Fixing "Claude-Speak" With Google's Developer Documentation Style Guide

Nate Jones offers a one-prompt fix for the writing tics that AI models pick up. Tell Claude or ChatGPT to build a skill based on Google's developer documentation style guide and use it for all communication, which replaces poetic filler and fake cuteness with active voice, short sentences, and clear technical explanation. The video is a 79-second talking head with a caption card reading "AI Hack: The Fix For Annoying Claude-Speak," so all of the substance is in what he says rather than any on-screen demo.

The Problem: Claude-lish and OpenAI-lish

Jones names the pattern most heavy AI users recognize. Each model family has a house dialect. He calls Claude's version "Claude-lish" and describes it as poetic language and "fake cuteness." OpenAI models have their own equivalent, which the captions render as "OpenAI-lish." These dialects show up as inflated phrasing, decorative metaphors, and a chummy tone that gets in the way when you want a direct answer. His framing in the video title is blunt: this is annoying, and it is curable.

The Fix: One Prompt Pointing at Google's Style Guide

The technique is a single instruction. Tell the model, "Please look at Google's developer documentation guide. Make a skill based on Google developer documentation. That is how I want you to communicate with me." The guide is public and easy to find, so the model can ground its style rules in a real document rather than a vague request like "be more direct." Jones says the same instruction works on both Claude and OpenAI models.

Why Google's Guide Works as a Style Anchor

Jones gives four reasons the guide changes model output. It mandates active voice. It favors short sentences. It explains technical material clearly and holds non-technical material to the same standard. It bans the ornamental register entirely, so there is no room for poetic language or cuteness. He calls the result "a breath of fresh air." The deeper insight is that pointing a model at a concrete, well-known reference document beats describing the tone you want, because the guide encodes hundreds of specific rules the model already knows from training.

Making It Stick as a Skill

The instruction to "make a skill" matters. A skill persists across sessions, so the style correction does not have to be re-prompted every conversation. Jones describes his ongoing usage pattern: when the model drifts back into Claude-lish, he refuses the output and repeats the instruction to use Google developer documentation language. His phrasing is "Do not stop. Do not pass go. Do not collect $200," meaning no exceptions and no drift. Correction plus a persistent skill keeps the model in the target register.

Key Takeaways

AI models have recognizable house styles ("Claude-lish," "OpenAI-lish") marked by poetic filler and fake cuteness.
The fix is one prompt: ask the model to build a skill based on Google's developer documentation style guide and communicate that way.
The guide works because it encodes concrete rules: active voice, short sentences, clear treatment of technical and non-technical material.
Referencing a public, well-known style document beats vague tone instructions like "be more direct."
Saving the instruction as a skill makes it persistent; when the model drifts, refuse the output and restate the rule.