"This doesn't sound like me."

If you've used AI for content creation, you've probably thought this. Maybe said it out loud. Maybe thrown your laptop across the room (okay, hopefully not).

The output is technically fine. The grammar is correct. The structure makes sense. But something is off. It's got that AI smell—that slightly corporate, overly helpful, suspiciously smooth quality that screams "a robot wrote this."

And for founders building personal brands or distinctive companies, that's a dealbreaker.

Here's the thing though: AI doesn't have to sound generic. It sounds generic because we're training it wrong.

Why AI Defaults to Bland

AI models are trained on massive amounts of internet text. That training creates a kind of "average" voice—a blend of millions of writing styles that ends up sounding like... nothing in particular. Professional, yes. Distinctive, no.

When you prompt AI without specific guidance, you get this average. And no matter how clever your prompt is, you can't fully override years of training with a few sentences of instructions.

"Write in a casual, friendly tone" doesn't cut it. Every AI user says that. The model has seen that instruction millions of times paired with millions of different examples. It doesn't know what casual and friendly means to you.

The Voice Training Method

Here's what actually works: showing, not telling.

Instead of describing your voice, you demonstrate it. Instead of giving adjectives, you give examples. Instead of hoping the AI figures it out, you train it explicitly.

In Lenny, this looks like building a voice profile from your actual writing. Not a paragraph describing your tone—but dozens of examples of your real content. Your blog posts, your emails, your social media, your internal docs.

The AI learns your patterns. Your sentence structures. Your favorite phrases. The way you start paragraphs. The rhythm of your thinking.

It's the difference between telling someone "I'm funny" and actually making them laugh.

The Process That Changed Everything

When I set up my voice training in Lenny, I started with three sources: my best-performing blog posts, my most personal emails to clients, and my unhinged Twitter threads.

Yes, the Twitter threads. Because that's where I'm most myself—not performing, not polishing, just thinking out loud.

I organized these into categories. "This is how I explain complex ideas." "This is how I tell stories." "This is how I make a point land." Each category became a reference point for different types of content.

The first output I generated after this training was... weird. In a good way. It had my tics. My rhythms. The way I interrupt myself mid-thought. The way I circle back to earlier points.

It sounded like me. Actually me. Not AI-pretending-to-be-casual. Me.

Beyond Voice: Training for Everything

Once you understand this principle, you can apply it everywhere.

Want AI that understands your audience? Don't describe them—show examples of content they've responded to.

Want AI that matches your brand positioning? Don't explain it—demonstrate it through your best marketing.

Want AI that knows your product inside and out? Don't summarize—upload your actual documentation, your sales calls, your customer success stories.

Every piece of context you provide makes the output more specifically, authentically, usefully yours.

The Authenticity Advantage

Here's why this matters beyond just "sounding right."

Audiences can feel authenticity. They might not be able to articulate why one piece of content resonates and another doesn't, but they know. The generic AI stuff gets scrolled past. The authentic stuff gets saved, shared, remembered.

In a world where everyone has access to AI, the competitive advantage isn't using AI—it's using AI in a way that's uniquely yours.

That requires training. Real training. Not prompts—systems.