Why LLMs Cite Academic Papers More Often Than Commercial Pages

The claim is familiar, but the evidence for it is narrower than it sounds. The large studies of AI citations, such as Pew’s and Ahrefs’, do not compare how often academic papers and commercial pages are cited across everyday queries. What the evidence does show is a set of tendencies that favor certain kinds of sources: models trained on data weighted toward higher-quality text, a measured bias toward already highly cited research when models generate references, and a higher share of official sources in some AI answers. Those tendencies help scholarly and official material. They do not mean commercial pages are disadvantaged by being commercial.

What the research does show

Models favor highly cited work when they generate references. In a study published in the Findings of NAACL 2025, Algaba and colleagues asked language models to suggest scholarly references for papers, relying only on what the models had learned, without search. The suggestions closely resembled human citation patterns but showed “a more pronounced high citation bias”: the models leaned toward papers that were already highly cited. The models also produced non-existent references that looked like real ones. This finding is about academic references in an academic task. It shows how models treat scholarly work, not how often they choose a paper over a product page.

Training data is weighted toward sources judged higher quality. OpenAI’s GPT-3 paper states that datasets “we view as higher-quality are sampled more frequently” during training. In that model, WebText2, one of the two book collections and Wikipedia were each seen more than once during training, while most of the general web crawl was seen less than once. That describes a 2020 model, but the principle favors carefully edited text of the kind found in scholarly and reference sources.

Official sources appear more often in some AI answers. Pew Research Center’s March 2025 data showed that 6% of the sources linked in AI summaries were .gov sites, compared with 2% of the links in standard results. Pew did not report a figure for academic sites, but the pattern shows that AI summaries can lean toward authoritative institutional sources more than standard results do.

Why the premise needs care

Three cautions apply before concluding that commercial pages lose out.

• Query type decides. For a medical, scientific or policy question, academic and official sources can be the most relevant. For a question about a product’s price or features, the company’s own page may be the best source. A comparison that mixes these query types says little.
• “Commercial” is not a quality judgment. A manufacturer’s technical documentation, a software company’s API reference or a retailer’s detailed product data can be the most accurate source on its subject. Google’s AI optimization guide contrasts commodity content with non-commodity content that offers first-hand, specific information. That distinction cuts across academic and commercial lines.
• Training and retrieval differ. The Algaba finding concerns references generated from a model’s training. When an AI system searches the web, which pages it cites depends on its search index and ranking, where commercial pages are ranked by the same systems as any other page.

What commercial publishers can take from this

The traits that help scholarly and official sources are ones a commercial page can have too.

1. Cite primary sources. Link to the studies, standards, regulations or official data behind your claims. The GEO research presented at KDD 2024 found that adding citations from reliable sources was among the methods that most improved a source’s visibility in generative answers, in the authors’ test environment.
2. Publish original, checkable information. Test results, specifications, methodology and data you collected yourself are what make a page worth citing, whoever publishes it.
3. Separate information from promotion. A page that explains how something works, with evidence, is a more useful source than one that mostly sells. Keep sales messages on pages designed for them.
4. Be precise and dated. State figures with units and dates, and update them when they change.
5. Earn references from authoritative sources. If researchers, standards bodies or reputable publications cite your work, that corroboration strengthens how your brand is represented in the public record models learn from.

Read the “academic beats commercial” claim as a statement about source quality, not source type. Scholarly and official material benefits from careful editing, citation and wide reference, and research on model-generated references shows a bias toward the most cited work. A commercial page built on the same qualities can compete on the same footing.