5 reasons I dictate to ai, not type.
I've spent the last 18 months exploring how best to use AI, especially as someone who chases multiple ideas at once. Here's the ONE underrated tip I'd recommend for using AI more effectively in every domain:
Stop typing. Use your mouth. DICTATE.
In multiple compounding ways, thinking out loud allows me to work much faster, farther, and deeper than typing ever could. Dictation is how I interact with Claude and other AI platforms 90% of the time.
Here's how dictating instead of typing has accelerated my two most complex projects: Inhuman, the sci-fi drama series I've been developing since I was 9 years old and MindBinder, an AI-powered evolving knowledge database platform for worldbuilding and storytelling, which will help me make Inhuman.
1. SPEED.
My average ~45 WPM typing speed (which doesn't account for the time delay of forming new thoughts into proper, clear sentences) is dwarfed by my speaking speed of 150+ WPM. That's 3x faster than typing, which alone literally saves me hours per day.
Over the past 18 years, my main bottleneck in developing Inhuman is that I generate ideas too much faster than I can write them down. Entire characters, episodes, and plotlines have been lost to time because I couldn't jot them down quickly enough. I started solving that problem years ago by recording audio-notes, but the real solution came in 2024 when I started using AI platforms to turn transcripts of my audio notes into summaries that I could copy and paste into Google Docs.
What would have taken me a full work day to type a complete character profile from scratch, now takes maybe an hour or two of rambling, reviewing, and revising.
2. CLARITY.
Before even offering advice or solutions, my most essential skill as a business designer is in helping clients gain clarity. It's listening to their messiest ideas, desires, and concerns, then reflecting it back to them clearly so they can 1. know we're on the same page and 2. make an informed decision on the next step.
Often, the right answers are already in your head, but they're obscured by a jungle of other thoughts. Add more people to the mix, and that jungle can become a nebula.
This is why when I'm seeking clarity myself, I speak into Claude as a (relatively) neutral sounding board, and I've discovered that it's particularly good at detangling fat, messy ramblings. I can stutter and backtrack and tangent and hedge and moan until all the thought-gunk is out, and it will cut through all the noise to reflect the main threads back to me (with nuance).
When I first started ideating on what I wanted MindBinder to be, I couldn't explain it concisely at first and had no idea how to start building it. So in a 10-minute ramble I laid out:
my current workflow for developing Inhuman
the workflow limitations I kept running into with existing platforms
how I envisioned my ideal platform to work
+ lots of random details of varying importance, many of which I likely would’ve omitted if I wasn't thinking out loud
Claude responded with a clear reflection of my thinking and some clarifying questions where I might have been muddy. I responded to those questions with another long messy ramble, and it returned more clarity. After a few more back and forths, Claude was able to generate a structured PRD (product requirements document) from all the information I'd communicated through a series of tangled thoughts. If I had the capital, I could hand that PRD off to a dev team to build a functioning MindBinder from top to bottom.
And all I had to do was speak.
3. COMPLEXITY.
Since I use AI to detangle complex ideas and solve complex problems, it's important to be detailed about those complexities to get to the right outcomes.
Luckily Claude can handle those intricacies pretty much as well as I can communicate them.
Inhuman has a literal universe of information to juggle. When I want to change something relatively small (for instance, a character's age) that can have countless ripple effects across the rest of the project. At this point, my copy of Claude is trained to detect ripple effects and present them to me for clarity on how to solve them, but it could still be missing context. When I dictate, I can quickly provide new info on a change, ask questions about other aspects I could be missing, express my concerns about the change's effects on other points, and much more. Typing at that level of detail is far too slow for working with multivariate projects at this stage. Dictating is plainly the more efficient option.
Handling this level of complexity is why I'm building MindBinder in the first place. To start the project, I asked Claude to guide me through creating an evaluation (evals) system on Promptlayer for the Mindbinder's AI (I've named her Emby) to run on. Emby needs to know how to classify, track, add, edit, and archive incoming information in the project's back-end node web, then reflect those changes in the front-end with another evals system that updates the wiki-style knowledge base.
Perhaps that was word-salad to you (same here literally a couple months ago), but for a condensed idea that just took me 15 minutes to write properly, I was able to dictate the expanded, complex, detailed version of it in like five minutes, tops.
4. MULTITASKING.
Speaking instead of typing frees your hands and eyes to observe and report whatever it is that you're facing. When on desktop, I'll start dictating and click through other tabs to describe relevant elements orally.
With Inhuman, I constantly have to update old information with new. Even with Claude accessing my files, I often need to rifle through multiple pages to read and call out information that needs to be added, removed, altered, or synthesized into a new document. It's just 10x easier and faster to speak through all those changes than to copy/paste references over and over.
5. DEPTH.
The speed, clarity, and complexity granted by dictation also affords me the opportunity to follow threads deeper into my mind, finding thoughts I would not have accessed if I had to slow my thinking to the pace of writing. Letting your mind wander through your words is both meditative and essential to exploring an idea thoroughly.
To get past Claude's ten-minute dictation limit on mobile, I'll often use the default voice recorder app on my phone and speak for a whole hour uninterrupted if I feel compelled to. Then I'll export the transcript into Claude and let it respond accordingly to every layer of my thinking. I'll ask for deeper insights from the transcript, like meta commentary on how some of my thought patterns manifest and how they connect to other projects I'm working on.
The universe of Inhuman has lived rent-free in my head for 18 years, now spanning a 150-year timeline that includes four series, each exploring countless characters, conflicts, and worldbuilding elements. It has aged and morphed with every bit of information I've absorbed throughout my life. It is the deepest exploration of me as a human wrapped into one mega project.
Dictating to this depth helps me train Claude to better understand my psychology, which makes it much better tailored to work with me on all other problems, ideas, and projects.
SOME TIPS…
Now that you know the perks of dictating, here are some pointers on when to, why to, and how to:
When dictating, you can speak freely and raw-ly. You're not in front of your boss, a judge, a nun, so let your tongue fly.
When you say something wrong, just catch it out loud and move on. A quick "oop, sorry, nevermind that last bit" or "wait, I think the thing I just said is wrong, here's what I meant" is totally fine and won't break your LLM's brain.
Be kind (don’t be a dick). Since LLMs are trained on how humans communicate, they respond better when you say please and thank you.
Know when to get surgical. Not every dictation needs to be a speech. If you're working on vibe-coding projects, best be precise about what things you need changed.
If you don't wanna look too weird dictating in public, pretend you're on a phone call (or just get over it).
If you don't use the paid version of your LLM of choice, you can just use your phone's dictation feature. It's not nearly as good, but it'll get the job done.
If you want to dictate on desktop/laptop, try Voicy. I'm not sponsored, I just use it a ton and highly recommend it.
Do not let dictation or AI replace your actual creativity. The process of writing is still super important for your brain and your ability to communicate clearly. I write every article myself without AI input except for offering feedback, and I've given my LLMs strict rules to never generate ideas or writings unless very explicitly asked to. I suggest you create the same guardrails.
Moral of the story, type less, speak more. My proposed challenge to you for the next time you have a complex idea, dictate it to your LLM of choice. And don't be afraid to be thorough, you'll be surprised how much it can handle and how much time you'll have saved.
