Ever read something online and wondered, “Was this written by a human or AI?” It’s a fair question these days, AI slop is everywhere. As I’ve leaned more into using AI to help with my writing, I realized I needed a clear system to stay transparent with my readers. It’s not about passing off AI work as my own; it’s about leveraging these incredible tools to bring better in-depth content faster.
To do that, I’ve developed a workflow and labels you’ll see on my posts: AI Generated, AI Assisted, and AI Reviewed. This post is a quick FAQ to explain what they mean and how I create content now.
- AI Generated: The core of the content, from structure to prose, was created by an AI model based on my detailed prompts and outlines. I’ve heavily edited and fact-checked it, but the initial draft was machine-made.
- AI Assisted: I wrote this content, but I used AI for specific tasks like brainstorming, refining sentence structure, or generating summaries of research material. Think of it as a super-powered thesaurus and research assistant.
- AI Reviewed: This is my own original work, but I’ve had an AI model review it for clarity, tone, and grammatical errors. It’s like having a final proofreader before hitting publish.
My Core Workflow
My goal is to blend the best of human creativity with the power of AI. Here’s a look at my new process.
1. Deep Research
It all starts with a question or an idea. My first step is what I call “Deep Research.” I don’t just rely on one source or one AI. I leverage deep research capabilities of several AIs.
Deep Research is my process of using large language models (OpenAI’s GPT series, Anthropic’s Claude, and Google’s Gemini) to search sources and the internet to build grounded responses with the latest information and context.
I gather all these different sources—links, articles, AI-generated deep research summaries—and put them into a central place.
2. Research and Ideation in NotebookLM
This is where the magic happens for me. I use Google’s NotebookLM as my digital research assistant. I upload all my sources, and it helps me make sense of everything.
NotebookLM is a research and writing assistant from Google. You can upload your sources (PDFs, text files, Google Docs, website links), and it creates a private model grounded in that specific information. It helps you summarize, ask questions, and brainstorm ideas based only on the materials you’ve provided.
I’ll read through my sources within NotebookLM and start jotting down initial thoughts and connections in the notes section.
3. Post Structuring with Advanced Reasoning Models
Once I have a good grasp of the material, I need a structure. I turn to an advanced AI model to help me outline a post. This isn’t just a simple “give me a blog post” prompt.
Advanced Reasoning with Chain of Thought (CoT) involves prompting a powerful AI model (like GPT-4o or Gemini Pro) to “think step-by-step.” Instead of just giving a final answer, it explains its reasoning process. This helps me iterate on the structure, consider different angles, and build a more logical and compelling narrative for the post.
I’ll go back and forth with the model, refining the outline until I’m happy with the flow and the key points I want to hit.
4. Validation and Skeleton Drafting
With a solid structure in place, I do two things simultaneously. First, I use NotebookLM’s audio feature to talk through content using my plan/structure.
NotebookLM Audio Overview is a feature where the tool generates an audio discussion based on your sources and notes. Listening to an AI-generated conversation about my topic helps me validate my understanding and sometimes sparks new ideas. It’s like having a quick chat with a knowledgeable colleague to see if my argument holds up.
At the same time, I write a “skeleton” post. This is a bare-bones draft that includes the main headings, key data points, and specific instructions for tone, style, and what to avoid (I’m looking at you, “but here’s the Twist…” and em dash).
5. Dual AI Generation and Human Merging
This is where things get really interesting. I take my skeleton post and detailed prompt and give it to two of the best writing models available (for me, right now, that’s OpenAI’s GPT4.5 and Google’s Gemini Pro 2.5). I get two different “takes” on the same content.
I then use another advanced reasoning model to compare the two drafts, highlighting the strengths of each. Finally, I sit down and manually merge them. I rewrite sections, add my own voice and personal anecdotes, and fact-check where required.
Comments and Reflections
So, what does this all mean?
The Good: This workflow allows me to tackle more complex topics and produce higher-quality, better-researched content more efficiently.
The Challenges: It’s not a perfect system. AI models can still “hallucinate” or generate plausible-sounding nonsense. That’s why the human editing, grounding and fact-checking step is crucial.