The Typographic Advantage: Why AI-Generated Content Demands Better Formatting

The Commodity Text Problem

AI-generated text arrives as undifferentiated plaintext. It has no typographic hierarchy, no spatial intelligence, no awareness of the physical or digital medium in which it will be read. A 300-page novel generated by a language model is, at the point of output, indistinguishable in its formatting from any other 300-page novel generated by the same or a competing model. The Markdown may be syntactically correct — headings marked with hashes, emphasis with asterisks — but syntactic correctness is not typographic design.

This is the commodity text problem. When production cost approaches zero, the supply of competent-but-undifferentiated text expands toward infinity. The reader, faced with an abundance of adequately written books, falls back on secondary signals to determine which ones deserve attention: the cover, the interior layout, the apparent care invested in the physical (or digital) object. Robert Bringhurst described typography's purpose as giving language "a durable visual form" (The Elements of Typographic Style, 4th edition, 2012). That durable form is precisely what AI-generated content lacks.

Presentation as a Trust Signal

The 2012 experiment by Errol Morris and David Dunning, published via the New York Times, demonstrated that typeface selection measurably affects the perceived credibility of statements. Identical claims presented in Baskerville were rated as more believable than the same claims presented in Helvetica, Comic Sans, or Georgia. The effect was modest in magnitude — approximately 1.5 percentage points — but statistically significant and consistent across a sample of over 45,000 respondents.

The Morris-Dunning finding has a specific implication for AI-generated content. If the text itself cannot be distinguished from human-written prose (and detection tools have proven unreliable), the reader's trust assessment defaults to extrinsic signals: publisher reputation, author platform, and — critically — the production quality of the artefact. A book with professional typesetting, correct leading, appropriate margins, and a well-designed title page signals that someone invested care in the object. That investment is a trust proxy. As explored in "The Architecture of Trust," the visual form of a document communicates credibility independently of its verbal content.

Colin Wheildon's comprehension research, discussed extensively in "The ROI of Legibility," adds a functional dimension: well-typeset text is not merely more credible, it is more comprehensible. Good comprehension rates rose from 12 per cent to 67 per cent when body text was set in a well-chosen serif face with appropriate leading. For AI-generated content competing in a saturated market, the difference between a reader who comprehends and one who abandons is the difference between a review and a refund.

What AI Cannot Do: Genre-Aware Spatial Intelligence

Language models operate on sequences of tokens. They have no concept of a page, a spread, a gutter margin, or the physical curvature of a bound spine. They cannot determine that a poetry collection requires preserved line breaks and generous vertical white space, that a cookbook needs ingredient blocks distinguished from method paragraphs, or that an academic monograph requires footnotes anchored to the baseline grid. These are spatial and genre-specific decisions that require knowledge of the output medium — print dimensions, binding method, paper stock, reading distance — that exists entirely outside the model's training distribution.

This gap is not a temporary limitation that will be solved by larger models. It is a category error. Typography is a design discipline concerned with the relationship between text, space, and the physical or optical properties of a substrate. A language model that produces Markdown has performed the textual work; the typographic work — choosing a typeface, calculating margins, setting leading, constructing a baseline grid, placing running heads — remains entirely unaddressed. As "The False Economy of the Software Default" argues, the assumption that default settings will handle this work is precisely the false economy that produces amateur-looking books.

The Formatting Gap in Practice

Examine the typical AI-assisted publishing workflow. The author generates or refines a manuscript using a language model, exports it as a Word document or Markdown file, and uploads it to a self-publishing platform. Amazon KDP applies its own automated formatting — a process that handles pagination and basic font embedding but makes no genre-specific typographic decisions. The result is a book that is technically readable but typographically inert: default margins, default leading, no baseline grid, no considered hierarchy, no relationship between the typeface and the genre's conventions.

The reader may not consciously identify what is wrong. But the cumulative effect of dozens of micro-failures — leading that is too tight, margins that are too narrow, chapter openings that lack visual ceremony, running heads that crowd the text block — registers as a vague sense that the book feels cheap. This perception is not irrational. It is an accurate reading of the production investment. The book was produced cheaply, and it looks like it. In a market where AI has equalised the cost of generating text, the books that succeed will be those where the author — or the author's tools — invested in the presentation that AI cannot provide.