Part 1: Introduction (Bakhtin and model collapse: How to use AI with expressive writing without generating AI slop)

by Tom Johnson on Jan 30, 2026 comments
categories: ai • ai-book-club • writing

• Part 1: Introduction
• Writing categories: expressive vs encyclopedic
• Next section

Part 1: Introduction

It seems like every time someone talks about AI and writing, it’s usually to slam AI slop and to say how much of a turnoff AI writing is. I agree that AI slop is terrible to read. It’s generic, void of unique voice, lacks first-person perspective and experience, and doesn’t say much at all. I also dislike reading it.

The problem, though, is that assuming all AI-assisted writing results in AI slop seems dismissive of the potential of AI. Is it possible to somehow use AI with more creative genres of writing in a way that doesn’t result in slop? In this post, I’ll explore ways to use AI to level up the quality of expressive writing without resulting in garbage output, and without robbing writers of the value of the writing process itself.

In brief, I argue that writers can use AI as a research assistant to add additional voices and perspectives into their writing; this contribution adds a heteroglossic quality that actually brings the content to life and deepens its meaning. AI can also be used to properly verify that you’re engaging with the authors in your essay fairly, representing their views correctly. In this post, I dive deeply into Bakhtin for my critical lens.

Because of the length of this post, I’ve split it into multiple pages, which you can see in the sidebar. Links to the next page appear at the bottom of each page.

Writing categories: expressive vs encyclopedic

First, let’s acknowledge that “writing” is such a broad category that applying blanket statements about AI and writing without qualifying what type of writing we’re talking about is generally misguided. This is a problem I encountered in Jonathan Warner’s More Than Words, which is otherwise an excellent book. Had Warner clarified the type of writing he was analyzing in the context of AI, it would have been easier for me to support his arguments.

In the corporate enterprise, a lot of writing happens, but very little of it is expressive. Few people care how AI-assisted the enterprise content is, as long as it’s accurate and clear. For example, take technical documentation for APIs. In this genre, you’re describing an API field, writing code, code comments, or internal procedures. By and large, when you have a scenario that’s highly technical and impersonal, with no expectations about arguments and perspectives, AI does a good job at the writing without slipping into soulless superlatives and flowery sentences.

In fact, in many cases, AI can help improve the accuracy and factuality of the information, acting as a verifier based on contextual data you provide. I routinely add API reference docs as context to my AI sessions when I want to verify release notes and other updates, for example. AI is generally welcome in enterprise writing scenarios, particularly with technical documentation, where the author is invisible, and where accuracy and clarity are more important than style.

However, in expressive genres of writing, such as personal essays, opinion articles, argumentative essays, blog posts, and more, AI is much less welcome. This is where using AI is met with accusations about AI slop, and rightly so.

My question is whether AI could also be helpful in more expressive genres, without becoming AI slop that everyone hates, including those producing it. I do think there’s a possible use case for using AI without losing your own voice, which seems to be the main concern. The strategy involves using AI as a research assistant to deepen your writing with a Bakhtinian “heteroglossia” strategy, which I’ll both explain and explore at length. In fact, the liveliness of the writing might emerge from this juxtaposition of multiple voices.

Next section

Continue on to the next section, Part 2: Heteroglossia.

Comment on LinkedIn