NOTES · 2026-04-28

When the email Claude wrote sounds smooth and your partner says "this didn't sound like you"

I stared at the draft Claude generated for my client. It was flawless, polite, and utterly hollow. When I read it aloud to my partner, they stopped me mid-sentence and said, "This didn't sound like you." That moment stung because it was true. I had let the machine scrub away my actual voice in favor of corporate smoothness. We often trade our authenticity for efficiency, especially when our brains are screaming for a shortcut. But when the output feels like a stranger wrote it, we lose the connection that makes our work matter. I built a specific editing drill to catch this before it leaves the inbox. We are going to reclaim our syntax from the algorithms that try to flatten us.

If the email sounds perfect but doesn't sound like you, it is not professional. It is fake.

The Smoothness Trap of AFM-2

I call this specific failure mode AFM-2: Voice Laundering. It happens when you feed a rough, jagged thought into Claude or ChatGPT and ask it to make it professional. The model obeys. It removes the sentence fragments you use when you are excited. It replaces your specific vocabulary with generic business speak. The result is a email that reads like it was written by a committee of people who have never met you.

For us, this is dangerous. Our communication style often carries the nuance of our neurodivergence. The directness, the associative leaps, the specific rhythm of our typing. When we let AI sanitize this, we are not just polishing; we are erasing. The smoothness is a mask. It hides the human on the other end, and it hides you from them. You might get the reply you wanted, but you lose the relationship you built.

Why Your Partner Noticed Immediately

People who know you have a mental map of your voice. They know how you start sentences. They know when you use a comma where a period should be. When I sent that sanitized draft to my partner for a quick glance, their brain flagged the anomaly instantly. It wasn't that the content was wrong. It was that the texture was off. It felt like wearing a suit that fit perfectly but smelled like someone else's laundry detergent.

This is the core of the problem with relying on default AI outputs. Tools like Gemini or Notion AI are trained on the average of all human writing. They converge on the mean. If you are neurodivergent, your natural communication style is rarely the statistical mean. When you accept the default output, you are actively moving away from your natural state toward a generic center. That is why it felt fake. It was a statistical average, not a human connection.

The Read-Aloud and One-Typo Protocol

To fix this, I developed a hard rule for any AI-assisted communication. Before I send anything, I must read it aloud. If I stumble over a phrase because it feels too formal or clunky in my mouth, I delete it. If I cannot say it without sounding like a robot, it stays deleted. This forces the text back through my own vocal cords, re-imprinting my rhythm onto the words.

The second part of the protocol is the one-typo edit. This sounds counterintuitive. Perfection is the goal of the machine, not the human. I intentionally look for places where the AI made the text too rigid and I break it. I might split a long sentence into two fragments. I might start a sentence with 'And' or 'But' if that is how I think. I introduce the friction that the AI tried to remove. This is not about making errors; it is about re-humanizing the text so the recipient hears me, not the server farm.

Voice Preservation Drills for Daily Use

You can train your tools to help rather than hinder. Instead of asking Claude to 'rewrite this professionally,' I now prompt it to 'keep my original tone and sentence structure, only fix the spelling.' I provide examples of my own writing in the context window. In Claude Projects or Custom GPTs, I upload samples of emails that got a good response. I tell the system explicitly: 'Analyze the sentence length and vocabulary of these samples. Mimic this style, do not smooth it out.'

Another drill I use is the 'ugly first pass' method. I write the rawest, messiest version of the email possible, typos and all. Then I ask the AI to organize the points but strictly forbid it from changing the voice. This keeps the core energy of the message intact. The AI becomes a secretary organizing my thoughts, not a ghostwriter replacing my soul. It requires more oversight, but the output remains mine.

When Smoothness Actually Hurts Your Brand

There is a misconception that professional means polished to a shine. In reality, trust is built on consistency. If your emails are usually direct and slightly chaotic, and suddenly you send a perfectly structured essay, the recipient feels unsettled. They wonder if you are angry, or if someone else is writing for you. Consistency of voice is a signal of safety and reliability.

I learned this the hard way when a client responded to a 'smoothed' email by asking if everything was okay. They thought I was distancing myself. The very tool I used to save time ended up costing me a relationship repair conversation. We have to remember that our quirks are often the very things people rely on to understand our intent. Stripping them away creates ambiguity, not clarity.

Building Your Own Voice Library

To prevent future laundering, I maintain a personal library of my best-written emails and documents. I store these in a dedicated space where my AI can reference them. When I need to draft something new, I point the model to this library. This ensures that the baseline for 'good writing' is my own history, not the internet's average. It acts as an anchor against the drift toward genericism.

This approach aligns with the broader philosophy of building systems that work for our specific brains. We are not trying to become neurotypical writers. We are trying to become more effective versions of ourselves. The technology should amplify our natural patterns, not erase them. If the tool makes you sound like everyone else, it is failing its primary function for you.

The Cost of the Fake Voice

Every time we send an email that doesn't sound like us, we reinforce the idea that our natural way of communicating is wrong. We tell ourselves that we need to be fixed. But the smoothness is a lie. It is a facade that requires constant maintenance. Eventually, the mask slips, or worse, we forget what was underneath it. We become the corporate drone the algorithm thinks we are.

Reclaiming your voice is an act of resistance. It is a declaration that your specific way of seeing and saying things has value. It takes courage to send an email that feels a bit rough around the edges but rings true. It takes discipline to edit the AI back into humanity. But the alternative is a world where we all sound the same, and that is a world I do not want to live in.

Key takeaways

AFM-2 Voice Laundering occurs when AI removes your unique neurodivergent syntax in favor of generic smoothness.
The Read-Aloud Protocol forces you to hear unnatural phrasing that your eyes might skip over.
Intentionally breaking perfect sentences restores the human rhythm and builds trust with the recipient.
Train your AI tools on your past writing samples to prevent them from defaulting to the statistical average.