Earlier this week, I was chatting with a friend who said they knew they needed to be using AI more often but didn’t quite know where to start. I shared some ideas with them on how to get the most out of AI, using three distinct approaches that non-technical people can comfortably use.
In fact, I’m using one of these approaches right now to write this post. I recorded myself talking through what I wanted to cover, fed the transcript to an LLM, and now we’re turning my meandering thoughts into structured prose. That’s the first tip I want to share with you.
Tip 1: The Speech-to-Transcript-to-Document Workflow
Talk first, structure later.
Your iPhone has Voice Memos. Your Android device has Recorder. Both apps generate transcripts automatically when you record yourself speaking, and that’s what makes this approach possible.
Here’s the workflow:
- Open the app, hit record, and talk through whatever you’re trying to write. Don’t edit yourself. Don’t worry about structure. Just do a “brain dump” about your topic into the recording.
- When you’re done, copy the transcript generated by the app and paste it into your favorite LLM, such as Claude, Gemini, ChatGPT, or something else.
- Give the LLM two pieces of context: WHO you’re writing for (your boss, your team, potential customers, etc.) and WHAT you want them to do after reading (approve your proposal, understand the process, buy your product, and so on). The actual prompt is simple: “Take this transcript and organize it into a coherent piece for [audience] that leads them to [action].” Optionally, provide an outline or attach a template for what the output should look like.
The LLM handles the restructuring. You handle the ideas. That’s the division of labor.
And yes, that’s how the first draft of this post was written.
Tip 2: Show the LLM what you want; let it figure out how
You can reverse-engineer the perfect prompt if you have examples of what “good” output looks like.
There are too many great videos and podcasts in the world and not enough time to watch/listen to them all. I wanted a reliable way to generate show notes from transcripts so I can skim the notes and decide if it’s worth watching/listening to the full thing.
I love how Lenny’s Podcast formats their episode summaries—clear structure, useful takeaways, scannable sections, links to references. So I downloaded the transcript of one of their meatier episodes, saved a copy of the corresponding show notes, and fed both to Claude.
My prompt: “Review the transcript and show notes I provided. What kind of prompt would I have to give you so you can take a different transcript and generate the equivalent show notes?”
Claude analyzed both pieces and wrote the prompt for me. I’ve since customized the prompt further to match my preferences (see example), but the foundation came from showing the LLM the input-output pair I wanted to replicate.
This works for any repeated task where you have examples of good output. Email responses. Project briefs. Status updates. Proposals. Requirements documents. Whatever it is, find one example you love, pair it with its source materials, and ask the LLM to reverse-engineer the prompt that performs the transformation.
Tip 3: Capture your coaching in custom instructions
Your best instructions emerge through conversation.
You start your chat with an LLM with a rough request. The LLM gives you something close. You course-correct: “Make it more direct.” “Focus on the second point.” “Use fewer adjectives.”
After fifteen minutes of back-and-forth, the LLM finally understands your style and needs. Then you close the chat and lose all that context.
Don’t!
At the end of any productive conversation with an LLM, add one final prompt: “I would like us to reflect on what we’ve learned in today’s session so we can add guidelines to my custom instructions for you. For example, I think we should make it a guideline to [specific way of working together 1]. Another guideline is [specific way of working together 2]. Are there any other guidelines that we should formalize?“
You can see an example below.
The LLM will analyze the patterns in your conversation and codify them into reusable guidelines. In ChatGPT, these go in Custom Instructions. In Claude, they go into Personal Preferences. In Claude Code, they go into your CLAUDE.md file. In Gemini, they go into “Your instructions for Gemini.” In all cases, you have a persistent set of contextual guidelines that will automatically be carried across future conversations.
Some LLMs try to mimic this by creating “memories” that store data about you, but the “memories” are hit-or-miss in my experience. Take control of the memory-making process by explicitly asking the LLM to perform the reflection and choosing what you want saved.
Each conversation teaches the LLM something new about how we prefer to work, what we value, and what we need. These lessons compound.
You already have what you need to start
You don’t need specialized tools or technical knowledge to apply these practical tips. You have a voice recorder on your phone. You have access to an LLM. You have examples of output you admire.
There’s no secret recipe or special sauce. You only need to know these three usage patterns exist and try them once.
Which tip will you try today?
Author’s meta-note: I did my verbal brain dump of this piece using Monologue, and this post is my first attempt at writing with the help of Spiral. I genuinely didn’t know what to expect from Spiral so I’m delighted to report that I went from blog post idea to verbal brain dump, and further to a complete, finished draft in 41 minutes. 🤯
