# Claude for Research > Research is a system of questions, not a search bar. Plans, sources, methods, and synthesis in language Claude can act on. For literature scans, interviews, competitive landscapes, and defensible conclusions. Source: https://www.juanmnl.com/notes/claude-for-research.html ## Frame the Question A fuzzy question gets a fuzzy answer. Sharpen these and Claude researches the right thing, at the right depth. ### The actual question State what you really want to know as a specific, answerable question, not a topic. Prompt: Not "tell me about EVs." But "How has EV battery cost per kWh changed 2015,2025, and why?" ### Scope & boundaries Geography, timeframe, population, sub-questions. Bounds keep the research focused and finite. Prompt: Scope: US market, last 5 years, consumer segment only. ### Depth A quick orientation vs an exhaustive, cited report. Say which, they're very different jobs. Prompt: I need a deep, multi-source report, not a quick summary. ### Recency How current it must be. Flag when only the latest data will do so Claude searches the live web. Prompt: Use only sources from the last 12 months. ### Purpose & audience A decision memo, a literature review, a blog post, the use shapes depth, tone, and format. Prompt: This feeds an investment decision, be skeptical and quantify. ### Vague → precise ## Working Modes The verb sets the job. Scoping, deep-diving, and fact-checking are different asks. ### Scope / map the landscape Get the lay of the land before diving, key players, debates, sub-topics, terms. Prompt: Map the landscape: main positions, key sources, and open questions. ### Deep dive Exhaustive investigation of one question across many sources, with citations. Prompt: Deep dive with citations, fan out across sources and verify claims. ### Fact-check / verify Confirm a specific claim against primary sources and flag what can't be confirmed. Prompt: Fact-check this claim, cite the primary source or say it's unverified. ### Compare / synthesize Pull multiple sources into one structured comparison or coherent view. Prompt: Compare these 4 approaches in a table: method, cost, evidence. ### Summarize Condense a long source or body of work. Specify length and what to keep. Prompt: Summarize this paper in 200 words, method, findings, limitations. ### Critique / red-team Have Claude attack the argument or find weaknesses in a body of evidence. Prompt: Red-team this conclusion, what's the strongest case against it? ### Monitor / track Keep tabs on a topic over time. Pairs well with a scheduled recurring brief. Prompt: Track developments and brief me weekly on what changed. ## Source Vocabulary Naming source quality gets you evidence, not vibes. Tell Claude what counts. ### Primary vs secondary Primary = original (study, dataset, filing); secondary = reporting on it. Prefer primary for claims. Prompt: Cite the primary source, not an article summarizing it. ### Peer-reviewed Vetted by experts before publication. The bar for scientific claims. Prompt: Use peer-reviewed studies where they exist. ### Preprint Posted before peer review, fast but unvetted. Useful, but flag the caveat. Prompt: If it's a preprint, label it as not yet peer-reviewed. ### Authoritative / primary org The body that owns the data (a statistics agency, regulator, the company itself). Prompt: Pull the number from the issuing agency, not a blog. ### Corroboration An independent second source confirming a claim. One source is a lead, not a fact. Prompt: Flag any claim with only one source, find corroboration. ### Recency & provenance When it was published and where it came from. Old or anonymous = lower trust. Prompt: Note each source's date and publisher beside the claim. ### Conflict of interest Who funded or benefits from a finding. Affects how much weight it carries. Prompt: Flag any funding or conflict of interest behind a study. ### Citation A traceable link or reference so you can verify it yourself. Prompt: Give inline citations with links for every factual claim. ## Search & Gather How Claude finds material. Steering the search shapes everything downstream. ### Search the live web For anything current, tell Claude to search rather than rely on training data. Prompt: Search the web, I need current figures, not your prior knowledge. ### Fan-out / breadth Cast wide across many sources before narrowing. Ask for multiple independent queries. Prompt: Fan out across several sources before drawing conclusions. ### Query terms & synonyms Have Claude vary the search language, jargon, plain terms, opposing framings. Prompt: Try multiple search phrasings, including critics' terms. ### Site / domain scoping Restrict to trusted domains (gov, edu, specific journals) when quality matters. Prompt: Prioritize .gov, .edu, and major journals. ### Date filtering Constrain to a time window so stale results don't crowd out current ones. Prompt: Limit results to 2023 onward. ### Follow the citation trail Trace a claim back to its origin; chase who cites a key paper. Prompt: Trace this stat to its origin and tell me where it actually came from. ## Critical Thinking The lenses that separate findings from noise. Name them and Claude reasons, not just collects. ### Triangulate Confirm a finding from multiple independent angles or methods. Prompt: Triangulate, does independent evidence agree? ### Correlation vs causation Two things moving together isn't one causing the other. A classic trap. Prompt: Is this causation or just correlation? What confounds it? ### Sample size & method How many, selected how. A vivid anecdote isn't data; a tiny study isn't proof. Prompt: Note the sample size and methodology for each study. ### Base rate The background frequency a claim should be judged against. Guards against scary-sounding stats. Prompt: What's the base rate here, for context? ### Bias & framing Selection, publication, and framing bias in how evidence is gathered and presented. Prompt: What biases might shape these sources? ### Steelman the other side Build the strongest version of the opposing view before judging. Beats a strawman. Prompt: Steelman the counter-argument, then weigh both. ### Uncertainty & confidence Ask Claude to state how sure it is and where the evidence is thin. Prompt: Mark each conclusion high / medium / low confidence. ### Distinguish fact / inference / opinion Keep what's established separate from what's interpreted or argued. Prompt: Separate established facts from your inferences. ## Synthesis & Output Naming the deliverable gets you a usable artifact instead of a wall of text. ### Executive summary The answer first, in a few sentences, with detail below. Lead with the conclusion. Prompt: Start with a 3-sentence exec summary, then the evidence. ### Literature review A structured survey of what's known, where sources agree, and where they conflict. Prompt: Write a literature review grouped by theme, noting disagreements. ### Annotated bibliography Each source listed with a short note on what it says and how reliable it is. Prompt: Give an annotated bibliography, one note per source. ### Comparison matrix Options as rows, criteria as columns. Makes tradeoffs scannable. Prompt: Put the options in a comparison matrix with sources cited. ### Key findings & gaps Bullet the takeaways and, crucially, what remains unknown. Prompt: List key findings and open gaps in the evidence. ### Cited report A full write-up where every claim links to its source. The gold standard. Prompt: Deliver a fully cited report, every claim links to a source.