Method 2: Voice Dump Synthesis

Thinking at the speed of speech

Last week I introduced the methods framework and walked through Method 1: The Handshake—how you configure the partnership before the real work starts. This week: the method I use most often.

Here’s a number that should change how you think about working with AI: you speak at roughly 150 words per minute. You type at about 40.

That’s not a small difference. That’s a 4x gap between the speed of your thinking and the speed of your input.

Every time you sit down to type a prompt, you’re running your ideas through a bottleneck. You’re filtering, editing, second-guessing, polishing before the AI ever sees what you’re actually thinking. By the time you hit send, you’ve already killed half your ideas—they didn’t survive the translation from thought to keyboard.

Voice Dump Synthesis is built on a simple premise: stop writing to the AI. Start thinking out loud and let the AI do the organizing.

You talk. You ramble. You contradict yourself. You explore dead ends. You say “wait, no, actually what I mean is...” and keep going. Then you hand the whole mess to the AI and tell it what to build from the raw material. The AI doesn’t need your ideas to arrive clean. It needs them to arrive at all.

Why the default fails.

Most people treat AI like a form to fill out. They try to write the perfect prompt: clear, structured, complete. And in doing so, they create exactly the problem they were hoping AI would solve.

The blank page bottleneck: you’re staring at the chat window trying to figure out how to phrase your question. Meanwhile, the actual idea is fully formed in your head—just not in a shape you’d feel comfortable typing. So you simplify it. You trim the nuance. You leave out the half-formed connections. What reaches the AI is a sanitized version of your thinking, and you get a sanitized response to match.

The nuance problem: the most valuable parts of your thinking are often the messiest. The tangent that turns out to be the real point. The contradiction that reveals a tension you hadn’t processed. The self-correction mid-sentence that shows you working something out in real time. When you type, you edit all of this out. When you speak, it stays in.

The dead voice problem: when your input is stiff and over-polished, the output is stiff and over-polished. The AI mirrors what it receives. Feed it a carefully constructed prompt and you get something that sounds like it was written by committee. Feed it an unfiltered brain dump and the output retains something of your actual voice, your cadence, your way of framing problems.

The division of labour.

Voice Dump Synthesis works because it separates two tasks that most people try to do at once: generating ideas and organizing them.

Your job is volume and substance. You externalize everything at full speed—the conclusions, the reasoning, the doubts, the side thoughts. You don’t worry about structure. You’re not writing. You’re just getting raw ideas out.

The AI’s job is order and container. It finds the signal and organizes it into whatever format you specify. Your incoherent twenty-minute voice note is a trivial parsing task for a system trained on the entire internet.

This division is liberating. You never have to be both the thinker and the organizer at the same time.

The moves.

In my practice, Voice Dump Synthesis has three moves most often. These are the things I find myself doing over and over. Take what’s useful, adapt what makes sense, try your own stuff—the goal is finding what aligns with how you like to work.

The Raw Dump is the permission move. You signal—to the AI and to yourself—that what’s coming will be messy. “I’m going to give you a messy, unedited brain dump. Don’t clean up my writing, don’t improve my ideas, don’t add your own. Just play back what’s already there.”

The permission matters more than it looks. Saying “this is going to be messy” isn’t just an instruction to the AI. It’s a release valve for your internal editor. It lowers the stakes of the input, which raises the volume of what you capture.

The Structural Request tells the AI what to build. You never say “clean this up” or “make this better.” You name the exact container you want. “Take this transcript and organize it into three distinct themes with supporting points under each.” Or: “Turn this ramble into a week-by-week project plan in table format.” The specificity of the container determines the quality of the output.

The Gap Check makes this iterative rather than one-shot. After you see your thinking organized, you spot the holes. “What’s missing from this? What have I implied but not stated? What would a skeptic immediately ask?” Then you fill the gaps—often with another voice dump. You’re converging on clarity through iteration rather than trying to nail it on the first attempt.

When to use it.

This method has the broadest trigger set in the framework. If you’re stuck, use it. If you’re overwhelmed, use it. If the idea feels too big or too interconnected to type out linearly, use it.

After a meeting or a walk ideas are freshest in the fifteen minutes after the experience that generated them. At the start of any major document, dump first, outline second. When you’re blocked, writer’s block is almost always an editing problem, not a thinking problem. The Raw Dump sidesteps it by removing the requirement that your expression be “right.”

In practice, this becomes the starting point for almost everything. It’s the method with the highest frequency of use because it solves the most universal problem: getting what’s in your head into a form you can work with.

What’s underneath this.

Voice Dump Synthesis is cognitive offloading. And if that sounds technical, or worrisome, consider that you already do this all the time.

When an idea lives only in your head, you can’t really examine it. It’s still part of you. The moment you externalize it—speak it out loud, see it as text on a screen—it becomes something separate. Something you can look at, rearrange, judge, improve.

This is the same thing that happens when you commit ideas to sticky notes in a workshop. You can’t hold the meaning of 50 different ideas in your head at once. But you can write each one on a Post-it, put them on a wall, and start grouping them. You can use structured processes—focused conversation, group sorting—to work through all 50 and come to conclusions.

What we’re doing with AI is mechanically the same thing. The difference is scale and speed. You can dump twenty minutes of raw thinking into a thread and have it organized in seconds. You can iterate three or four times in the span of a single work session. But the cognitive move is identical: externalize the ideas so you can work with them.

Once your thinking is external, you stop being a struggling writer trying to get words on a page. You become a strategic editor working with material that already exists. The cognitive load drops. The quality goes up.

And there’s a compounding effect. People I work with who adopt this as a regular practice tell me they start thinking differently, even when they’re not recording. The habit of externalizing loosens the grip of perfectionism. You stop waiting for the fully formed idea and start capturing thinking in motion.

That’s the promise. Not that the AI will think for you, but that it will hold your thinking in a form you can actually use—so you can focus on the part that only you can do: having the ideas in the first place.