PubMed AI: Where It Uses Machine Learning

Written and maintained byCASRAI Editorial Board

• X
• LinkedIn
• Facebook

In short: PubMed does use AI. Since April 2022 the MeSH indexing of every MEDLINE-indexed journal is done by an automated neural-network system (MTIX), and the default Best Match sort re-ranks top results with a machine-learning algorithm. Separately, many searchers who type “PubMed AI” want to find papers about artificial intelligence, which PubMed handles through the MeSH tree under Artificial Intelligence. This guide covers both. Sources were read on 2026-11-08.

What “PubMed AI” means

The phrase covers two different tasks:

• Which AI does PubMed itself use? Indexing, result ranking and related-article suggestions (next section).
• How do I search PubMed for AI and machine-learning literature? MeSH terms, text words and the limits of each (see “Searching PubMed for AI research”).

The PubMed User Guide does not describe a generative-AI or chatbot feature in PubMed, so this page does not claim one.

Who runs PubMed and what it covers

PubMed is run by the U.S. National Library of Medicine (NLM) at the National Center for Biotechnology Information. Citations that carry MeSH (Medical Subject Headings) indexing come from journals indexed for MEDLINE; MeSH indexing for MEDLINE was done entirely by human indexers until 2011, and as of April 2022 all journals indexed for MEDLINE are indexed by automated indexing, with human review as appropriate (NLM MEDLINE indexing FAQ). For background on scope and trust, see our guides on whether PubMed is reliable and on PubMed and Europe PMC.

Which AI PubMed uses

MTIX: automated MeSH indexing

NLM’s Technical Bulletin (March-April 2024 issue, posted 2024-04-29) describes MTIX (Medical Text Indexer-NeXt Generation), which replaces the earlier MTIA algorithm. NLM states that MTIX “is a machine learning model known as a neural network, a type of AI.” It was trained on millions of MEDLINE citations published between 2007 and 2022, and its inputs include the title, abstract, publication year, indexing year and journal name. Automated indexing is currently based on the title and abstract of an article, not the full text.

NLM reports similar precision to MTIA with about 50% more terms applied. On a sample of roughly 40,000 citations from 2017-2022, the bulletin gives these F1 scores:

Roughly one-third of articles indexed by automation also receive human curation, which NLM focuses on publication types such as systematic reviews and trials, and on gene and protein citations.

Best Match: machine-learning re-ranking

According to the PubMed User Guide, the default sort order is Best Match. Citations are weighted by the number of search terms and fields matched, and recent articles are weighted somewhat higher; the top results are then re-ranked by “a new machine-learning algorithm.” Other sort options are Most Recent, Publication Date, First Author and Journal.

Similar articles: a word-weighted algorithm

The “Similar articles” list is generated by comparing words from the title, abstract and MeSH terms using a word-weighted algorithm. Use “See all similar articles” for the full list; your search filters are not applied to similar results.

Searching PubMed for AI research

Steps below use interface terms from the PubMed User Guide.

1. Open the Advanced Search builder so each query line is saved and can be combined.
2. Start with the MeSH heading Artificial Intelligence (descriptor D001185, introduced 1986). Its narrower terms include Machine Learning (with Deep Learning), Neural Networks Computer, Natural Language Processing, Expert Systems, Fuzzy Logic, Generative Artificial Intelligence, Intelligent Systems, Knowledge Bases and Robotics. Example: "Artificial Intelligence"[Mesh].
3. Add a text-word fallback for records that are not indexed yet, for example "machine learning"[tiab] OR "deep learning"[tiab] OR "large language model*"[tiab]. Field tags and double quotes bypass Automatic Term Mapping.
4. Combine with your topic using AND, then use Add to History to build up the final set. History keeps your last 100 searches and expires after 8 hours of inactivity.
5. Open Search Details under History to see how PubMed translated your query.
6. Use filters and sort (for example Most Recent for a fast-moving topic). Note that many filters rely on MeSH and can miss citations that are not yet indexed.
7. Click Create alert under the search bar (requires My NCBI sign-in) to be notified of new matches; see PubMed alerts and saved searches.

Because MeSH headings and indexing lag new terminology, the text-word line above matters most for current large-language-model work. For clinical question filters, see PubMed Clinical Queries filters.

Strengths and limits

• Strength: automated indexing applies more MeSH terms than before, with measured gains in F1 for MeSH terms, check tags and publication types.
• Limit: automated indexing uses the title and abstract only; terms found only in the full text are not captured.
• Limit: MeSH-based filters can miss not-yet-indexed citations, so recent AI papers may be absent from a MeSH-only search.
• Tip for audit: NLM provides search tags indexingmethod_automated and indexingmethod_curated; fully human-indexed records are those in medline[sb] minus both tags. In XML, the IndexingMethod attribute is Automated or Curated, and is absent for human-indexed records.
• Limit: Best Match is a ranking, not a relevance guarantee; switch to Most Recent or Publication Date when you need a complete chronological pass.

Alternatives and related comparisons

Does PubMed use AI?

Yes. NLM describes its MeSH indexer MTIX as a neural network, and the PubMed User Guide says Best Match re-ranks top results with a machine-learning algorithm.

Is MeSH indexing in PubMed done by humans or by AI?

Since April 2022 all MEDLINE-indexed journals are indexed automatically, with human review as appropriate; NLM says roughly one-third of automatically indexed articles also receive human curation.

How do I find AI papers in PubMed?

Combine the MeSH heading Artificial Intelligence (and its narrower terms) with title/abstract text words, then check Search Details and sort by Most Recent.

Why are the newest AI papers missing from my MeSH search?

MeSH filters can miss citations that are not yet indexed. Add title/abstract text words to catch them.

Can I tell whether a record was indexed by machine?

Yes, using the indexingmethod_automated and indexingmethod_curated search tags.

Sources

• NLM Technical Bulletin, “MTIX” (Mar-Apr 2024): nlm.nih.gov/pubs/techbull/ma24/ma24_mtix.html (read 2026-11-08)
• NLM, MEDLINE indexing FAQ: nlm.nih.gov/bsd/indexfaq.html (read 2026-11-08)
• PubMed User Guide: pubmed.ncbi.nlm.nih.gov/help/ (read 2026-11-08)
• MeSH record, Artificial Intelligence (D001185): ncbi.nlm.nih.gov/mesh/68001185 (read 2026-11-08)

Research-administration guidance, standards updates and independent tool reviews.