Skip to main content
Onur AksoyOnur Aksoy — Homepage

Negative Belief Shifts after LLM Arguments and the Limits of Lexical Cues

Reanalysis of Anthropic persuasion (3,939 ratings of 1,313 arguments) and GEO-bench (4,996 Google pages): LLM texts match human average shifts yet trigger 221/1,120 backfires that word counts fail to predict.

#BeliefChange#LLM#Persuasion#LexicalCues#EquivalenceTesting#GEO#OpenScience
Negative Belief Shifts after LLM Arguments and the Limits of Lexical Cues

Source Files & Production Assets

Directly downloadable source files and 20TB high-speed vault archive

20TB Storage Vault
PDF DOCUMENT438.3 KBSHA: sha256:426...

Preprints.org Open Access Preprint Manuscript (PDF)

Direct Download ⤓
SOURCE CODEGitHub Repo

Open Analysis Code and Reproducibility Repository (GitHub)

Open Drive ↗

Introduction: The Blind Spot Behind Average Persuasion Scores

Evaluations comparing human and large language model (LLM) persuasion routinely summarize performance with a single number: mean attitude change. At first glance, average shifts suggest functional parity. Yet means conceal opposing movements within the same distribution. If an AI-generated argument raises aggregate support while pushing nearly one in five readers further away from the advocated stance, summary averages hide an operational liability.

Published as an open-access preprint under DOI 10.20944/preprints202611.1914.v1 on Preprints.org, this study reanalyses two public benchmarks from first principles: Anthropic/persuasion and GEO-bench. Extending the empirical rigor established in our analyses of entity reconciliation and synthetic technical debt with deterministic gates led by Onur Aksoy (Hackonomist), this work tests both argument-level belief distributions and the predictive limits of surface-level lexical heuristics.

flowchart LR
    subgraph Inputs["1. Public Benchmark Corpora"]
        A1["Anthropic/persuasion: 1,313 Arguments, 3,939 Ratings, 56 Claims"]
        G1["GEO-bench: 4,996 Google Top-Five Pages, 1,000 Queries"]
    end

    subgraph Methods["2. Argument and Query-Level Statistical Pipeline"]
        A2["TOST Equivalence Testing and Argument-Level Backfire Rates"]
        A3["Claim-Stratified CMH Test and GEE Regression"]
        G2["Within-Query Friedman Rank Tests and Intent Mapping"]
        A1 --> A2 --> A3
        G1 --> G2
    end

    subgraph Findings["3. Empirical Outcomes"]
        R1["Claude 221/1,120 Negative Shifts vs Human 1/174"]
        R2["Unseen Claims TF-IDF Out-of-Fold R2 <= 0.003"]
        R3["GEO-bench Rank Variation Weak, Intent Dominates"]
        A3 --> R1
        A3 --> R2
        G2 --> R3
    end

Argument-Level Reanalysis: Equivalence Margins and Backfire Asymmetry

The Anthropic/persuasion corpus contains 1,313 arguments across 56 policy claims written either by humans or by five Claude models. Because each argument received three independent ratings (3,939 ratings total), aggregating and modeling outcomes at the argument level removes repeated-rating dependence.

At the mean level, Claude 3 Opus arguments shifted reader support by 0.48 Likert points compared with 0.54 points for human arguments. The difference is not statistically significant (p = 0.37). However, formal equivalence testing using two one-sided tests (TOST) places a strict boundary on this claim: equivalence holds only at a wide ±0.20 Likert margin (p = 0.010), and fails at the tighter ±0.15 threshold (p = 0.066).

Examining the tails of the distribution uncovers a stark divergence. Out of 1,120 Claude arguments, 221 (19.7%) shifted at least one reader away from the advocated position. Across 174 human-written arguments, that opposite-direction shift occurred only once (1/174, or 0.6%; p = 3e-14).

This asymmetry cannot be attributed to cherry-picked claims, a single model release, or skewed baseline beliefs. A claim-stratified Cochran-Mantel-Haenszel (CMH) test confirms higher negative-shift rates for LLM text in 54 out of 56 claims (CMH p = 6e-10), holding under matched starting attitudes and across all five Claude variants.

Dimension Claude Arguments (LLM) Human Arguments Statistical Test & Result
Mean Belief Shift (Claude 3 Opus) 0.48 Likert points 0.54 Likert points Difference non-significant (p = 0.37)
TOST Equivalence Margin Equivalent at ±0.20 (p = 0.010) Reference baseline Not supported at ±0.15 (p = 0.066)
Negative Belief Shift (Backfire) 221 / 1,120 (19.7%) 1 / 174 (0.6%) Argument-level disparity (p = 3e-14)
Cross-Claim Consistency Negative shifts in 54 / 56 claims Negative shifts in 1 / 56 claims Stratified CMH (p = 6e-10)
Out-of-Sample Lexical Prediction Fixed word lists & full TF-IDF Shared feature space out-of-fold R² <= 0.003

Why Lexical Heuristics Fail to Predict Persuasion or Rank

Both persuasion playbooks and generative engine optimization (GEO) guidelines frequently prescribe inserting surface-level lexical markers: quantitative statistics, research terminology, certainty modifiers, and inclusive collective-reference words. Testing these cues directly against both benchmarks reveals three structural limits:

  1. Zero predictive power for backfire: Under generalized estimating equations (GEE), none of the recommended lexical categories predicted negative belief shifts in LLM-authored arguments.
  2. Confounder collapse: Weak bivariate correlations with average attitude change largely disappeared once prompt type and initial reader attitude were included as controls.
  3. Full bag-of-words ceiling: To rule out blind spots in fixed dictionaries, a full TF-IDF bag-of-words model was evaluated via cross-validation on unseen claims. Neither curated word lists nor the full lexical space predicted attitude change (out-of-fold R² <= 0.003).

The same pattern holds in search visibility. Analyzing 4,996 Google top-five result pages across 1,000 queries in GEO-bench using within-query Friedman rank tests shows minimal variation by search position. Lexical feature counts track query intent rather than ranking advantage, with only citation markers occurring slightly more frequently at rank 1 than at rank 5 (q = 0.014).


Architectural Implications for Production LLM Systems

Sprinkling statistical terms or assertive phrasing over generated prose neither guarantees reader persuasion nor secures top search placement. More importantly, selecting language models for high-stakes communication, advisory workflows, or public-facing messaging solely on average score parity introduces silent failure modes across nearly one in five arguments.

All analysis code, curated word lists, plotting scripts, and computed JSON outputs are publicly archived in the open reproducibility repository on GitHub. Both the full preprint PDF and source repository links are available in the download vault above.

FAQ & Technical Deep Dive

Frequently Asked Technical Questions

Why can LLM arguments carry higher risk even when average persuasion scores match human writers?↓

Mean persuasion scores mask opposing movements that cancel out across readers. Claude 3 Opus shifted support by 0.48 Likert points versus 0.54 for humans (p = 0.37), yet 221 of 1,120 Claude arguments (19.7%) moved at least one reader away from the advocated stance compared with 1 of 174 human arguments (0.6%).

Are opposite-direction belief shifts limited to one model variant or a few polarizing topics?↓

No. The backfire gap appeared in 54 of the 56 policy claims (claim-stratified Cochran-Mantel-Haenszel p = 6e-10). Matched initial attitudes and comparable negative-shift rates held across all five Claude models in the benchmark.

Do word-level cues like statistics, certainty, or inclusive terms explain persuasion or backfire?↓

They do not. None of the lexical cues predicted negative shifts in generalized estimating equation (GEE) models, and associations with mean attitude change largely vanished once prompt type and initial attitude were controlled. Neither fixed word lists nor a full TF-IDF bag-of-words model predicted change on unseen claims (out-of-fold R² <= 0.003).

How do lexical cues behave across Google top-five rankings in GEO-bench?↓

Within-query Friedman analysis across 4,996 Google top-five pages for 1,000 queries shows minimal variation by rank. Lexical density primarily reflects query intent rather than search position, with only citation markers appearing more often at rank 1 than at rank 5 (q = 0.014).

Onur Aksoy (a.k.a. Hackonomist)
Author & Systems Architect

Onur Aksoy (Hackonomist)

Ankara-based Software Developer, Digital Strategist & Systems Architect. Specialist in deterministic state machines, local-first distributed topologies, low-level Rust network sniffers, and zero-server media engines.

Enterprise Architecture Solutions
Ankara & Global Systems Engineering