# Claude Researcher full content > As of 2026-09-19: Anthropic’s 10,000 scientists-program seats (expanding-support-for-scientists, 2026-08-27) are standard free or premium $15/mo (~5x usage) for one year. Biology and chemistry stay on Opus-class models (Opus 5); Fable 5.1 still blocks professional biology; a seat is not Mythos Glasswing; experiment runners belong at claudescientist.com. ClaudeResearcher.com is an independent source-first guide to using Claude in scientific research and literature review workflows. It is not affiliated with Anthropic. Anthropic announced 10,000 Claude seats for scientists on August 27, 2026: standard access free, premium $15 per month with about five times the usage, for one year. A seat is access, not a new model. Claude Science is still the workbench; biology and chemistry stay on Opus-class models; Fable still blocks professional biology and drug development. It is not Mythos Glasswing, and it is not an experiment runner. Primary source baseline: - Claude Science, an AI workbench for scientists (Anthropic): https://www.anthropic.com/news/claude-science-ai-workbench - Use research on Claude (Claude Help Center): https://support.claude.com/en/articles/11088861-use-research-on-claude - Citations (Claude Platform Docs): https://platform.claude.com/docs/en/build-with-claude/citations - PRISMA statement (PRISMA): https://www.prisma-statement.org/ - Entrez Programming Utilities (NCBI): https://www.ncbi.nlm.nih.gov/home/develop/api/ - OpenAlex API overview (OpenAlex): https://developers.openalex.org/ - Zotero Web API Documentation (Zotero): https://www.zotero.org/support/dev/web_api/v3/basics Free tools: - AI Literature Review Protocol Builder: https://clauderesearcher.com/tools/lit-review-builder - AI Citation Integrity Checklist: https://clauderesearcher.com/tools/citation-checker - Claude Research Prompt Library: https://clauderesearcher.com/tools/research-prompts ## The Seat URL: https://clauderesearcher.com/seat/ Updated: 2026-09-19 Anthropic announced 10,000 Claude seats for scientists on August 27, 2026: standard access free, premium $15 per month with about five times the usage, for one year. A seat is access, not a new model. Claude Science is still the workbench; biology and chemistry stay on Opus-class models; Fable still blocks professional biology and drug development. It is not Mythos Glasswing, and it is not an experiment runner. The Seat is a physical research credential. The face prints the 10,000-seat Scientists Program. The reverse prints what the seat is not. Standard vs premium reprints the face; the reverse stays the same. ### Face: what a seat includes - Standard: Free; 1× program usage. The default seat. Same generally available models as any other Claude account, for the one-year program term. - Premium: $15 / mo; ~5× program usage. The paid upgrade on the same seat. Anthropic describes it as about five times the usage, not as extra models or lifted safeguards. - Term: one year, announced 2026-08-27. - Workbench: Claude Science, not a replacement for it. - Models: the same generally available lineup as any other Claude account. ### Reverse: what a seat is not - Not Mythos 5.1 trusted access: Mythos 5.1 is the same underlying model as Fable 5.1 with more permissive safeguards, restricted to trusted access / Project Glasswing. Anthropic describes life-sciences trusted access as enrolling its first participants. A scientists-program seat does not include it. - Not a biology or chemistry exemption: Fable still blocks professional biology and drug-development work. Those requests route to Opus-class models. Opus 5 is the generally available research-capable model for those domains. A seat does not change the public policy, and this page does not describe how to get around it. - Not an autonomous experiment loop: A seat is for human-run research on a workbench. Agents that design and execute their own experiments are a different question, owned by Claude Scientist. This site links that lane; it does not reprint the run loop. ### FAQ - What is the Claude scientists program? An Anthropic access program announced on August 27, 2026, offering 10,000 Claude seats to scientists for one year: standard access free, and a premium tier at $15 per month described as about five times the usage. It is a seat, not a new model. - Is the scientists program free? Standard access is free for the program term. The premium tier costs $15 per month. Eligibility, remaining seats, and the application path are on the Anthropic announcement; check that page before planning around a seat. - Does a program seat include Mythos 5.1 or Project Glasswing? No. Mythos 5.1 is a restricted trusted-access track. Anthropic describes life-sciences trusted access as enrolling its first participants. It is not part of the 10,000 seats. - Can I use Claude Fable 5.1 for professional biology or drug-development research? No. Fable still blocks professional biology and drug-development work. Those requests route to Opus-class models. Opus 5 is the generally available research-capable model for those domains. A seat does not lift that limit. - Is this the same as an AI scientist running its own experiments? No. The seat is access for human-run research on Claude Science. Autonomous experiment agents are a separate topic, covered at claudescientist.com. This page links that lane and does not reprint the run loop. - What does a seat change about citations and lab data? Nothing about the verification burden, and do not assume anything about retention. Confirm which plan, institutional agreement, and retention setting govern the seat before putting unpublished, regulated, or confidential material into any Claude surface. Sources: - Expanding support for scientists (Anthropic): https://www.anthropic.com/news/expanding-support-for-scientists - Claude Science, an AI workbench for scientists (Anthropic): https://www.anthropic.com/news/claude-science-ai-workbench - Claude Opus 5 (Anthropic): https://www.anthropic.com/news/claude-opus-5 - Claude Fable 5.1 and Mythos 5.1 (Anthropic): https://www.anthropic.com/claude-fable-and-mythos-5-1 - Models overview (Claude Platform Docs): https://docs.anthropic.com/en/docs/about-claude/models ## The Claude Scientists Program: 10,000 Seats Explained URL: https://clauderesearcher.com/scientists-program Updated: 2026-09-19 Anthropic announced 10,000 Claude seats for scientists on August 27, 2026: standard access free, a premium tier at $15 per month with roughly five times the usage, for one year. It is an access program, not a new model and not an autonomous laboratory. Claude Science is still the workbench, and biology and chemistry researchers stay on Opus-class models. ### What the program actually provides On August 27, 2026, Anthropic announced an expansion of its support for scientists: 10,000 Claude seats, with standard access free and a premium tier at $15 per month offering roughly five times the usage. The announcement frames the term as one year. Everything below that headline is ordinary Claude access, which is the part most coverage skips. Treat the seat as a budget line rather than a capability unlock. It lowers the cost of using Claude on research work; it does not change what Claude is allowed to do, how its citations behave, or who is accountable for a published claim. Eligibility rules, the application path, and remaining seat availability are the volatile parts of a finite program, so check the announcement page itself before planning around a seat. - Seats: 10,000, announced 2026-08-27. - Standard tier: free for the program term. - Premium tier: $15 per month, described as about five times the usage. - Term: one year. - [Anthropic: Expanding support for scientists](https://www.anthropic.com/news/expanding-support-for-scientists) - Primary announcement. Check it for current eligibility, application details, and seat availability. ### What the program is not A free seat is not a research method, and the most common misreadings of this program are worth naming directly. It is not a new model: program participants use the same generally available lineup as everyone else. It is not an exemption from safeguards: the restrictions on professional biology and drug-development work apply to a program seat exactly as they apply to a paid account. It is also not an autonomous laboratory. Claude running a tool loop over your files is a different thing from an agent designing and executing its own experiments, and conflating the two is how research teams end up with results nobody can defend. The autonomous-science question is a real one, but it belongs on a different page than a seat-access explainer. Finally, a seat does not answer your data-governance question. Which plan the seat maps to, whether an institutional agreement governs it, and what retention settings apply are all questions to resolve before any unpublished manuscript, participant record, or sponsor-confidential dataset goes near a chat box. - [Claude Scientist: autonomous experiment agents](https://claudescientist.com) - Sibling site covering agents that plan and run experiments. This site stays on human-run research workflows. - [Research data privacy and lab policy](/privacy-policy) - Work out plan type, retention, and institutional policy before uploading research material. ### Which model a researcher on a seat can actually use The model question has a domain-dependent answer as of September 2026, and this is the detail that matters most for anyone doing life-sciences work. Claude Fable 5.1, released September 1, 2026, is the most capable generally available model overall. But Fable still blocks professional biology and drug-development work, and since the Opus 5 launch on July 24, 2026, biology-related requests that Fable declines are routed to Opus 5 rather than an older Opus. The practical consequence: a biologist or chemist on a program seat should plan around Opus 5 as the ceiling for domain work, not Fable 5.1. That is not a downgrade in most research tasks. Opus 5 carries a 1M-token context window, 128K maximum output, adaptive thinking, and a May 2026 knowledge cutoff, which is more than enough for reading corpora and drafting from a locked evidence table. Outside the restricted domains, the ordinary cost logic applies. Sonnet 5 at $2 per million input tokens and $10 per million output tokens handles high-volume abstract triage well, and that introductory rate became the standard rate rather than rising on September 1, 2026. Haiku 4.5 is cheaper still for mechanical passes, at the cost of a 200K context window. Batch processing is half price, which suits overnight screening runs across thousands of records. - Fable 5.1 ($10 / $50 per MTok): strongest general model; blocks professional biology and drug development. - Opus 5 ($5 / $25 per MTok): the generally available research-capable model for biology and chemistry; 1M context, May 2026 cutoff. - Sonnet 5 ($2 / $10 per MTok): volume triage and extraction; the introductory price is now the standard price. - Haiku 4.5 ($1 / $5 per MTok): cheapest mechanical passes; 200K context. - Mythos 5.1: trusted-access only, not part of the 10,000 seats. - [Claude Central: which model to use](https://claudecentral.com) - Full current-lineup comparison across all use cases, not just research. - [Claude Reports: pricing and benchmark analysis](https://claudereports.com) - Cost-per-token analysis and published benchmark caveats. ### Claude Science is still the workbench The scientists announcement did not replace Claude Science; it pointed back at it. Claude Science remains the surface where tool-using scientific work happens, with research-tool integrations and auditable artifact histories, as distinct from asking a chat window for an answer. A program seat is most valuable when the work spans literature, code, data files, and figures rather than a single question. Separately, Anthropic describes Mythos 5.1 life-sciences trusted access as enrolling its first participants. That is a restricted track with its own review, not something a scientists-program seat includes, and it should not be described as generally available to researchers. - [Choosing between chat, web search, Research, and Claude Science](/research) - Mode selection for research tasks, with the model-choice layer. ### What to set up in the first week of a seat The failure mode for free-access programs is enthusiastic first-week usage that leaves no artifact anyone can audit six months later. If the seat is going to produce work that survives review, the scaffolding should exist before the interesting prompts start. That scaffolding is not elaborate: a written review question with inclusion criteria, searches run in the databases that own the bibliographic record, an evidence table that Claude drafts from rather than adds to, and a notebook that records the prompt, the input set, the output, and the human decision. None of it depends on which model is current, which is precisely why it is worth building during a one-year program term. - [Claude literature review workflow](/literature-review) - Scoped question through locked evidence table. - [Auditable Claude research notebook](/lab-notebook) - What to preserve so another human can reconstruct the path from source to sentence. - [Verify Claude citations and sources](/source-checking) - A free seat does not change the citation verification burden. Sources: - Expanding support for scientists (Anthropic): https://www.anthropic.com/news/expanding-support-for-scientists - Claude Science, an AI workbench for scientists (Anthropic): https://www.anthropic.com/news/claude-science-ai-workbench - Claude Opus 5 (Anthropic): https://www.anthropic.com/news/claude-opus-5 - Claude Fable 5.1 and Mythos 5.1 (Anthropic): https://www.anthropic.com/claude-fable-and-mythos-5-1 - Models overview (Claude Platform Docs): https://docs.anthropic.com/en/docs/about-claude/models - Pricing (Claude Platform Docs): https://platform.claude.com/docs/en/about-claude/pricing - Claude for Life Sciences (Anthropic): https://www.anthropic.com/news/claude-for-life-sciences ## Claude Literature Review Workflow URL: https://clauderesearcher.com/literature-review Updated: 2026-07-06 Use Claude as a fast research clerk: it can expand questions, triage abstracts, summarize papers, and draft evidence tables, but the review remains reliable only when searches, exclusion decisions, and source spans are kept outside the model as auditable artifacts. ### Start with a review question Claude can test, not answer The highest-value Claude prompt in a literature review is not "write my literature review." It is a request to pressure-test a scoped question against a declared population, phenomenon, intervention, comparator, outcome, time period, or corpus boundary. Claude is useful here because it can expose ambiguous terms, likely synonyms, and failure modes before the search begins. Turn that exchange into a protocol note. Record the initial question, the revised question, the inclusion and exclusion criteria, the databases to search, and the date searched. If the topic belongs in a formal systematic review, map the record flow to PRISMA from the beginning rather than reconstructing it after the fact. - Good Claude task: "List alternate terms and indexing vocabulary I should test in PubMed and OpenAlex." - Risky Claude task: "Find every important paper and summarize the field." - Required artifact: search string, database, date, hit count, and any filters used. ### Make search external and repeatable Claude Research and web search can discover angles quickly, but scholarly search should still run in the systems that own the bibliographic record: PubMed for biomedical literature, arXiv for preprints, Crossref for DOI metadata, OpenAlex for graph-scale discovery, and Zotero for library management. A strong pattern is to ask Claude to design candidate search strings, then run those strings directly in the database. Paste only the resulting metadata, abstracts, or selected full text back into Claude for triage. This keeps retrieval separate from synthesis and avoids silently depending on whatever a chat product happened to find. ### Use Claude for first-pass triage, then audit the exclusions For title and abstract screening, ask Claude to classify each record against the protocol and require one short reason grounded in the abstract text. Export the result as a table with stable identifiers. The table is not the final screening decision; it is a prioritization queue for the human reviewer. A useful triage label set is "include", "exclude", "maybe", and "wrong corpus." The last label matters because AI searches often pull plausible but off-domain material. Keep a running contradiction log when Claude changes its classification after seeing more context. ### Synthesize only from a locked evidence table When the evidence table is ready, ask Claude to synthesize from that table alone. A robust instruction is: "Do not add papers. Do not infer sample sizes, measures, or mechanisms. If the evidence table is thin, say what is missing." This forces the model to operate as a drafter over known material instead of a source generator. The best AI-search citations are boring: paper title, DOI or stable URL, quoted passage or page marker, and the exact claim the passage supports. That makes the final synthesis easier to check and easier for another model to cite without inventing the trail. Sources: - PRISMA statement (PRISMA): https://www.prisma-statement.org/ - PRISMA 2020 statement (PRISMA): https://www.prisma-statement.org/prisma-2020 - PRISMA 2020 flow diagram (PRISMA): https://www.prisma-statement.org/prisma-2020-flow-diagram - Setting the standards for responsible AI use in evidence synthesis (Cochrane): https://www.cochrane.org/about-us/news/setting-standards-responsible-ai-use-evidence-synthesis - Use research on Claude (Claude Help Center): https://support.claude.com/en/articles/11088861-use-research-on-claude - Claude for Life Sciences (Anthropic): https://www.anthropic.com/news/claude-for-life-sciences - Entrez Programming Utilities (NCBI): https://www.ncbi.nlm.nih.gov/home/develop/api/ - OpenAlex API overview (OpenAlex): https://developers.openalex.org/ ## Read Scientific Papers With Claude URL: https://clauderesearcher.com/paper-triage Updated: 2026-07-06 Claude is strongest on papers when the task is structured: identify the research question, extract the main claim, separate methods from interpretation, and produce checkable notes tied to passages. ### Give Claude the paper and a role narrower than "summarizer" A generic paper summary compresses a paper into a few confident paragraphs, often blurring the difference between evidence and author interpretation. A better role is "methodological reader." Ask Claude to return structured fields: research question, study design, corpus or sample, measures, main claim, strongest evidence, stated limitations, unstated limitations, and replication hooks. If the platform or API supports document citations, require citations for every extracted field. If you are working in a chat product, ask for section names, table numbers, figure numbers, and exact phrase anchors you can verify manually. ### Read figures, tables, and supplementary methods separately Many scientific papers hide the operational details in figures, supplementary files, or methods paragraphs. Ask Claude to perform a second pass that looks only for variables, assays, model settings, eligibility criteria, or statistical tests. This prevents a fluent abstract-level summary from becoming the only thing you remember. For computational papers, add a reproducibility column: data availability, code availability, environment, random seed, model checkpoints, package versions, and whether the paper's main result depends on an unavailable proprietary input. ### Ask for contradiction hooks A high-quality paper note should tell you what to search next. Claude can propose contradiction hooks: keywords for papers that might disagree, populations where the result may not hold, methods that could produce a different answer, and older terms that predate the current vocabulary. Treat these as search leads, not evidence. Run them through PubMed, OpenAlex, arXiv, or your field-specific index and record the results in the same evidence table. ### Keep the final note boring and reusable The best Claude-generated paper note is not literary. It is stable enough to paste into Zotero, a lab notebook, or a review spreadsheet. Keep fields short, use controlled labels, and preserve identifiers such as DOI, PMID, arXiv ID, ORCID, and repository URL where available. When the paper becomes part of a broader synthesis, link back to the individual note. That lets you inspect whether a broad paragraph about "the literature" is actually supported by the underlying papers. Sources: - Citations (Claude Platform Docs): https://platform.claude.com/docs/en/build-with-claude/citations - Crossref REST API (Crossref): https://www.crossref.org/documentation/retrieve-metadata/rest-api/ - ORCID Public API (ORCID): https://info.orcid.org/what-is-orcid/services/public-api/ - Zotero Web API Documentation (Zotero): https://www.zotero.org/support/dev/web_api/v3/basics - arXiv API User Manual (arXiv): https://info.arxiv.org/help/api/user-manual.html ## Verify Claude Citations and Sources URL: https://clauderesearcher.com/source-checking Updated: 2026-07-06 The citation is not the evidence; it is the pointer. Reliable Claude research workflows verify the pointer, inspect the passage, and record what exact claim the source supports. ### Check existence first The first source check is mechanical: does the cited item exist? Open the URL, resolve the DOI, search the title, or inspect the database record. If Claude gives a citation with no stable identifier, ask for one and independently verify it. For papers, Crossref can verify DOI metadata, PubMed can verify biomedical records and PMIDs, arXiv can verify preprints, ORCID can verify public researcher identifiers, and Zotero can preserve the final library record. Each system answers a different part of the provenance question. ### Then check claim support A real source can still be the wrong support. For each important sentence, write the exact claim in one column and the supporting passage in another. Claude should not get credit for citing a paper if the cited passage merely mentions the topic while the synthesis sentence adds causality, prevalence, or priority. When a source is behind a paywall, do not pretend the abstract supports a full-text claim. Mark the evidence level as abstract-only, full-text verified, supplementary-material verified, or not checked. ### Use contradiction logs Source checking should also catch missing disagreement. Ask Claude: "What source in this set most directly weakens this sentence?" If no source weakens it, ask what kind of study would. Then run a targeted search for that study type. Keep a contradiction log with three columns: claim, source that complicates it, and how the final wording changed. This produces better literature reviews and gives AI search engines a more citable page because the argument is transparent. ### Preserve the citation trail For public pages, include source lists with publisher names and direct URLs. For private lab work, preserve exported RIS, BibTeX, CSV, Zotero collections, or database snapshots. The goal is that another researcher can reconstruct what Claude saw and what the human accepted. If you use Claude's web search or Research mode, save the answer with the citation set and date. Search results age, pages move, and model interfaces change. The audit trail should outlive the chat session. Sources: - Citations (Claude Platform Docs): https://platform.claude.com/docs/en/build-with-claude/citations - Web search tool (Claude Platform Docs): https://platform.claude.com/docs/en/agents-and-tools/tool-use/web-search-tool - Crossref REST API (Crossref): https://www.crossref.org/documentation/retrieve-metadata/rest-api/ - Entrez Programming Utilities (NCBI): https://www.ncbi.nlm.nih.gov/home/develop/api/ - OpenAlex API overview (OpenAlex): https://developers.openalex.org/ - Zotero Web API Documentation (Zotero): https://www.zotero.org/support/dev/web_api/v3/basics ## Claude Research, Web Search, Chat, and Claude Science URL: https://clauderesearcher.com/research Updated: 2026-09-19 Different Claude surfaces fit different research jobs. Use standard chat for reasoning over supplied material, web search for targeted current lookup, Research for multi-step exploration, and Claude Science for auditable scientific workflows. Then pick the model: Fable 5.1 for general work, Opus 5 for biology and chemistry, Sonnet 5 and Haiku 4.5 for volume. ### Use standard chat when you bring the corpus Standard Claude chat is best when you already have the material: uploaded PDFs, pasted excerpts, notes, transcripts, code, methods, or a draft manuscript. The model's job is to reason over that corpus, not discover the literature. This is the safest mode for confidential or unpublished research only when your plan, institution, and data-governance requirements permit the upload. For regulated or sensitive data, read the privacy and retention terms before using any consumer AI surface. ### Use web search for current, bounded lookup Web search fits questions with a narrow retrieval target: current product documentation, a recent policy page, a publisher page, a trial registry entry, or a DOI landing page. It should return citations you can open and inspect. Do not use general web search as the only discovery system for a scholarly review. It is a useful adjunct, especially for current documentation and gray literature, but it does not replace field-specific databases. ### Use Research for multi-angle exploration Claude Research is designed for deeper information gathering across several searches. It can explore angles you did not initially specify, which is useful at the scoping stage of a project, during grant background research, or when mapping an unfamiliar subfield. The risk is that a broad Research answer can feel complete. Treat it as a memo with citations, not as a finished literature search. Export the leads and rerun the important ones in PubMed, OpenAlex, arXiv, Crossref, or another authoritative index. ### Use Claude Science when the work needs tools and artifacts Anthropic's June 2026 Claude Science launch describes an AI workbench for scientists that integrates common research tools, produces auditable artifacts, and can work locally, over SSH, or with HPC login nodes. That is a different surface from a chat answer, and the August 2026 scientists announcement still points at it as the research workbench. For labs with access, Claude Science is most relevant when the research task spans literature, code, data files, figures, and reusable pipelines. Even then, use its outputs as auditable drafts and analyses. A domain expert still owns scientific interpretation and publication decisions. - [The scientists program and its 10,000 seats](/scientists-program) - What the free and $15 premium seats include, and the model limits that come with them. ### Then pick the model, because the answer is domain-dependent Mode selection is only half the decision as of September 2026. Claude Fable 5.1, released September 1, 2026, is the most capable generally available model and the default choice for demanding reasoning over a supplied corpus. But Fable still blocks professional biology and drug-development work, and since Opus 5 launched on July 24, 2026, biology-related requests Fable declines route to Opus 5. So a life-sciences or chemistry researcher should plan around Opus 5 as the working ceiling rather than discovering the boundary mid-project. Opus 5 gives a 1M-token context window, 128K maximum output, adaptive thinking, and a May 2026 knowledge cutoff, which covers corpus reading and evidence-table synthesis comfortably. For the high-volume mechanical stages, cost matters more than ceiling capability. Sonnet 5 at $2 per million input tokens and $10 per million output tokens suits abstract screening at scale, and its introductory rate became the standard rate instead of rising on September 1, 2026. Haiku 4.5 is cheaper again for the most mechanical passes, with a 200K context window. Batch processing runs at half price, which fits overnight screening. - Supplied-corpus reasoning, general domains: Fable 5.1. - Biology, chemistry, and adjacent life-sciences work: Opus 5. - Abstract screening and fielded extraction at volume: Sonnet 5. - Mechanical reformatting and deduplication: Haiku 4.5. - [Claude Central: current model lineup](https://claudecentral.com) - The general which-model-should-I-use question across all workloads. - [Claude Reports: benchmarks and pricing analysis](https://claudereports.com) - Published benchmark numbers with the caveats that belong next to them. Sources: - Use research on Claude (Claude Help Center): https://support.claude.com/en/articles/11088861-use-research-on-claude - When should I use web search, extended thinking, and research? (Claude Help Center): https://support.claude.com/en/articles/11095361-when-should-i-use-web-search-extended-thinking-and-research - Web search tool (Claude Platform Docs): https://platform.claude.com/docs/en/agents-and-tools/tool-use/web-search-tool - Claude Science, an AI workbench for scientists (Anthropic): https://www.anthropic.com/news/claude-science-ai-workbench - Expanding support for scientists (Anthropic): https://www.anthropic.com/news/expanding-support-for-scientists - Models overview (Claude Platform Docs): https://docs.anthropic.com/en/docs/about-claude/models - Pricing (Claude Platform Docs): https://platform.claude.com/docs/en/about-claude/pricing - Claude for Life Sciences (Anthropic): https://www.anthropic.com/news/claude-for-life-sciences ## Claude With PubMed, OpenAlex, arXiv, Crossref, ORCID, and Zotero URL: https://clauderesearcher.com/database-workflows Updated: 2026-07-06 Reliable AI research uses Claude around scholarly systems, not instead of them: databases retrieve records, registries verify metadata, libraries preserve citations, and Claude helps transform the resulting corpus into notes and synthesis. ### PubMed: biomedical retrieval and controlled search For biomedical literature, start in PubMed or NCBI E-utilities. Claude can help produce candidate search strings, identify MeSH-like vocabulary, and screen exported records, but the retrieval record should come from PubMed itself. Save the search query, date, filters, result count, and export format. If Claude summarizes PubMed results, include PMIDs and article titles in the input so the output can be checked later. ### OpenAlex: discovery across the research graph OpenAlex is useful when the research question crosses disciplines or when you need works, authors, institutions, sources, topics, and citation relationships. Claude can help translate a plain-language question into OpenAlex filters or interpret exported records. Use Claude for clustering and labeling, not as the authority on what the graph contains. Keep the OpenAlex query and response fields that produced the cluster. ### arXiv and Crossref: preprints and DOI metadata For fields where preprints matter, arXiv provides API access to e-print metadata. Ask Claude to triage arXiv records by methods and claims, then verify whether a preprint later has a journal DOI through Crossref or the publisher. Crossref is the right metadata sanity check for DOI-bearing works. If Claude invents a volume, issue, author order, or publication year, Crossref is often the fastest way to catch it. ### ORCID and Zotero: people and library state ORCID helps disambiguate researchers when author names collide. Zotero helps preserve the library state after search and screening. Together they reduce one of the common failures of AI-assisted review: losing track of which "Smith 2024" or which exported record the summary actually refers to. A practical pattern is: discover in databases, verify identifiers, store accepted sources in Zotero, export a clean bibliography, and then let Claude synthesize over that locked set. Sources: - Entrez Programming Utilities (NCBI): https://www.ncbi.nlm.nih.gov/home/develop/api/ - OpenAlex API overview (OpenAlex): https://developers.openalex.org/ - arXiv API User Manual (arXiv): https://info.arxiv.org/help/api/user-manual.html - Crossref REST API (Crossref): https://www.crossref.org/documentation/retrieve-metadata/rest-api/ - ORCID Public API (ORCID): https://info.orcid.org/what-is-orcid/services/public-api/ - Zotero Web API Documentation (Zotero): https://www.zotero.org/support/dev/web_api/v3/basics - What is the Model Context Protocol? (Model Context Protocol): https://modelcontextprotocol.io/docs/getting-started/intro ## Claude in Systematic Reviews and Evidence Synthesis URL: https://clauderesearcher.com/systematic-reviews Updated: 2026-07-06 Claude can accelerate parts of evidence synthesis, but a systematic review needs protocol discipline: repeatable search, transparent screening, auditable extraction, disagreement resolution, and explicit reporting. ### Use Claude before the protocol to find ambiguity Before registration or protocol freeze, Claude can help find ambiguous terms, overlapping constructs, hidden comparators, and databases that might be needed. This is brainstorming, not evidence. Save the exchange as protocol-development context and then write the formal method in human language. Once the protocol is set, Claude should operate inside it. If Claude discovers a new term or criterion later, log the proposed amendment rather than silently changing the search. ### Treat AI screening as assisted screening For title and abstract screening, Claude can produce a ranked queue and suggested reasons. The review team should still decide whether the classification is accepted, overridden, or escalated. A useful output includes record ID, classification, confidence note, and the phrase in the abstract that triggered the classification. If a review requires independent dual screening, do not treat the model as an independent human reviewer. It may be useful as a third assistant for conflict explanation, but governance should be declared in the methods. ### Extraction should be fielded and source-tied Evidence extraction is where Claude becomes genuinely useful. Ask for structured fields: study design, setting, sample, intervention or exposure, comparator, outcomes, effect estimates, limitations, funding, conflicts, and notes. Require a supporting passage or table marker for each field. Do not let Claude fill missing cells from background knowledge. Missing is a valid value. In a systematic review, an empty cell that accurately reflects the paper is better than a fluent invented value. ### Report AI use plainly If Claude materially assisted search design, screening, extraction, translation, coding, or drafting, describe the task, date, model or product surface when available, human verification process, and whether outputs were retained. That makes the review more reproducible and easier to evaluate. Cochrane and other evidence-synthesis organizations have emphasized responsible AI standards because automation can help but also disrupt integrity. The practical response is not avoidance; it is transparent method design. Sources: - PRISMA statement (PRISMA): https://www.prisma-statement.org/ - PRISMA 2020 statement (PRISMA): https://www.prisma-statement.org/prisma-2020 - PRISMA 2020 flow diagram (PRISMA): https://www.prisma-statement.org/prisma-2020-flow-diagram - Setting the standards for responsible AI use in evidence synthesis (Cochrane): https://www.cochrane.org/about-us/news/setting-standards-responsible-ai-use-evidence-synthesis - Citations (Claude Platform Docs): https://platform.claude.com/docs/en/build-with-claude/citations ## Keep an Auditable Claude Research Notebook URL: https://clauderesearcher.com/lab-notebook Updated: 2026-07-06 A Claude research notebook is the bridge between a helpful chat session and a defensible research artifact. It records what was searched, what Claude saw, what changed, and what a human verified. ### Separate the notebook from the chat Chat history is convenient but not a durable research record. A notebook should live in a place your lab already trusts: a versioned repository, electronic lab notebook, secure shared drive, Zotero notes, or a project-management system with exports. For every meaningful AI-assisted step, record the date, product surface, prompt, input files or source set, output, and human decision. If the step produced a figure, table, or code file, save the file and the generation path together. ### Track source state as a first-class artifact Research often fails because source state is implicit. The notebook should include search strings, database names, result counts, exported metadata, included/excluded record IDs, and the date of each search. For living reviews or fast-moving fields, add a planned refresh cadence. When Claude is used with connectors, MCP, or Claude Science, document which external systems were reachable. This is especially important for proprietary datasets, lab pipelines, and internal notebooks. ### Preserve correction history The most valuable AI notebook entries are not the ones where Claude was right. They are the corrections: a hallucinated citation, a misread figure, a missed exclusion criterion, or a synthesis paragraph that overstated causality. Record the correction and the prompt or guardrail you added afterward. Over time, this becomes a local methods asset. Your lab learns which prompts, source formats, and verification checks work for its field. ### Make the final artifact cite the notebook A manuscript, internal memo, or grant background section should be able to point backward to its evidence table and notebook. That does not mean publishing every prompt. It means preserving enough that an internal reviewer can reproduce the path from source to sentence. For public-facing work, disclose the AI-assisted steps at the level required by the venue, journal, funder, or institution. Sources: - Claude Science, an AI workbench for scientists (Anthropic): https://www.anthropic.com/news/claude-science-ai-workbench - What is the Model Context Protocol? (Model Context Protocol): https://modelcontextprotocol.io/docs/getting-started/intro - Zotero Web API Documentation (Zotero): https://www.zotero.org/support/dev/web_api/v3/basics - OpenAlex API overview (OpenAlex): https://developers.openalex.org/ - Citations (Claude Platform Docs): https://platform.claude.com/docs/en/build-with-claude/citations ## Claude Research Data Privacy and Lab Policy URL: https://clauderesearcher.com/privacy-policy Updated: 2026-07-06 Before putting research material into Claude, identify the account type, data classification, retention controls, institutional policy, and whether human subjects, health data, unpublished IP, or third-party confidential data are involved. ### Classify the data before the tool Researchers often ask "is Claude safe?" too late. Start with the material: published paper, unpublished manuscript, grant draft, code, participant data, protected health information, internal lab protocol, confidential sponsor data, or proprietary dataset. Each category has different rules. If the content is public and already citable, the main risks are accuracy and provenance. If the content is unpublished, regulated, contractual, or personally identifying, the first question is governance, not prompting. ### Account type changes the answer Anthropic's public privacy and help-center materials distinguish commercial products such as Claude for Work and the Anthropic API from consumer products. For commercial products, Anthropic says inputs and outputs are not used to train models by default. Team, Enterprise, Education, cloud, and API usage may also be governed by institutional contracts. Do not assume a personal paid account has the same controls as an institutional account. Researchers should ask their administrator which plan, data-processing agreement, retention policy, and model-improvement setting applies. ### Retention is part of research risk Data retention is not only a compliance setting; it affects reproducibility and confidentiality. Enterprise controls can define how long conversations and project data remain available. A lab should decide what belongs in Claude history, what belongs in the official notebook, and what should never leave approved systems. For human-subjects or clinical material, consult IRB, privacy, and legal guidance before using any AI assistant. De-identification, minimum necessary data, and approved environments matter more than clever prompts. ### Write a lab AI-use checklist A practical lab policy can be short: allowed data classes, prohibited data classes, approved accounts, required disclosure language, citation-verification standard, retention expectations, and escalation path for uncertain material. The policy should also state that Claude output is not a source of truth. It is an assistant output that must be verified against primary sources, code, data, or expert review before use. Sources: - Is my data used for model training? (Anthropic Privacy Center): https://privacy.claude.com/en/articles/7996868-is-my-data-used-for-model-training - Configure custom data retention controls for Enterprise plans (Claude Help Center): https://support.claude.com/en/articles/10440198-configure-custom-data-retention-controls-for-enterprise-plans - Who owns and manages the data of my team? (Claude Help Center): https://support.claude.com/en/articles/9265372-who-owns-and-manages-the-data-of-my-team - Citations (Claude Platform Docs): https://platform.claude.com/docs/en/build-with-claude/citations ## Claude Research Prompts That Preserve Evidence URL: https://clauderesearcher.com/prompts Updated: 2026-07-06 The best Claude research prompts are not magic words. They are forms that force the model to declare scope, cite passages, mark uncertainty, and keep human-verifiable fields. ### Search-design prompt Use this before touching a database: "I am reviewing [topic] in [population/corpus] from [date range]. Generate candidate search terms, controlled-vocabulary candidates, synonyms, adjacent terms, and likely false positives. Return a table with term, why it matters, where to test it, and what it might accidentally retrieve." Follow-up: "Convert the strongest term groups into database-specific query drafts for PubMed, OpenAlex, and arXiv. Do not claim these searches have been run." ### Paper-triage prompt Use this after exporting records: "Classify each record against these inclusion criteria: [criteria]. Use labels include, exclude, maybe, wrong corpus. Give one abstract-grounded reason and quote no more than a short phrase from the abstract as the anchor. If the abstract is insufficient, label maybe." This prompt makes uncertainty visible. The "maybe" pile is where expert time should go. ### Evidence-extraction prompt Use this with included full text: "Extract the following fields only from the provided source: study design, sample/corpus, method, intervention/exposure, comparator, outcome, effect or finding, limitation, funding/conflict, and passage marker. Use not reported when absent. Do not infer values from background knowledge." Follow-up: "Which extracted fields are weakest, ambiguous, or dependent on a figure/table I should inspect manually?" ### Synthesis and contradiction prompt Use this after the evidence table is locked: "Draft a synthesis from the evidence table only. For each paragraph, list the source IDs that support it. Add a separate section called Claims to weaken, naming the evidence that most complicates each paragraph." This turns Claude into an argument drafter and checker, not a replacement for literature judgment. ### Reproducibility prompt For computational papers or Claude Science outputs, use: "Create a reproducibility checklist from this method and artifact history. Include data availability, code availability, environment, package versions, random seeds, hardware or compute assumptions, external tools, and any step that cannot be reproduced from the record." The output should become a notebook entry and, when appropriate, a reviewer question. Sources: - Citations (Claude Platform Docs): https://platform.claude.com/docs/en/build-with-claude/citations - Web search tool (Claude Platform Docs): https://platform.claude.com/docs/en/agents-and-tools/tool-use/web-search-tool - PRISMA statement (PRISMA): https://www.prisma-statement.org/ - Entrez Programming Utilities (NCBI): https://www.ncbi.nlm.nih.gov/home/develop/api/ - OpenAlex API overview (OpenAlex): https://developers.openalex.org/ - Zotero Web API Documentation (Zotero): https://www.zotero.org/support/dev/web_api/v3/basics ## AI Hallucinated Citations: A Researcher Verification Protocol URL: https://clauderesearcher.com/ai-hallucinated-citations Updated: 2026-07-06 A hallucinated citation is not only a fake paper. It can be a real paper attached to the wrong claim, a distorted author list, an impossible DOI, or an abstract-only source treated like full-text evidence. ### Treat every AI citation as unverified until it passes two checks The first check is existence: can you resolve the URL, DOI, PMID, arXiv ID, publisher page, database record, or library item? The second check is claim support: does the source actually say what the sentence needs it to say? Researchers often stop after the existence check. That misses the most common high-friction failure in AI-assisted writing: a real source is attached to an overstated, outdated, or unrelated claim. Verification has to operate at the sentence level. - Existence check: source can be opened or resolved in a trusted system. - Metadata check: title, authors, venue, year, and identifier match. - Support check: passage, table, figure, or section supports the exact claim. ### Classify the failure before fixing it A fabricated citation needs deletion or replacement. A mismatched citation needs a source that actually supports the claim, or the claim needs to be rewritten. A weak citation may be usable only if the evidence level is disclosed, such as abstract-only or metadata-only. The fastest way to avoid false confidence is to keep a failure label in the notebook. Use labels such as nonexistent source, bad identifier, metadata mismatch, source does not support claim, abstract-only support, retracted or corrected source, and contradiction found. ### Use registries and libraries for the mechanical checks Crossref is useful for DOI metadata, PubMed for biomedical records and PMIDs, arXiv for preprint identifiers, OpenAlex for broad discovery metadata, and Zotero for preserving the final library state. Claude can help format checks, but these systems should carry the authoritative record. If a source has no stable identifier, search by title and author in the field database or publisher site. If you still cannot find it, do not cite it as real. Ask Claude to mark the source as unresolved rather than producing a polished replacement citation. ### Record the human decision The final verification artifact should be short: citation, claim, identifier, source status, supporting passage or section marker, evidence level, contradiction status, and human decision. This note can live in Zotero, a spreadsheet, a manuscript comment, or a project notebook. When a citation fails, preserve the failure. A correction history is valuable because it shows which prompts, source formats, or workflows are producing risk. Over time, the lab learns where Claude is reliable and where it needs tighter guardrails. Sources: - Citations (Claude Platform Docs): https://platform.claude.com/docs/en/build-with-claude/citations - Web search tool (Claude Platform Docs): https://platform.claude.com/docs/en/agents-and-tools/tool-use/web-search-tool - Crossref REST API (Crossref): https://www.crossref.org/documentation/retrieve-metadata/rest-api/ - Entrez Programming Utilities (NCBI): https://www.ncbi.nlm.nih.gov/home/develop/api/ - OpenAlex API overview (OpenAlex): https://developers.openalex.org/ - Zotero Web API Documentation (Zotero): https://www.zotero.org/support/dev/web_api/v3/basics ## Claude Researcher Updates URL: https://clauderesearcher.com/updates Updated: 2026-09-19 This page tracks dated changes to the site and the Claude research surface. The most recent sweep, on September 19, 2026, added the scientists program explainer and refreshed model guidance for Fable 5.1 and Opus 5. ### September 19, 2026 - scientists program explainer published Added a dedicated explainer for the scientists program Anthropic announced on August 27, 2026: 10,000 Claude seats, standard access free, a premium tier at $15 per month described as roughly five times the usage, for a one-year term. The page covers what the seats include, what they do not, and the model limits that come with them. The editorial emphasis is deliberate. A free seat lowers the cost of research work; it does not change citation verification, data governance, or who is accountable for a published claim. The explainer says so plainly rather than treating the announcement as a capability unlock. - [The Claude scientists program: 10,000 seats explained](/scientists-program) - New page. - [Anthropic: Expanding support for scientists](https://www.anthropic.com/news/expanding-support-for-scientists) - Primary source, dated 2026-08-27. ### September 19, 2026 - model guidance refreshed for Fable 5.1 and Opus 5 The research mode-selection page now carries a model-choice layer, because mode selection alone stopped being a complete answer. Claude Fable 5.1 arrived on September 1, 2026 as the most capable generally available model, but it still blocks professional biology and drug-development work. Since the Opus 5 launch on July 24, 2026, biology-related requests that Fable declines route to Opus 5, which makes Opus 5 the generally available research-capable model for those domains. Two stale framings were removed in the same pass. Claude Science is no longer described only through its June 2026 beta language, since the August scientists announcement still points at it as the research workbench. And the site now records that Sonnet 5 introductory pricing at $2 per million input and $10 per million output became the standard rate rather than rising on September 1. Anthropic also describes Mythos 5.1 life-sciences trusted access as enrolling its first participants. That is a restricted, separately reviewed track, and the site treats it as such rather than as something researchers can generally obtain. - [Research modes and model choice](/research) - Updated page. - [Anthropic: Claude Fable 5.1 and Mythos 5.1](https://www.anthropic.com/claude-fable-and-mythos-5-1) - Primary source, dated 2026-09-01. - [Anthropic: Claude Opus 5](https://www.anthropic.com/news/claude-opus-5) - Primary source, dated 2026-07-24. ### September 19, 2026 - adjacent changes noted, not adopted Three product changes since the July baseline matter to researchers without belonging to this site’s lane. Anthropic announced text watermarking on August 14, 2026 for future Claude models, describing it as adding no hidden characters or extra tokens and as not traceable to a person, organization, or conversation. Researchers drafting with Claude should know it exists; the full explanation lives on a sibling site. Claude memory now spans chat and Cowork with topic-level edit and delete, on by default for Free, Pro, and Max accounts and admin-gated and off for Team and Enterprise until a user enables it. The research-relevant implication is narrow: memory is not a lab notebook and should not become the only record of a search or a decision. Claude in Chrome also reached general availability on paid plans. None of these changed a workflow page on this site. They are recorded here so the next sweep does not rediscover them. - [Claude Helps: watermarking and Claude in Chrome](https://claudehelps.com) - Plain answers on watermarking and everyday Chrome usage. - [Claude Memories: how memory works across surfaces](https://claudememories.com) - Memory controls, scope, and defaults. - [Auditable Claude research notebook](/lab-notebook) - Why chat memory is not a substitute for a research record. ### July 6, 2026 - initial source baseline ClaudeResearcher.com launched with a source baseline covering Anthropic's Claude Science announcement, Claude Research help article, Claude Platform citations documentation, web-search documentation, Claude for Life Sciences announcement, PRISMA 2020 materials, Cochrane responsible-AI evidence-synthesis note, and official API documentation for PubMed/NCBI, OpenAlex, arXiv, Crossref, Zotero, and ORCID. The main editorial change from ordinary AI-research commentary is the workflow boundary: Claude is framed as a research assistant that operates around authoritative systems and auditable notebooks. ### What will be tracked here Future updates should record changes that affect research workflows: Claude citation behavior, Research availability, Claude Science platform changes, scientific connectors, data-retention guidance, and major updates to research-reporting standards. Routine model-ranking news and general Claude product launches should be covered by sibling sites such as Claude Weekly or Claude Reports unless they directly change literature review or scientific workflow practice. ### Source review checklist A weekly review should open the primary Anthropic documentation used by the site, check whether the Claude Science availability language changed, confirm citation and web-search docs still describe the same fields, and inspect whether PRISMA, Cochrane, or key scholarly APIs changed relevant guidance. Every update should preserve the date, source URL, affected page, and editorial action taken. Sources: - Expanding support for scientists (Anthropic): https://www.anthropic.com/news/expanding-support-for-scientists - Claude Fable 5.1 and Mythos 5.1 (Anthropic): https://www.anthropic.com/claude-fable-and-mythos-5-1 - Claude Opus 5 (Anthropic): https://www.anthropic.com/news/claude-opus-5 - Pricing (Claude Platform Docs): https://platform.claude.com/docs/en/about-claude/pricing - Claude text watermarking (Anthropic): https://www.anthropic.com/news/claude-text-watermark - Claude Science, an AI workbench for scientists (Anthropic): https://www.anthropic.com/news/claude-science-ai-workbench - Citations (Claude Platform Docs): https://platform.claude.com/docs/en/build-with-claude/citations - PRISMA 2020 statement (PRISMA): https://www.prisma-statement.org/prisma-2020 - Setting the standards for responsible AI use in evidence synthesis (Cochrane): https://www.cochrane.org/about-us/news/setting-standards-responsible-ai-use-evidence-synthesis ## About Claude Researcher URL: https://clauderesearcher.com/about Updated: 2026-07-06 ClaudeResearcher.com is an independent educational publication about using Claude in scientific research and literature review workflows. It is not affiliated with Anthropic. ### What this site covers ClaudeResearcher.com covers practical research workflows: literature review, paper triage, source verification, scholarly databases, systematic reviews, research notebooks, data handling, and prompts that preserve evidence. It does not try to be a general Claude news site, a model benchmark site, a Claude Code tutorial, or a speculative autonomous-science publication. Those topics are better handled by sibling sites in the Claude Network. ### Source policy Claims about Claude features are tied to Anthropic, Claude Help Center, Claude Platform Docs, or the Model Context Protocol documentation. Claims about literature review and evidence synthesis are tied to PRISMA, Cochrane, or official scholarly infrastructure documentation. The site avoids fake testimonials, invented benchmarks, and unsupported claims that Claude can replace expert review. When a page gives a workflow, it also names what the human researcher remains responsible for. ### Independence disclaimer ClaudeResearcher.com is an independent educational publication. It is not affiliated with, sponsored by, or endorsed by Anthropic. Claude is a trademark of Anthropic, PBC. Product availability, model names, plan details, and privacy controls can change; readers should verify current details against official sources. Nothing on this site is legal, medical, research-integrity, IRB, or compliance advice. For sensitive research data, use institutional policy and professional review. Sources: - Claude Science, an AI workbench for scientists (Anthropic): https://www.anthropic.com/news/claude-science-ai-workbench - Use research on Claude (Claude Help Center): https://support.claude.com/en/articles/11088861-use-research-on-claude - Citations (Claude Platform Docs): https://platform.claude.com/docs/en/build-with-claude/citations - PRISMA statement (PRISMA): https://www.prisma-statement.org/ - Setting the standards for responsible AI use in evidence synthesis (Cochrane): https://www.cochrane.org/about-us/news/setting-standards-responsible-ai-use-evidence-synthesis ## AI Literature Review Protocol Builder URL: https://clauderesearcher.com/tools/lit-review-builder Updated: 2026-07-06 Turn a research question, database plan, criteria, and screening stages into a reusable protocol plus stage-specific Claude prompts. The tool runs entirely in your browser. Target queries: - ai literature review workflow - claude literature review workflow - literature review protocol builder How it works: - Enter the review question and corpus boundary. - Choose scholarly databases and add field-specific sources. - Write inclusion and exclusion criteria. - Select screening stages and evidence fields. - Generate the protocol and Claude prompt pack. - Copy or download the markdown for your project notebook. FAQ: - Does this tool search PubMed, OpenAlex, or arXiv? No. It prepares a protocol and prompts. Run the actual searches in the scholarly databases and record the search strings, dates, filters, and result counts. - Can I use the output as a systematic review protocol? Use it as a draft aid, not a replacement for review-team judgment, registration requirements, or institutional methods review. ## AI Citation Integrity Checklist URL: https://clauderesearcher.com/tools/citation-checker Updated: 2026-07-06 Use this checklist when Claude or another AI assistant gives you a citation. It produces a risk score and a short action plan while keeping all notes in your browser. Target queries: - ai hallucinated citations - verify ai citations - claude citation checker How it works: - Paste the citation, URL, DOI, PMID, arXiv ID, or title. - Write the exact claim the citation is supposed to support. - Answer the verification checklist from the source itself. - Review the risk score and required next actions. - Copy the markdown note into your evidence table or notebook. FAQ: - Does the checklist prove that a citation is correct? No. It helps you structure verification. You still need to inspect the source, passage, and metadata yourself. - Why include retraction and contradiction checks? A source can exist and support a narrow claim while still being unsuitable because it is retracted, superseded, or contradicted by stronger evidence. ## Claude Research Prompt Library URL: https://clauderesearcher.com/tools/research-prompts Updated: 2026-07-06 Find a prompt by research stage, output type, and source discipline. Each prompt includes adaptation notes that keep Claude tied to evidence instead of inventing citations. Target queries: - claude prompts for research - research prompt library - claude literature review prompts How it works: - Choose the research stage or leave it set to all prompts. - Filter by task or search for a phrase. - Open a prompt and read the adaptation notes. - Copy the prompt and replace bracketed fields with project details. - Save the final prompt and Claude output in your research notebook. FAQ: - Are these prompts enough to prevent hallucinated citations? No prompt can guarantee that. The prompts reduce risk by requiring source anchors, missing-value labels, and verification steps. - Can I use these prompts with non-Claude tools? Usually yes. They are written for Claude Researcher workflows, but the evidence rules apply to any AI assistant used in scholarly work.