How to Remove 'AI Slop' and Refine Your Model's Writing Style

Modern AI models are undeniably powerful, but they often suffer from a recognizable, formulaic style—characterized by excessive lists, repetitive phrases, and a stiff, robotic tone. While these "LLMisms" can make AI-generated content feel like filler, there are several effective strategies to improve the quality of your output.

Shift Your Workflow

The most effective way to improve AI writing is to stop treating it as a primary generator and start utilizing it as an assistant. Instead of asking a model to write an entire draft, task it with research, outlining, or drafting individual sections. By breaking the work into smaller components, you can more easily curate the best parts and maintain a consistent, human voice throughout the piece.

Master the Art of Prompting

If you rely on models for drafting, you must move beyond generic instructions. Here is how to tighten up your prompts:

Provide Style Examples: Models follow examples much better than they follow abstract instructions. Instead of asking for a specific tone, provide several paragraphs of your own writing (or writing you admire) and instruct the model to "write in this style."
Explicitly Ban Tropes: Create a custom "style manual" that lists specific behaviors you want to avoid—such as contrastive negation or excessive listing—and ask the model to adhere to these constraints.
Encourage Self-Evaluation: After the AI produces a draft, ask it to critique its own work based on your stylistic requirements. Instruct it to identify and rewrite sections that sound like "AI slop."

Curate a Style Guide

Building a personal repository of writing tropes to avoid is highly recommended. By formally labeling patterns you dislike—like the "Westworld host" cadence or repetitive sentence structures—you create a reference that can be fed into long-context prompts. Over time, building this library ensures that your AI assistant learns exactly what "usability" means for your specific projects.

Ultimately, the best AI output comes from a tight feedback loop. By combining clear examples, explicit prohibitions on stylistic flaws, and a human-in-the-loop editing process, you can move past the robotic default and generate content that reads as if it were crafted by a person.