Mariam Gogidze on Training AI for Authentic LinkedIn Voice

In a recent LinkedIn post, Mariam Gogidze explores the challenges of using generic AI tools for content creation, particularly on professional platforms like LinkedIn. She argues that while AI offers potential efficiencies, its default output often lacks the distinct personality and specific nuance required to connect with an audience. Gogidze highlights a common pitfall: AI trained on the vast, undifferentiated data of the internet tends to produce content that is grammatically correct but devoid of individual character.

Gogidze recounts her own experiences attempting to steer AI tools towards her specific voice. Initial attempts using prompts like “write in my voice” or “be more casual” proved ineffective, yielding what she describes as “corporate-speak with exclamation points” or “forced casualness.” She found that even detailed style guides or lengthy prompts were inconsistent in their results.

“The fundamental issue: General AI optimises for AVERAGE. Great content requires SPECIFIC.”

The Problem with Generic AI Output

The core of Mariam Gogidze’s argument centers on the inherent nature of large language models trained on broad datasets. As she points out, these models are designed to identify and replicate common patterns, leading to an output that mirrors the “entire internet” rather than a specific individual’s style.

“ChatGPT is trained on the entire internet. It writes like… the entire internet,” Gogidze states in her post. She elaborates on the characteristics of this generic output: “Perfect grammar. Zero personality. Maximum ‘I am an AI language model’ energy.” This lack of distinctiveness, she contends, is immediately recognizable to an audience attuned to authentic voices.

Finding a Solution: Training AI on Personal Data

Frustrated with the limitations of generic AI, Gogidze sought a different approach. She discovered a tool, MagicPost, developed by Naïlé Titah, which focuses on training AI models using an individual’s existing content. This method shifts the paradigm from instructing AI on how to write to demonstrating how the user already writes.

Gogidze explains the effectiveness of this personalized training method:

“Instead of: Telling AI how to write. I started: Showing AI how I already write. Feed it your actual posts. Let it learn YOUR patterns. Not LinkedIn’s average patterns.”

She contrasts the output of a generic AI with one trained on her specific voice. While a standard AI might produce a generic hook like, “Ready to transform your LinkedIn strategy? Here are 5 game-changing tips! 🚀”, her personalized AI generated a more distinctive and revenue-focused statement: “I wasted 6 months on LinkedIn growth tactics that looked impressive in screenshots but generated zero revenue.”

A New Workflow for Authentic Content

Mariam Gogidze outlines a revised workflow that integrates AI more effectively into her content creation process. This involves:

• Importing a substantial number of high-performing past posts to allow the AI to learn her unique writing “DNA.”
• Generating initial drafts that capture her voice and style.
• Editing these AI-generated drafts for specificity, significantly reducing the time compared to writing from scratch.

Gogidze emphasizes that this approach results in content that is not entirely AI-generated but rather a hybrid: “They’re 70% AI foundation + 30% human refinement.” This blend allows for greater consistency without sacrificing authenticity.

The Future of AI in Content Creation

Ultimately, Mariam Gogidze’s post suggests that the effectiveness of AI in content creation hinges on the training data. “People don’t want ‘perfect’ content. They want CONSISTENT content that sounds like you,” she argues. The key question for content creators, according to Gogidze, is not whether to use AI, but rather whether they are using AI trained on their personal voice or on a generalized digital echo.

“The real question isn’t: ‘Should I use AI for content?’ It’s: ‘Am I using AI that knows MY voice, or everyone’s voice?'”

Gogidze is currently conducting a 60-day test with MagicPost and plans to share her findings. She invites others to share their experiences with AI for LinkedIn content and what challenges they have encountered.