Vibe Writing

A friend said something to me recently that I haven’t been able to shake.

I was drafting a description of my research project to apply for computational resources. It was one of those dense, technical summaries that require precision, framing, and just the right tone. He glanced over and said, casually, “Why don’t you just outsource it to ChatGPT? Like everybody else does.” He called it vibe writing. That phrase stuck with me.

Outsourcing writing to an AI is not the same as asking for feedback or using a spell checker. It represents a shift in authorship. Instead of wrestling with your own sentences, you supply a rough idea and delegate articulation to a language model. The model returns something fluent, structured, and confident. You tweak a few words. It sounds good. You submit.

The result is efficient. But what exactly have you outsourced?

Much of the ethical conversation around AI companions focuses on emotional outsourcing: people turning to AI for validation and simulated intimacy without the vulnerability and reciprocity that human relationships require. I think there is a parallel risk in such an outsourcing. When we repeatedly rely on a language model to “sound articulate,” we begin to doubt our own ability to do so.

Writing is not merely a channel of communication; it is a tool for organizing one’s own thoughts. The struggle to find the right phrase is often the moment when we discover what we actually believe. When that struggle is bypassed, something subtle changes: we still produce language, but we are no longer developing the capability to generate it ourselves.

Over time, this affects confidence. If every important email, statement, or reflection passes through an “AI filter” before it feels acceptable, we start to internalize a new standard: unassisted writing equals inadequate writing. The model becomes the definition for clarity, and our own drafts feel clumsy by comparison.

That comparison matters. If I begin to believe that my unmediated articulation is inferior, I will hesitate even more in conversations, defer more quickly in debates, or rely on pre-formulated phrasing instead of developing my own.

Social capability depends on spontaneous articulation. In real-time dialogue, there is no “regenerate” button. You cannot pause a difficult conversation to optimize tone and structure. If we grow accustomed to AI-assisted expression, ordinary human speech, with its hesitations, revisions, and imperfections, may start to feel uncomfortable. We may speak less freely because we are no longer practiced at thinking out loud.

This is not an argument against using language models. They can be powerful scaffolding tools, helping structure ideas, suggest alternative phrasings, or identify logical gaps. The key distinction lies in whether the tool supplements or replaces our cognitive effort.

“Vibe writing” sounds harmless. Yet behind that convenience lies a deeper question: are we using language models to strengthen our voice, or are we slowly outsourcing it?

(Vibe written with ChatGPT 5.2)