How to Make AI Content Sound Like You, According to Dan Sherrard-Smith

In a recent LinkedIn post, Dan Sherrard-Smith addresses the pervasive issue of generic AI-generated content, particularly on platforms like LinkedIn. He argues that much of the AI output sounds like “AI slop,” contributing to a “boring, vanilla” online environment. Sherrard-Smith then shares a detailed method for training the AI model Claude to adopt a user’s unique writing voice, aiming to combat this trend and help individuals grow their personal brand.

Sherrard-Smith emphasizes the problem with standard AI prompts:

Most founders type “write a LinkedIn post about X” into a blank chat. Claude spits out something polished. But it’s vanilla, forgettable and boring.

To overcome this, Sherrard-Smith proposes a two-step setup using the Claude desktop application and its “Cowork” feature. This approach involves creating a dedicated folder on one’s computer to store custom instructions for the AI.

Building a Personal Voice Profile for AI

The first crucial step, according to Sherrard-Smith, is to construct a “voice file.” This file, named voice.md and placed within the designated Cowork folder, serves as a direct input for Claude before it handles any task. As Sherrard-Smith explains, this file should contain three key elements:

1. Five sentences the user has personally written, ideally sourced from recent LinkedIn posts.
2. A list of at least ten words the user intentionally avoids using.
3. A specific rule regarding sentence length, such as Sherrard-Smith’s own guideline: “Max 12 words per sentence. One idea. Stop.”

This structured approach provides Claude with clear parameters to mimic the user’s style. Sherrard-Smith notes that this file should be updated monthly to maintain accuracy.

Add three things to it: i/ 5 sentences you’ve actually written. Copy them from your last LinkedIn post. ii/ 10 words you never use. (add more as you refine this skill) iii/ Your sentence length rule. Mine: “Max 12 words per sentence. One idea. Stop.”

Creating a Custom AI Skill

Following the creation of the voice file, Sherrard-Smith outlines how to build a specific AI “Skill” within Claude’s Cowork tab. He suggests naming this skill “Write Like Me.” The core of this skill is a precise prompt that instructs Claude to prioritize the user’s voice profile.

The prompt Sherrard-Smith recommends is:

“Read voice.md first. Match the sentence length, word choice, and rhythm exactly. If you use any word from my banned list, flag it. Draft this: [task].”

By implementing this, Sherrard-Smith argues, users can ensure that AI-generated content originates from their established voice rather than the AI’s default, generic output. He points out that this method helps prevent the kind of “AI slop spam” that is diluting content quality on professional networks.

Addressing Limitations and Enhancements

While Sherrard-Smith’s method is designed to capture written voice, he acknowledges its limitations. He states that Claude cannot replicate spoken cadence or rhythm. For users whose voice is heavily influenced by these elements, Sherrard-Smith suggests recording and transcribing spoken words, or even using transcripts from professional calls, and including them in the voice.md file.

He concludes by highlighting the efficiency of his method. While setting up the skill takes an initial 20 minutes, Sherrard-Smith believes it is a worthwhile investment that ultimately saves time and maintains brand authenticity in content creation.