The Last Fingerprint: How Markdown Training Shapes LLM Prose

TL;DR

This paper investigates how Markdown training influences large language models' use of em dashes, revealing that em dash frequency is a signature of training data and fine-tuning procedures.

Contribution

It establishes a mechanistic link between Markdown training, structural internalization, and em dash usage, providing a diagnostic tool for fine-tuning effects.

Findings

Em dash frequency varies widely across models, from 0.0 to 9.1 per 1,000 words.

Models instructed to avoid Markdown still produce em dashes, indicating a latent tendency.

Meta's Llama models do not produce em dashes, showing differences in fine-tuning.

Abstract

Large language models produce em dashes at varying rates, and the observation that some models "overuse" them has become one of the most widely discussed markers of AI-generated text. Yet no mechanistic account of this pattern exists, and the parallel observation that LLMs default to markdown-formatted output has never been connected to it. We propose that the em dash is markdown leaking into prose -- the smallest surviving unit of the structural orientation that LLMs acquire from markdown-saturated training corpora. We present a five-step genealogy connecting training data composition, structural internalization, the dual-register status of the em dash, and post-training amplification. We test this with a two-condition suppression experiment across twelve models from five providers (Anthropic, OpenAI, Meta, Google, DeepSeek): when models are instructed to avoid markdown formatting, overt features (headers, bullets, bold) are eliminated or nearly eliminated, but em dashes persist -- except in Meta's Llama models, which produce none at all. Em dash frequency and suppression resistance vary from 0.0 per 1,000 words (Llama) to 9.1 (GPT-4.1 under suppression), functioning as a signature of the specific fine-tuning procedure applied. A three-condition suppression gradient shows that even explicit em dash prohibition fails to eliminate the artifact in some models, and a base-vs-instruct comparison confirms that the latent tendency exists pre-RLHF. These findings connect two previously isolated online discourses and reframe em dash frequency as a diagnostic of fine-tuning methodology rather than a stylistic defect.