Em Dashes, AI Writing, and the Quiet Typography Debate in Digital Blogging

While the rise of generative AI has dominated headlines with concerns over plagiarism, originality, and authorship, a quiet revolution in digital typography may offer a more nuanced lens through which to evaluate machine-generated content. Simon Willison, a respected software developer and blogger, recently reflected on a decade-old Python script that automatically converts hyphenated spaces into proper em dashes (—) in his blog posts. This seemingly trivial line of code — s = s.replace(' - ', u'\u2014') — has become an accidental litmus test for human authenticity in an age where AI writing is often detected through stylistic fingerprints.

Willison, whose blog has been a staple of the developer community since the early 2000s, acknowledges that while he does not use large language models to generate content, his readers sometimes suspect otherwise. The em dash script, unchanged since at least 2015, is the one element he admits might carry an "LLM smell." Why? Because automated typographic corrections — particularly the consistent, almost obsessive use of em dashes over hyphens — are a hallmark of AI-generated text trained on meticulously edited corpora. Human writers, by contrast, often use hyphens, en dashes, or even spaced em dashes inconsistently, depending on mood, platform, or haste.

This observation intersects with broader digital publishing trends. As content creators increasingly rely on tools like Grammarly, Hemingway, or AI editors to polish prose, the line between human intent and algorithmic imposition blurs. According to research on digital typography in web publishing, consistent use of typographic elements such as em dashes, curly quotes, and non-breaking spaces correlates strongly with professionally edited content — and increasingly, with AI-generated output. While these features enhance readability, their uniformity can ironically signal artificiality in contexts where organic variation is expected.

Interestingly, the confusion around the term "zh-Hans" in the source materials — which incorrectly appear to be pulled from unrelated Chinese-language health articles on Mayo Clinic’s website — underscores a deeper issue: the proliferation of misaligned or scraped content across the web. The presence of Chinese-language metadata and unrelated知乎 (Zhihu) snippets in English-language reports suggests either automated content aggregation errors or attempts to manipulate SEO through multilingual keyword stuffing. This further complicates the task of distinguishing authentic, human-authored content from algorithmically assembled or AI-enhanced material.

Willison’s case is not about whether em dashes are "right" or "wrong," but about the cultural and technical assumptions we embed in our tools. His script, written in Django’s templating system, was a pragmatic solution to a real problem: inconsistent formatting across platforms. Yet today, it functions as a digital artifact — a timestamped signature of human curation in a world increasingly shaped by automation. As AI writing tools become more sophisticated, subtle stylistic choices like punctuation may become the new watermark — not for detection, but for recognition.

For journalists and bloggers alike, the takeaway is clear: authenticity is not just about content origin, but about the invisible labor behind the text. A perfectly placed em dash might be the result of a five-line Python script — or of a writer’s lifelong habit of polishing prose. The difference matters. In an era where every comma could be AI-generated, the most human thing a writer can do may be to leave a few imperfections — or to proudly maintain a 10-year-old script that refuses to change.

As the debate over AI and authorship evolves, typographic consistency may become the new frontier of digital forensics — not to catch fraud, but to honor the quiet, deliberate choices that still distinguish human voice from machine mimicry.