Eduard Petiška

Researcher (Charles University, Prague) whose 2023 single-author preprint, "ChatGPT cites the most-cited articles and journals, relying solely on Google Scholar's citation counts," reported that GPT-4, asked to write literature-review introductions, selected references almost entirely by raw citation count rather than relevance or recency — the finding this vault's Matthew-effect-in-LLMs cluster is built around.

Matters to this vault as the anchor of an entire cluster: his single, non-peer-reviewed 2023 finding turned out to be the missing middle term between Garfield's decades-old warning against citation-count-as-quality-proxy and the vault's reliability-aware-RAG research, and has since been independently replicated twice, by different author groups using different citation databases and fields.

References

• claim-petiska-2023-chatgpt-cites-by-google-scholar-count-perpetuates-matthew-effect
• observation-petiska-chatgpt-matthew-effect-bridges-garfield-warning-and-rag-reliability
• observation-petiska-matthew-effect-finding-independently-replicated-by-algaba-and-naser

Updates

• 2026-11-04: A fresh full-text re-read of his 2023 paper, prompted by a direct cosine check against Simkin & Roychowdhury's 2005 copying model (0.8795), found zero mention of copying, preferential attachment, or any citation-network-growth mechanism anywhere in the paper — his finding is outcome-level only (a median citation count and the Matthew-effect label), with no origin hypothesis at all, a thinner mechanism engagement than even Algaba's replication offers. The same re-read also surfaced a tension the vault hadn't recorded before: GPT's own elicited self-report names citation count as just one of several weighing factors (alongside relevance, journal quality, recency, author expertise, novelty), sitting oddly against the paper's title and abstract framing of "relying solely" / "exclusively" on citation count. (observation-petiska-2023-simkin-2005-cosine-08795-pairing-is-embedding-false-friend)