming evidence appeared. [source]
Fix: force negation queries; allocate at least 20% of queries to "X failures" / "X criticism." [source]

Source echo chambers

Detection: all sources from same ecosystem, same time period, or same citation network. [source]
Fix: actively seek competing ecosystems, critical reviews, alternative approaches; check if multiple sources trace to the same original claim. [source]

Sycophantic convergence (multi-agent)

Detection: agents uncritically adopt peer views in debate. [source]
Fix: anonymize debate contributions; use architecturally heterogeneous models; enforce explicit evidence requirements - agents may only update beliefs when presented with NEW evidence. [source]

Premature convergence

Detection: research stopped after 2–3 agreeing sources on a topic deserving thorough investigation. [source]
Fix: define stopping criteria BEFORE beginning; require source diversity (different authors, publications, years); run the "What would change my mind?" test. [source]

Scope creep

Detection: research expanding far beyond original question; token budget exhausted on tangents. [source]
Fix: maintain original research map; only expand when tangent directly impacts a mapped sub-question; apply "Will this change the decision?" test. [source]

Required sections

Executive summary (2–4 sentences): Research question, headline finding, confidence level. [source]
Methodology disclosure: Sources consulted, search strategies, tools, depth, limitations. [source]
Findings per sub-question: Organized by theme, not source. Each: claim + evidence + citations + confidence + caveats. [source]
Evidence quality assessment: Rate sources; flag citation chains. [source]
Consensus vs outlier claims: Separate multi-source consensus from single-source claims. [source]
Knowledge gaps: Sub-questions inadequately answered, unresolved conflicts, thin evidence areas. [source]
Source list with quality ratings: URL, access date, source type, brief quality note. [source]

Token Economics

Multi-agent research uses approximately 15x more tokens than single-agent chat. Pre-allocate budget proportional to sub-question importance. Reserve 15% as contingency for unexpected findings. [source]
Diminishing returns signal: if the last 3 sources added zero new claims, you have likely reached saturation. [source]

Sources

Core references (May 2026): [source]
Deep Research Agents Survey - arXiv:2506.18096 [source]
Deep Research Survey - arXiv:2508.12752 [source]
Progressive Confidence Estimation - arXiv:2604.05952 [source]
Dual-Perspective Retrieval - arXiv:2602.18693 [source]
Multi-Agent Consistency - arXiv:2603.24481 [source]
Identity Bias in Multi-Agent Debate - arXiv:2510.07517 [source]
Council Mode: Mitigating Hallucination - arXiv:2604.02923 [source]
Anthropic Multi-Agent Research System [source]

Children

Question Decomposition (frontier)
Source Evaluation Heuristics (frontier)
Multi-Source Synthesis (frontier)
Subagent Research Patterns (frontier)
Anti-Pattern Detection (frontier)

Frontier under this node: Anti-Pattern Detection, Multi-Source Synthesis, Question Decomposition, Source Evaluation Heuristics, Subagent Research Patterns

← the whole tree · 3D view· how to read this page