Should you talk or type to AI tools like ChatGPT and Claude?
Short answer
Talk. Speaking is at least three times faster than typing by my estimate, and the bigger win is that you stop self-censoring, so the model gets more context. I use Wispr Flow for dictation, don't type more than a sentence anymore, and even code by voice. Type only exact strings like URLs, numbers and code.
If I could give every business owner one AI habit, it's this: speak instead of type. Voice dictation is the one tool I will never not have for the rest of my life. Going back to typing would be like going back to the stone ages.
"I don't type more than one sentence anymore. When I get to sentence two my brain goes, dude, stop, just talk."
Why is talking better than typing for AI?
Two reasons, and the second one matters more.
- Speed. You speak at least three times faster than you type. That's my estimate, not a lab measurement.
- You stop self-censoring. When you type, no matter how fast you are, you edit as you go because your fingers can't keep up with your head. You curate. You cut context you think doesn't matter. When you talk, you just go. I call it brain dumping.
The second point is why this works so well with AI. You have to flip your brain from Google to AI. In Google, typing too much gets you a bad result. In any LLM, the more information you give it, the better. Rambling is a feature here.
What tool do you use for dictation?
Wispr Flow. It works inside any app, so it doesn't matter if I'm in ChatGPT, Claude, a coding agent, Slack or email. I've quoted it at about $15 a month and I tell every single person to get it.
One way to get it cheaper: I tell a lot of people about Lenny's bundle, a newsletter subscription that includes a year of Wispr Flow plus a stack of other AI tools. I've said a year of Wispr Flow alone is worth about $240, and the bundle includes it plus the others. The AI companies give it away because that newsletter's audience is exactly who they want, which is kind of our business model in some ways.
Does this work for coding, too?
Yes. I never touched a line of code until 2026, and I code entirely with my voice. I talk to the coding agent the way I'd talk to a developer sitting next to me: what I want, what's broken, what it should look like. Most of my instructions to agents arrive as long, unedited voice notes. The rambling is deliberate.
When should you still type?
- Exact strings. Dictation is poor for URLs, code snippets, exact numbers and spellings of names. Paste or type those.
- Anything that becomes a commitment. When a spoken ramble turns into a task, a plan or a promise, have the AI reflect it back to you in writing first and check it. Spoken intent is great input. It still deserves a review before it becomes action.
How do I start on Monday?
- Install a dictation tool that works system-wide.
- Next time you open ChatGPT or Claude, hold the key and explain the task like you're talking to a smart friend. Who it's for, what good looks like, what you've tried, what annoys you.
- Don't clean it up. Send it.
- For anything serious, end with "don't do it yet, write me the best prompt for this." That's meta-prompting, and it pairs perfectly with dictation.
Answered by TJ Larkin. From Local Media HQ podcast episodes and AI training sessions in 2026. Last updated .