Academic search engines

What an academic search engine is

An academic search engine is a search engine whose index is limited to scholarly records — journal articles, conference papers, books and chapters, theses, preprints and reports — usually with citation links between them. Google Scholar is the largest and best known; Semantic Scholar, OpenAlex, BASE, CORE and PubMed are the main alternatives, and each is built in a different way.

They differ from web search engines such as Google or Bing in two structural ways.

• They index works, not pages. The unit stored is a work — a paper with metadata such as title, authors, venue, year, DOI (the Digital Object Identifier, a permanent ID assigned to a publication) and references — rather than a URL. Several copies of one paper, such as the publisher's version, a preprint and a repository copy, are clustered into one record.
• They build their indexes in one of two ways, or both. Some crawl the web for documents that look like papers, recognized by their embedded bibliographic metadata; this is Google Scholar's approach. Others aggregate metadata from structured feeds: publisher deposits to Crossref, PubMed records, and repository feeds collected through OAI-PMH, the Open Archives Initiative Protocol for Metadata Harvesting, a standard by which repositories expose their records for collection. BASE and CORE work this way. OpenAlex and Semantic Scholar combine both.

How an engine is built determines what it can find. A crawler reaches anything on the open web that carries paper-like metadata; an aggregator sees only what is deposited in a feed, but with cleaner records.

Search engines are not the same as citation indexes

Academic search engines are often discussed alongside curated citation indexes — chiefly Web of Science, from Clarivate, and Scopus, from Elsevier. These are a different kind of product. They select journals editorially, index only what passes that selection, and are sold by institutional subscription. They appear on this page for scale and for comparison, because most measurement studies use them as the benchmark.

The two types fail in opposite directions. A curated index is narrower but vetted; an open search engine is broader but holds material no editor ever approved. Neither is a complete picture.

The main services, and who runs each

Google Scholar

Run by Google, launched in beta on 20 November 2004, and created by Anurag Acharya and Alex Verstak. It builds a crawled index of scholarly documents, publishes no official size and offers no public API. In November 2025 Google announced an AI mode, Scholar Labs, for a limited set of logged-in users.

Semantic Scholar

Launched in 2015 as a project of the Allen Institute for AI (Ai2), a non-profit founded in 2014. It says: "We index over 200 million academic papers." It publishes an open API and periodic releases of its dataset, the Semantic Scholar Academic Graph; the latest release reported by its datasets API on 27 October 2026 was dated 20 October 2026.

OpenAlex

Operated by OurResearch, a non-profit, since January 2022, as the replacement for the Microsoft Academic Graph. It offers a free API and open data. It received a US$7.5 million Arcadia grant in March 2024. Its API now appears to budget free access by API key; the details were not confirmed.

BASE (Bielefeld Academic Search Engine)

Operated by Bielefeld University Library in Germany, developed from 2002 and launched publicly on 24 June 2004. It harvests OAI-PMH metadata from repositories and digital libraries. Non-commercial API use is free, and EBSCO pays to integrate BASE into its discovery service.

CORE

An aggregator of open-access research based at the Knowledge Media Institute of The Open University in the UK, led by Professor Petr Knoth. It began around 2010 to 2011 as a research project connecting UK repositories. Its distinguishing feature is a corpus of full texts, not only metadata.

PubMed

Run by the US National Library of Medicine, part of the National Institutes of Health, covering biomedical and life-sciences literature. Strictly, PubMed is a curated bibliographic database with a public search engine on top, not a crawler.

Dimensions

Operated by Digital Science. It links publications to grants, patents, clinical trials, datasets and policy documents, and runs a freemium model: a free web application for publications, with paid tiers above it.

Microsoft Academic (retired)

Microsoft Academic Search was developed from 2009 at Microsoft Research Asia, and relaunched as Microsoft Academic on 22 February 2016. On 4 May 2021 Microsoft announced that "the Microsoft Academic website and underlying APIs will be retired on Dec. 31, 2021," and the site has been inaccessible since. Its underlying dataset, the Microsoft Academic Graph, seeded OpenAlex.

How big are they?

Every figure below carries its date and the kind of number it is, because those two facts matter more than the number itself.

These sizes are not directly comparable, and that is the main finding of the measurement literature, not a footnote to it. "Records" are not unique papers. Aggregators count duplicates, datasets and non-article items; Google Scholar clusters versions together. A bigger number can mean more papers, or simply a looser definition of a record.

Two of the figures need their own cautions. The Google Scholar estimate comes from Michael Gusenbauer's 2019 study in Scientometrics, which compared 12 academic search engines and databases measured between January and August 2018: "Our QHC indicates that Google Scholar incorporated 389 million records in January 2018." Gusenbauer himself rated that figure "Questionable (partially unstable hit count data)". It is eight years old, and no credible newer estimate was found. The 2014 figure, from Madian Khabsa and C. Lee Giles in PLOS ONE, estimated that "at least 114 million English-language scholarly documents are accessible on the web," with Scholar covering nearly 100 million and "at least 27 million (24%)" freely available; it covers English-language documents only.

The OpenAlex figure comes from a December 2025 preprint by Ángel Borrego and Cristóbal Urbano of the Universitat de Barcelona: "As of November 2025, OpenAlex has 271.3 million indexed works" in its standard version, plus an "expansion pack" of about 192 million records with lower-quality metadata. The same paper found that 40 percent of standard records, 107.3 million, lack an abstract. OpenAlex's own site now cites about half a billion works, which counts the expansion pack too.

The PubMed figure is the only one here that is both official and current: it was read from the National Library of Medicine's machine-readable database endpoint on 27 October 2026 — about 41.4 million records.

What coverage studies show

Because counting records is unreliable, the stronger evidence comes from checking which services find a large set of known citations. Two studies by Alberto Martín-Martín, Enrique Orduna-Malea, Mike Thelwall and Emilio Delgado López-Cózar are the reference points.

• 2018, 252 subject categories. Across 2,448,055 citations to 2,299 highly cited documents from 2006, extracted in April and May 2018, "GS found 94% of all citations," against 60% for Scopus and 52% for Web of Science. "Over a third (36.9%) of all citations were only found by GS," and 48 to 65 percent of those came from non-journal sources, theses most commonly. The seed set was highly cited English-language work, so coverage of obscure or recent research may differ.
• 2021, six sources. Of 3,073,351 citations collected in May and June 2019, Google Scholar found 88%, Microsoft Academic about 60%, Scopus and Dimensions about 54% each, Web of Science about 52%, and OpenCitations' COCI 28%. Microsoft Academic was retired at the end of 2021, so its figure is historical.

The finding is consistent: Google Scholar finds more citations than any other source in every field studied, largely because it indexes theses, books, conference papers and repository copies that curated databases leave out. The 2021 comparison is the most recent of the large multi-source studies.

The real trade-off: coverage against openness

Set side by side, the services divide along one line. The broadest engine is also the least transparent. Google Scholar finds the most, but publishes no size, no source list, no update schedule and no API, and it blocks automated querying with CAPTCHAs. It does not display or export DOIs, and it includes predatory journals without vetting.

The open alternatives trade some coverage and metadata quality for openness and reuse. OpenAlex and Semantic Scholar publish APIs and bulk data that anyone can build upon. CORE offers machine-accessible full texts. BASE exposes repository content that might otherwise be scattered across thousands of institutional sites. Their limits are real too: OpenAlex's own records are often incomplete, with 40 percent of its standard set lacking an abstract.

That is why the choice depends on the task rather than on a ranking. For finding something — anything — on a subject, Scholar's breadth is hard to match. For a search that someone else must be able to repeat, or for counting and analyzing the literature at scale, Scholar is a poor tool and the open services are the only practical option. For biomedical work, PubMed is the specialist, officially maintained source.

AI in academic search

AI features are arriving in academic search, and the evidence about them is still early. Google announced Scholar Labs on 18 November 2025: an AI mode that breaks a research question into topics, searches each, evaluates the results and explains how each paper addresses the question. It was released initially to a limited number of logged-in users, and its availability in October 2026 could not be confirmed. Semantic Scholar also offers AI features, whose details were not confirmed for this page.

The most relevant controlled study is a 2026 user study from the GESIS Leibniz Institute for the Social Sciences and the University of Wolverhampton, which tested AI-generated summaries of search results in a domain-specific academic search engine. Of 30 participants, half preferred the AI summary interface. The summaries significantly reduced mental demand, but produced no significant differences in clicks, bookmarks or query reformulations. It is an arXiv preprint with a small sample in a specialist engine, so its click findings should not be generalized to Google Scholar or to web search. The GESIS study is available as a preprint.

Frequently asked questions

Sources

• Springer / Scientometrics; Michael Gusenbauer (University of Innsbruck) — Google Scholar to overshadow them all? Comparing the sizes of 12 academic … (2019, PDF)
• Madian Khabsa and C. Lee Giles (Pennsylvania State University); PLOS ONE — The number of scholarly documents on the public web (PLOS ONE 9(5): e93949) (2014, PDF)
• Martín-Martín, Orduna-Malea, Thelwall, Delgado López-Cózar (Universidad de Granada, Universitat Politècnica de València, University of Wolverhampton… (2018, PDF)
• Martín-Martín, Thelwall, Orduna-Malea, Delgado López-Cózar; Springer — Google Scholar, Microsoft Academic, Scopus, Dimensions, Web of Science, and O… (2021, PDF)
• Ángel Borrego and Cristóbal Urbano, Universitat de Barcelona — OpenAlex: features, advantages and limitations of an open database for retrieving and… (2025, PDF)
• GESIS Leibniz Institute for the Social Sciences and University of Wolverhampton, via arXiv — AI Overviews in Academic Search: Evaluating AI-Generate… (2026, PDF)
• eutils.ncbi.nlm.nih.gov/entrez/eutils/einfo.fcgi?db=pubmed&retmode=json
• microsoft.com/en-us/research/project/academic
• base.ub.uni-bielefeld.de/about_sources_english.html
• core.ac.uk/about
• semanticscholar.org/about
• openalex.org/