# Too Good to Be Real? Diagnosing and Reducing the Gap Between AI Preference and Real User Engagement

arXiv: https://arxiv.org/abs/2609.18282 · HTML: https://mlsanity.com/paper/2609.18282/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 28 of its 46 references that are in the map.
- First posted on arXiv 16 Sep 2026.

## Builds on (most cited first, 28 of 28)

- [Training language models to follow instructions with human feedback](https://mlsanity.com/paper/2203.02155/) (2022)
- [LLaMA: Open and Efficient Foundation Language Models](https://mlsanity.com/paper/2302.13971/) (2023)
- [Direct Preference Optimization: Your Language Model is Secretly a Reward Model](https://mlsanity.com/paper/2305.18290/) (2023)
- [Deep reinforcement learning from human preferences](https://mlsanity.com/paper/1706.03741/) (2017)
- [From Crowdsourced Data to High-Quality Benchmarks: Arena-Hard and BenchBuilder Pipeline](https://mlsanity.com/paper/2406.11939/) (2024)
- [Self-Preference Bias in LLM-as-a-Judge](https://mlsanity.com/paper/2410.21819/) (2024)
- [Preference Leakage: A Contamination Problem in LLM-as-a-judge](https://mlsanity.com/paper/2502.01534/) (2025)
- [Winning Arguments: Interaction Dynamics and Persuasion Strategies in Good-faith Online Discussions](https://mlsanity.com/paper/1602.01103/) (2016)
- [Parsing Argumentation Structures in Persuasive Essays](https://mlsanity.com/paper/1604.07370/) (2016)
- [Eliminating Position Bias of Language Models: A Mechanistic Approach](https://mlsanity.com/paper/2407.01100/) (2024)
- [Self-Boosting Large Language Models with Synthetic Preference Data](https://mlsanity.com/paper/2410.06961/) (2024)
- [AMPERSAND: Argument Mining for PERSuAsive oNline Discussions](https://mlsanity.com/paper/2004.14677/) (2020)
- [Semantic-Aware Logical Reasoning via a Semiotic Framework](https://mlsanity.com/paper/2509.24765/) (2025)
- [Logical Phase Transitions: Understanding Collapse in LLM Logical Reasoning](https://mlsanity.com/paper/2601.02902/) (2026)
- [LoRA-Mixer: Coordinate Modular LoRA Experts Through Serial Attention Routing](https://mlsanity.com/paper/2507.00029/) (2025)
- [Coupling Macro Dynamics and Micro States for Long-Horizon Social Simulation](https://mlsanity.com/paper/2604.05516/) (2026)
- [MVP: Winning Solution to SMP Challenge 2025 Video Track](https://mlsanity.com/paper/2507.00950/) (2025)
- [IntervenSim: Intervention-Aware Social Network Simulation for Opinion Dynamics](https://mlsanity.com/paper/2604.06600/) (2026)
- [Learning to Substitute Words with Model-based Score Ranking](https://mlsanity.com/paper/2502.05933/) (2025)
- [SimToken: A Simple Baseline for Referring Audio-Visual Segmentation](https://mlsanity.com/paper/2509.17537/) (2025)
- [GODBench: A Benchmark for Multimodal Large Language Models in Video Comment Art](https://mlsanity.com/paper/2505.11436/) (2025)
- [Seeing Further and Wider: Joint Spatio-Temporal Enlargement for Micro-Video Popularity Prediction](https://mlsanity.com/paper/2604.20311/) (2026)
- [Learning to Control Summaries with Score Ranking](https://mlsanity.com/paper/2604.17197/) (2026)
- [Calibrating Model-Based Evaluation Metrics for Summarization](https://mlsanity.com/paper/2604.17200/) (2026)
- [HotComment: A Benchmark for Evaluating Popularity of Online Comments](https://mlsanity.com/paper/2604.25614/) (2026)
- [ExoActor: Exocentric Video Generation as Generalizable Interactive Humanoid Control](https://mlsanity.com/paper/2604.27711/) (2026)
- [ViralBERT: A User Focused BERT-Based Approach to Virality Prediction](https://mlsanity.com/paper/2206.10298/) (2022)
- [Interpretable Stylistic Variation in Human and LLM Writing Across Genres, Models, and Decoding Strategies](https://mlsanity.com/paper/2604.14111/) (2026)

## Abstract (arXiv)

Large language models are increasingly used to generate and evaluate online content, yet it remains unclear whether the qualities they associate with higher engagement match what real users respond to. We study this question using 1.17 million answers to 25,978 questions from Zhihu, Quora, and Reddit, comparing real platform answers and AI-generated answers across four within-question engagement levels. We introduce Ontological Preference Measurement, which represents answers along three dimensions: logic, affect, and expression. We find a systematic gap between AI preference and real user engagement: as target engagement increases, LLMs add more explicit logical structure, while real user engagement is more strongly associated with affective and expressive salience. We call this tendency logic overbinding. Based on this diagnosis, we propose Ontology-Masked Reasoning Autoencoding (OMRA), a controlled intervention that masks and reconstructs over-explained spans while preserving stance, factual content, and coherence. Across four LLM families, OMRA reduces the measured gap by an average of 54.4%. In human evaluation, OMRA wins 62.4% of pairwise preference judgments against matched real platform answers, even though the real answers are more often judged to be human-written.