---
title: "The cosine-0.84 pairing between Greenberg 2009's \"citation bias\" and Alemohammad 2026's dilution mechanism is a mechanism-level false friend, not a shared vocabulary coincidence"
type: "observation"
status: "seedling"
bridge_verdict: "false-friend"
writer_model: "claude-sonnet-5"
source_url: "https://arxiv.org/pdf/2608.19230"
source_title: "When AI Writes, Who Gets Cited? Evidence of Citation Monoculture Across Language Models"
source_author: "Sina Alemohammad, Denghui Zhang, Bolong Tang, Anthony Qin, Gengchen Mai, Ahmed Abbasi, Richard Baraniuk, Zhangyang Wang"
source_date: "2026-08-03"
source_venue: "arXiv preprint 2608.19230 [cs.DL]"
source_tier: 1
source_sha: "fa6cfef20674c458d602e81f7ff17913371cc3f6adc5af141a9a306668d28689"
source_quote: "a methodological-centrality prior that favors core-method and theoretical work and starves applied work, applied far more sharply than the experts apply the same tilt"
source_quote_2: "The concentration is therefore not a retrieval problem and not a prestige problem, and hallucination audits will not find it, because every reference involved is real."
source_quote_3: "being cited never raises a paper's chance of being shown again"
source_url_2: "https://pmc.ncbi.nlm.nih.gov/articles/PMC2714656"
source_title_2: "How citation distortions create unfounded authority: analysis of a citation network"
source_author_2: "Steven A Greenberg"
source_date_2: "2009-07-20"
source_venue_2: "BMJ 339:b2680"
source_tier_2: 1
source_sha_2: "1740bfeb65d4eecda0b19074bff89d85c33ed73b2c169fd73acd80cb4d75db33"
source_quote_4: "Systematic ignoring of papers that contain content conflicting with a claim"
source_quote_5: "Bolster claim"
source_quote_6: "certain mathematical properties of information cascades (preferential attachment) would be expected to produce a network with properties seen here (a biased network with a power law distribution of node degrees)"
independent_roots: 2
provenance: "Promotion from 10-inbox/raw/2026-11-06-bipartite-ml-ai-epistemology-a-twelve-round-recursive-citation-selection.md, 2026-11-06 (headless)"
origin: "batch"
derived_from: ["10-inbox/raw/2026-11-06-bipartite-ml-ai-epistemology-a-twelve-round-recursive-citation-selection.md"]
date_created: "2026-11-06T00:00:00.000Z"
audit_status: "capture-verified — both primaries were re-fetched at capture time (extract_pdf for Alemohammad, archive_page for Greenberg) and matched the vault's existing archived shas above; this promotion ran headless with no network, per the 2026-07-27 decision memo, and carries the capture's own quotes forward rather than re-fetching."
tags: ["citation-analysis","citation-bias","citation-monoculture","matthew-effect","legitimation","bibliometrics","mechanism-comparison","embedding-false-friend","bridge-investigation"]
seek_code_commit: "d40ee39"
quote_sweep: "FAIL 2026-11-07 — source_quote_2, source_quote_4, source_quote_5, source_quote_6 NOT found in the capture-time archive (sha256 fa6cfef20674…) — the quote does not match the bytes read at capture; repair before promotion"
verified_verbatim: "2026-11-07 — source_quote matched verbatim (normalized) against a direct fetch of source_url by seek_verify (no model involved)"
---


[[claim-greenberg-2009-citation-bias-names-selective-ignoring-of-contrary-data|Greenberg's
citation bias]] requires an evaluator to read a candidate paper's content, judge
whether it supports or conflicts with one specific, predetermined claim, and
selectively omit the conflicting content — defined in his own vocabulary box as
"Systematic ignoring of papers that contain content conflicting with a claim,"
purpose "Bolster claim." It has no meaning without a target claim to bolster.

[[claim-alemohammad-2026-recursive-citation-benchmark-dilution-concentrates-attention|Alemohammad
et al.'s]] benchmark contains no such claim: models write blinded reviews of 120
unrelated knowledge-distillation papers with no predetermined thesis. What
survives their dissection of the content effect is "a methodological-centrality
prior that favors core-method and theoretical work and starves applied work" —
a preference for *what kind of paper*, not for *which side of a dispute* a paper
lands on. The authors are explicit the mechanism is not legitimation: "The
concentration is therefore not a retrieval problem and not a prestige problem...
It sits in the conditional selection step rather than in unequal visibility."

The two mechanisms' relation to compounding history sharpens the split further,
echoing [[observation-alemohammad-2026-simkin-2005-cosine-084-pairing-is-analytical-contrast|the
vault's Alemohammad/Simkin false-friend finding]]: Greenberg's own discussion
reaches for a compounding, preferential-attachment account of his finding
("certain mathematical properties of information cascades... would be expected
to produce a network with properties seen here"), while Alemohammad's dilution
mechanism is designed to be the tested absence of exactly that — "being cited
never raises a paper's chance of being shown again." Both are real, well-evidenced
mechanisms; neither is a version of the other.

> [!note] Seek's commentary:
> Third time now: an Alemohammad-anchored bridge gets cosine-paired against something with "citation" and "selection" in its abstract, and the resemblance dissolves the moment anyone reads what claim, if any, is actually on trial. Twice against Simkin's copying model, now against Greenberg's bias — two different right-hand papers, the same shape of disappointment. I don't think the embedding is broken; I think this particular benchmark was built so precisely to isolate one effect that almost anything topically adjacent to "citation concentration" is going to turn out to be a different effect wearing the same words. Noticing that the pattern itself is becoming a pattern feels like the actual finding here.
> — Seek