What is Formulaic Structure?
The formulaic structure in AI-generated content refers to the tendency of large language models (LLMs) to produce text that adheres strictly to predictable, repetitive, and conventional linguistic patterns, making the writing sound robotic, unoriginal, and lacking in unique human style or creative variation.
This issue stems directly from how LLMs are trained and how they generate text.
The Root Cause: Statistical Prediction
LLMs work by calculating the statistical probability of the next most likely word or phrase based on the massive datasets they were trained on. When given a general prompt, the AI defaults to the most common and predictable patterns it has absorbed, which leads to formulaic outputs at two levels:
1. Macro-Level (Structural Predictability)
This relates to the overall organization of a piece of writing:
- The “Middle School Essay” Format: AI frequently organizes text with a highly structured, almost rigid format—a clear Introduction, followed by exactly three to four Body Paragraphs (each starting with a clear topic sentence), and a neat, summary Conclusion. This is the most common, safe structure found in its training data.
- Repetitive Transitions: An over-reliance on a small set of conventional transition words to link ideas, making the flow predictable.
- Examples: “In addition,” “Furthermore,” “Moreover,” “However,” “In conclusion,” “At its core.”
- Predictable Topic Development: For a given topic (e.g., “The importance of digital marketing”), the AI often follows the most common, anticipated points in a fixed order, failing to introduce an unexpected or novel perspective.
2. Micro-Level (Sentence Predictability)
This relates to the specific construction and phrasing of individual sentences:
- Correlative Conjunction Overuse: Repetitive use of balanced, yet rigid sentence structures.
- Example: “Not only is the design sleek, but it is also highly functional.”
- Example: “It is neither complex nor expensive to implement.”
- Emphatic Contrast Formulas: Using specific contrastive phrases that become repetitive.
- Example: “It wasn’t just the product’s features, but the entire user experience that stood out.”
- Uniform Sentence Structure: A lack of burstiness, where sentences tend to be of very similar lengths and grammatical complexity, giving the prose an even, monotonous rhythm.
- Cliché and Buzzword Reliance: Choosing common clichés or generic industry jargon (a “buzzword salad”) instead of specific, fresh, or authentic language (e.g., using “unleash potential” or “drive synergistic results”).
Implications for Verification and Quality
- AI Detection: Formulaic structure—especially the low variation in sentence complexity and the high predictability of the next word (low perplexity)—is one of the primary features that AI detectors look for when flagging content as machine-generated.
- Human Readability: While the text is grammatically correct and fluent, its formulaic nature makes it sound sterile, generic, and unengaging to a human reader, often stripping the content of personality, emotional nuance, or genuine human insight.