Declaration of Independence Flagged as 99% AI-Generated

Hey, here's this week's AI Brief!

💼 In Today’s 5-min AI Brief

• Your AI Toolkit: Build an AI writing system that sounds like you

• Prompt of the Week: Extract your voice into a reusable guide

• Last Week in AI: Claude Cowork, Healthcare AI, Anthropic's $10B raise

• What I'm Learning: Building an audience voice database

🛠️ Your AI Toolkit

Stop Trying to Beat AI Detectors. Build This Instead.

An AI detector just flagged the Declaration of Independence as 99% AI-generated. Written in 1776. 250 years before AI existed.

Meaning… AI detectors just don’t work. 

Even OpenAI discontinued their AI writing detector tool due to low accuracy.

Here's what's actually happening: AI detectors look for patterns like organized structure, formal wording, and logical flow. So the better your writing is, the more likely it gets flagged. That's partly why "beating" detectors is the wrong approach.

Meanwhile, AI writing has tells that humans spot instantly. Em-dashes everywhere. Phrases like "game-changer" and "noteworthy." Generic structure that could've been written by anyone. And of course, “it’s not this, it’s that”.

You might think the fix is a magic prompt. Tell ChatGPT to avoid those patterns, and yes, you'll get less AI-sounding writing. But the remaining problem is it still won’t sound like YOU.

The better approach: Build an AI writing system that captures your voice from the start.

Here's the framework:

1. Extract your voice into context files. Analyze your best writing. Document your sentence patterns, signature phrases, and the words you actually use.

2. Train AI on your real work. Feed it 3-5 pieces that sound unmistakably like you. Not your most polished stuff. Your most you stuff.

3. Create reusable workflows. Build templates for each content type you create regularly, so you're not starting from scratch every time.

There's no shortcut. But once it's built, it works.

The full setup guide can be found in my free AI Workflow Library if you’re interested.

💭 Prompt of the Week

Voice Extraction Prompt

This is step one of building your AI writing system. Before you can train AI to write like you, you need to know what "like you" actually means. This prompt analyzes your writing and outputs a voice profile you can reuse forever.

Grab 3-5 pieces of content that sound unmistakably like you, paste them in, and run this:

Save the output. That voice guide becomes the foundation for every AI writing task going forward. Instead of fighting detectors, you're building a system that sounds like you from the start.

💡 Last Week In AI

• OpenAI and Anthropic Race Into Healthcare

  • Both companies launched health features within days of each other. OpenAI's ChatGPT Health connects to Apple Health and medical records. Anthropic's Claude for Healthcare also has Apple Health connections but goes deeper with connections to medical databases like PubMed.

  • I’ll definitely be checking this out because I love tracking my health data haha. Also, these launches give us a tip on how AI model companies are honing in on addressing security concerns.

• Claude Cowork: Your First Real AI Coworker

  • Anthropic released Cowork, and this one's different from your usual Claude chat. Give it access to a folder on your Mac, and it can actually take action: organize files, turn receipt screenshots into expense spreadsheets, draft reports from scattered notes. You can even launch multiple agents to work in parallel.

  • Currently in research preview. Just expanded from Max subscribers to Pro ($20/month) as of this week.

• Anthropic Raises $10B at $350B Valuation

  • Anthropic signed a term sheet with Coatue and GIC. OpenAI sits at $500B for comparison.

  • I think this reflects what users already know. Claude has the best coding model right now, many say the best writing, and it's my personal favorite. Features like Cowork, Claude Code, and Skills are shipping first from Anthropic. And it feels like they’re the innovators that everyone now copies.

📖 What I’m Learning

Building an Audience Voice Database

I've been interviewing people in my target audience and documenting exactly what they say, thanks to AI meeting notes. I soon realized this was gold for not only understanding what they want better, but also for context for my AI systems to be able to pull in.

Phrases like: "People are just scared of the word AI" and "My sister was like, 'What does automate mean?'" and "It's very, learn on your own. They give us licenses to Coursera... but it's all just initiative."

This week, I started to log each quote with the pain point it reveals, what stage they're at in their AI journey, and what potential service or educational lesson ideas it sparks. 

Just like extracting our own voices for our voice guidelines, I’m now seeing the power of creating and extracting your audience's voice to know what to offer and write about. Both are amazing contexts for our AI’s. And if you're creating offers or media/content for a specific audience, try collecting their words to fuel your creation.

Final Thoughts

The theme this week is really to take the time to do the foundational work of extracting things out into documentation for your AI systems. That takes more work upfront, but it's the only approach that’s going to consistently represent your voice.

-Thaddeus