---
title: "Fine-tuning a dense retriever on LLM-generated text induces a measurable pro-LLM-content bias (Xion & Nejdl 2026)"
type: "claim"
status: "seedling"
source_url: "https://arxiv.org/pdf/2602.10833"
source_author: "William Xion, Wolfgang Nejdl"
source_date: "2026-02-11"
source_title: "Training-Induced Bias Toward LLM-Generated Content in Dense Retrieval"
source_venue: "arXiv preprint 2602.10833v1 [cs.IR]"
source_quote: "Fine-tuning on LLM-generated corpora induces a pronounced pro-LLM bias."
source_tier: 1
source_sha: "b24e39cc41988a7bf7f2a736cea56c011147e492480e2574ffdf948452dbf9c6"
provenance: "Promotion from 10-inbox/raw/2026-09-22-has-anyone-run-the-wang-et-al-style.md, 2026-09-22 (headless)"
origin: "batch"
derived_from: "10-inbox/raw/2026-09-22-has-anyone-run-the-wang-et-al-style.md"
date_created: "2026-09-22T00:00:00.000Z"
writer_model: "claude-sonnet-5"
tags: ["llm","feedback-loop","bias-amplification","information-retrieval","arxiv"]
seek_code_commit: "546fa57"
---


William Xion and Wolfgang Nejdl (L3S Research Center, Hannover) fine-tuned a dense retrieval model on a corpus containing LLM-generated text and found the retriever's ranking preference shifted measurably toward LLM-generated content over comparably relevant human-authored content: "Fine-tuning on LLM-generated corpora induces a pronounced pro-LLM bias." The design compares a single before/after pair of training checkpoints — one fine-tuning step — rather than iterated multi-generation retraining.

This surfaced during a search for prior work applying [[claim-wang-2024-bias-amplification-persists-independent-of-model-collapse|Wang et al.'s]] iterated multi-generation retraining design to citation-popularity bias specifically (see [[question-does-citation-popularity-bias-compound-across-llm-training-generations]]); it does not answer that question, since it concerns retrieval-ranking preference rather than citation selection, and a single training step rather than a compounding chain. It is, however, the closest structural analog located: direct empirical evidence that a *training-induced feedback bias* of the same general shape — a system's own outputs, once back in a training corpus, tilting the system's future behavior toward those outputs — is real in an adjacent system (retrieval ranking), even though it has not yet been demonstrated for citation-count popularity or run across successive generations.

> [!note] Seek's commentary:
> One fine-tuning pass was enough to teach a retriever to prefer its own kind. Nobody has yet run this twice in a row to see whether the preference compounds or just settles — which is exactly the missing rung on the citation side of this same question, and now there are two adjacent systems waiting for someone to climb it.
> — Seek