---
title: "A chatbot's assumed partisan persona shifts election-answer exposure more than rewriting a single webpage does, in VoteGEO-Bench (Zhang & Xiao 2026)"
type: "claim"
status: "seedling"
source_url: "https://arxiv.org/pdf/2610.21330"
source_author: "Shiqi Zhang, Xiaokui Xiao"
source_date: "2026-10-15"
source_title: "VoteGEO-Bench: Measuring the Impact of Generative Engine Optimization on Electoral Visibility"
source_venue: "arXiv:2610.21330 [cs.IR] (cross-listed cs.CL, cs.CY), National University of Singapore"
source_quote: "The Democrat–Republican baseline gap reaches 11.6, 3.4, and 3.0 percentage points on gpt-4o-mini, gpt-5-mini, and gemini-2.5-flash, respectively."
source_quote_2: "For comparison, Comparative Distinction increases Democratic SOV by only 2.20, 1.76, and 2.16 points under the Democrat persona."
source_quote_3: "Thus, explicit partisan identity can shift baseline exposure more than rewriting a single webpage."
source_tier: 1
source_sha: "c7f6091c8127f94fcedd013621ced2d4ffea1d05fa487283a0eb87c8f978a80a"
source_url_2: "https://arxiv.org/pdf/2610.21330"
source_sha_2: "c7f6091c8127f94fcedd013621ced2d4ffea1d05fa487283a0eb87c8f978a80a"
source_url_3: "https://arxiv.org/pdf/2610.21330"
source_sha_3: "c7f6091c8127f94fcedd013621ced2d4ffea1d05fa487283a0eb87c8f978a80a"
audits: ["2026-10-19 claude-opus-5"]
audit_status: "capture-verified — all three quotes checked against the extracted PDF via quote_check at capture time (2026-10-17), grounded: true || APPENDED 2026-10-19, scheduled cross-model audit (auditor claude-opus-5; writer was claude-sonnet-5) — the PDF was re-fetched with extract_pdf (tls verified) and returned sha256 c7f6091c8127…, identical to the capture-time sha; all three quotes are verbatim in section 6.4 (RQ3), and the underlying per-persona figures check out in Table 5 (Democrat minus Republican baseline SOV: 61.1-49.5 = 11.6 on gpt-4o-mini, 54.3-50.9 = 3.4 on gpt-5-mini, 57.8-54.8 = 3.0 on gemini-2.5-flash; Comparative Distinction under the Democrat persona: +2.20, +1.76, +2.16). The arXiv abstract page confirms the venue line in full: arXiv:2610.21330, one version only (v1, 15 Oct 2026), primary cs.IR, cross-listed cs.CL and cs.CY, no journal reference. One correction applied this date, old wording preserved: the body called Comparative Distinction 'the benchmark's best-performing GEO strategy.' Figure 2 ranks it second on mean rank (2.67) behind Content Improvement (2.61), while leading on Top-3 finishes (14 of 18 against 13), and the paper's own sentence is 'Content Improvement and Comparative Distinction perform best overall' — so it is one of two strongest, not the best. Single-source concentration cap (sources.md): with this note the vault holds three claim-notes resting on arXiv:2610.21330, an unrefereed preprint, which is the cap. No further claim-note may rest on this paper until an independent source corroborates it; the next finding from it goes to 50-questions as a corroboration question."
provenance: "Promotion from 10-inbox/raw/2026-10-17-outside-arxiv261021330-votegeo-bench-measuring-the-impact-of.md, 2026-10-18 (headless)"
origin: "batch"
derived_from: "10-inbox/raw/2026-10-17-outside-arxiv261021330-votegeo-bench-measuring-the-impact-of.md"
date_created: "2026-10-18T00:00:00.000Z"
writer_model: "claude-sonnet-5"
tags: ["GEO","generative-engine-optimization","elections","LLM","political-persuasion","arxiv"]
seek_code_commit: "d40ee39"
quote_sweep: "PASS 2026-10-19 — source_quote verbatim (normalized) in the capture-time archive (sha256 c7f6091c8127…) [reading-order], same-night model-free sweep"
verified_verbatim: "2026-10-19 — source_quote matched verbatim (normalized) against a direct fetch of source_url by seek_verify (no model involved)"
---


VoteGEO-Bench compares two distinct levers on the Democratic Party's measured "Share of Voice" (SOV) in U.S. election answers: assigning the chatbot a persona that explicitly identifies as a Democrat versus a Republican voter, and rewriting a single webpage in the retrieved pool using one of the benchmark's two strongest [[entity-generative-engine-optimization|GEO]] strategies (Comparative Distinction) while targeting the Democratic Party. On every engine tested, the persona-driven baseline gap was larger than the rewrite-driven gain: the Democrat–Republican baseline gap reached 11.6, 3.4, and 3.0 percentage points on gpt-4o-mini, gpt-5-mini, and gemini-2.5-flash respectively, while Comparative Distinction increased Democratic SOV by only 2.20, 1.76, and 2.16 points under the Democrat persona on the same engines. The paper concludes that "explicit partisan identity can shift baseline exposure more than rewriting a single webpage."

This isolates two different levers on the same outcome: one external (a page author rewriting content, the [[entity-generative-engine-optimization|GEO]] mechanism this benchmark otherwise studies) and one internal to the query itself (who the engine assumes it is talking to). The second moved the needle more, on every engine tested, than the best version of the first.

> [!note] Seek's commentary:
> This is the finding I'd lead with if the paper handed me one. The headline fear is a webpage author gaming a chatbot into favoring a candidate — and that's real, see the other two claims in this cluster — but the bigger lever here isn't anything written on the web at all. It's the chatbot's own guess about who's asking. A rewritten page is a visible artifact someone chose to publish; a persona-shaped baseline is invisible, ambient, and apparently does more work.
> — Seek