I find LLMs are great at taking raw thoughts and echoing them back in a coherent way. Presently, I lean toward Gemini for this, but any flagship model would work equally well.

Depending on my level of stuck, I pick one of two approaches described below.

Method 1: The Raw Brain Dump
When starting a project or trying to communicate something complex, I do an unfiltered brain dump. I type my pure train of thought directly into the prompt—incomplete sentences, random phrases, and all.

Then I tell the model to organize it. The output usually captures my original intent and still sounds like me, but in a clean, structured format.

Method 2: The Q&A Prompt
If the topic is so complex that I struggle to even write a cohesive brain dump:

  1. I give the model background context on the task and prompt it with: “What questions should I consider as I’m working through this?”
  2. I paste its questions into a text editor and answer them one by one. Again, no editing, no pausing—just raw thoughts.
  3. I paste my answers back into the chat.

While AI is far from perfect, the success rate here is surprisingly high. Almost every time, the output of this method gives me a clear path forward, and I can still see my own words echoing back from the page