PubMed Search on the platform

PubMed Search lets you find biomedical literature — over 36 million citations from PubMed/MEDLINE — using natural language or Boolean queries. Reach for it to survey a target or disease, investigate a drug mechanism, or gather papers before starting a campaign. Ask the model to extract entities from the returned abstracts and it reads them directly — no separate entity-extraction tool call is needed. See what you can achieve.

Try it

What PubMed Search does

PubMed Search is a literature retrieval tool that queries the PubMed/MEDLINE database maintained by the National Center for Biotechnology Information (NCBI). PubMed indexes over 36 million citations from biomedical journals, conference proceedings, and preprint servers. On the platform, PubMed Search runs when you ask Ivy in chat (there is no form or job for it); it calls NCBI directly, runs entirely on CPU with no GPU requirements, returning results in 3-30 seconds depending on query complexity.

The tool uses the NCBI E-utilities API under the hood: a PubMed search runs ESearch to find matching PubMed IDs (PMIDs), then ESummary and EFetch in parallel to retrieve metadata and full abstracts for the matched records. The platform handles API key management, rate limiting, and query construction automatically.

Citations

Sayers EW, Beck J, Bolton EE, Brister JR, Chan J, Comeau DC, et al. Database resources of the National Center for Biotechnology Information. Nucleic Acids Research. 2024;52(D1):D33-D43. doi: 10.1093/nar/gkad1044

Key capabilities

• Natural language search: Type a question in chat and the platform translates it into a structured PubMed query
• Boolean queries: Use AND, OR, NOT operators and field tags like [Title], [Author], [MeSH Terms] for precise filtering
• Date filtering: Restrict results to a specific publication date range
• Article type filtering: Filter by reviews, clinical trials, meta-analyses, or case reports
• Author and affiliation search: Find papers by specific researchers or institutions
• Result chaining: Ask the model to extract entities directly from the returned abstracts, or feed results into Literature Passports

How it works technically

PubMed Search on the platform wraps three NCBI E-utilities endpoints across two search modes:

1. ESearch (esearch.fcgi): Accepts a search query and returns a list of matching PubMed IDs. Supports the full PubMed query syntax including field tags, Boolean operators, and date ranges.

2. ESummary (esummary.fcgi): Takes the list of PMIDs from ESearch and returns structured metadata for each article, including title, authors, journal, publication date, and DOI.

3. EFetch (efetch.fcgi): Returns full abstract text, MeSH descriptor terms, and author keywords for the matched PMIDs.

A plain search calls all three (ESummary and EFetch in parallel), so its results carry abstracts and MeSH terms. A filtered search — by author, affiliation, date range, or article type — calls only ESearch and ESummary, so its results have no abstract or MeSH terms. Field tags written into the query itself (for example 2024[DP]) keep a search plain.

Query translation

When you type a natural language query in chat, the orchestrator analyzes your intent and constructs an appropriate PubMed query. For example:

• "Find papers about EGFR resistance to osimertinib" becomes EGFR AND resistance AND osimertinib
• "Reviews on GLP-1 receptor agonists published in 2024" becomes GLP-1 receptor agonists AND review[PT] AND 2024[DP]
• "Papers by John Smith on kinase inhibitors" becomes Smith J[AU] AND kinase inhibitors

You can also provide raw PubMed syntax directly if you prefer explicit control.

Rate limiting

NCBI enforces rate limits on E-utilities (3 requests/second without an API key, 10 requests/second with one). The platform manages this automatically using a token-based rate limiter shared across all NCBI tools (PubMed, Entrez, BLAST enrichment). You do not need to configure or manage API keys.

When to use PubMed Search

Primary use cases

Literature review for a target or disease

• Search for publications related to a therapeutic target before starting a drug discovery campaign
• Identify key papers, review articles, and recent findings in a specific disease area
• Track publication trends over time to gauge research interest

Drug mechanism investigation

• Find papers describing mechanism of action for existing drugs
• Identify resistance mechanisms and combination strategies
• Locate preclinical and clinical data supporting drug-target hypotheses

Competitive intelligence

• Monitor publications from competitor research groups
• Track clinical trial publications for drugs in development
• Identify emerging targets and therapeutic approaches through publication trends

Data sourcing for passports

• Literature Passport aggregates PubMed results with LLM-driven entity extraction
• Target Passport uses literature data to assess target validation level
• Therapeutic Area Passport incorporates publication trends for disease landscaping

When to use alternatives

Integration with the platform ecosystem

Upstream data sources

PubMed Search accepts input from:

• Free-text chat queries: Natural language descriptions of what you are looking for
• Therapeutic Area Passport: Disease names extracted during passport generation
• Target Passport: Gene and protein names used as search terms

Downstream applications

PubMed results feed into:

• Entity extraction: The model reads retrieved abstracts directly and pulls out genes, proteins, compounds, and diseases — no separate tool call
• Literature Passport: Comprehensive literature analysis combining search, entity extraction, and trend analysis
• Semantic Scholar: Retrieve citation graphs for key papers identified in PubMed results
• Target Passport: Literature evidence supporting target druggability assessment

Workflow chaining examples

Literature-to-target pipeline:

1. PubMed Search for "KRAS G12C inhibitors"
2. Model extracts protein names directly from the returned abstracts
3. UniProt fetch for extracted protein names
4. Target Passport generation

Competitive intelligence pipeline:

1. PubMed Search for a specific drug name
2. Clinical Trials Search for the same compound
3. Therapeutic Area Passport combining both data sources

Scope and limitations

Coverage

• Database size: 36+ million citations (as of 2026)
• Update frequency: Daily additions from contributing journals
• Content scope: Biomedical and life science journals, plus selective coverage of chemistry, physics, and engineering journals relevant to biomedicine
• Historical depth: Records dating back to 1966, with some older citations

Limitations

• Abstracts only: PubMed provides titles and abstracts, not full-text articles. For full text, use Europe PMC or access through institutional subscriptions.
• English bias: While PubMed indexes journals in many languages, the majority of indexed content is in English. Non-English abstracts may have limited discoverability.
• Indexing lag: Newly published articles may take days to weeks to appear in PubMed, depending on the journal's submission practices.
• MeSH lag: MeSH term indexing can lag behind initial publication by several weeks. Very recent articles may not have MeSH annotations.
• No citation counts: The basic PubMed search does not return citation counts. For citation data, chain into Semantic Scholar.

Performance characteristics

• Typical runtime: 3-10 seconds for standard queries, up to 30 seconds for complex Boolean queries returning many results
• Maximum results: 100 articles per query (20 by default; ask for more or fewer)
• Deterministic: Same query returns the same results for the same database state (results may change as new articles are indexed)

References

• NCBI E-utilities documentation: https://www.ncbi.nlm.nih.gov/books/NBK25500/
• PubMed Help: https://pubmed.ncbi.nlm.nih.gov/help/
• PubMed search syntax: https://www.ncbi.nlm.nih.gov/books/NBK3827/

Chat Usage

Example Prompts

• "Search PubMed for CRISPR gene editing in cancer treatment"
• "Find recent publications on GLP-1 receptor agonists from the last 2 years"
• "Search PubMed for 'EGFR AND resistance AND osimertinib'"

Workflow Chaining

• Feeds into: LLM entity extraction (direct, no separate tool call), Literature Passport
• Accepts from: Free-text queries, disease names from Therapeutic Area Passport

FAQ

Read next