ROIpad ← Back to Search

Insight for: Show HN: We fingerprinted 178 AI models' writing styles and similarity clusters

A dataset and analysis of 178 AI models' writing styles, identifying similarity clusters and distinctiveness based on 3,095 standardized AI responses.

Analyzed: Apr 9, 2026

This analysis provides quantitative insights into AI model stylistic differentiation and convergence. Identifying 'clone clusters' with high cosine similarity highlights potential commoditization or lack of unique voice among certain models. The finding that Gemini 2.5 Flash Lite writes 78% like Claude 3 Opus at 185x less cost presents a significant cost-optimization opportunity for businesses prioritizing stylistic similarity over other model attributes. Meta's 'strongest provider house style' indicates brand-specific stylistic consistency, which could be a differentiator. The impact of specific prompts on writing convergence ('satirical fake news') and divergence ('count letters') offers valuable data for prompt engineering and model evaluation. This research informs strategic model selection, cost management, and understanding the inherent stylistic biases and capabilities of various LLMs, critical for applications requiring specific tone or avoiding detection.

stylometric fingerprint lexical richness sentence structure punctuation habits formatting patterns discourse markers clone clusters cosine similarity z-normalized feature vectors composite metric independent signals distinctiveness ratio writing convergence writing divergence Pearson correlation response length correlation cross-prompt consistency aggregate cosine similarity Node.js z-score normalization

Hacker News Post

Parent Entity

Show HN: We fingerprinted 178 AI models' writing styles and similarity clusters

Score: 52