Talk to AI with your voice: dictation and voice mode for busy CEOs

Two ways to use Claude when you cannot type. The commute, the walk, the gap between meetings: all of it is live time now.

What you'll have when you're done

A working pattern for talking to Claude instead of typing at it. You will know the difference between the two voice options, which one to reach for in which situation, and the one technique that makes both of them actually useful instead of novelty features you try once.

No terminal. No setup. Plain app on your phone.

AI is still a desk-only tool for most CEOs

You open Claude when you sit down at your laptop. You close the laptop, the AI stops. The commute, the walk, the dead twenty minutes between calls: that time disappears.

Here is what that costs you. Every day, you are carrying half-formed thoughts that never get processed. Decisions you are circling but not resolving. Drafts you are putting off because you are never at a desk when the thinking is actually happening. The AI that could help is sitting idle because you think you need a keyboard to use it.

You do not. Claude is on your phone, and the phone surface is where most of the best AI moves happen for a CEO in motion. The two voice options below are the on-ramp.

What you need first

• The Claude app on your phone. iOS or Android. If you have not done the basic setup yet, the first 30 minutes article walks through it.
• A paid plan for reliable access to voice features. The free tier works for light use but you will hit limits fast during a commute.
• A real situation to test on. Have a decision you are sitting with, a draft you are avoiding, or a meeting coming up you have not prepped for. That is your first prompt.

That is it. No integration, no download beyond the app.

Step by step

First: the two voice options, defined

These are different features. Know which one you are using.

Dictation (speech-to-text): you speak, the app transcribes your words into the text box, and Claude replies in text as usual. Tap the mic icon in the app's input field, or use your phone keyboard's mic button on any platform. Claude reads your words and responds in writing, the same as if you had typed.

Best for: long brain-dumps, drafting on the move, capturing a thought precisely when you want a written answer you can read, copy, or forward.

Voice mode (spoken conversation): a hands-free back-and-forth where you talk and Claude talks back out loud. No typing, no reading. You have a conversation.

Best for: thinking out loud, hands-free Q&A during a commute or walk, brainstorming when your hands and eyes are occupied. Voice features have broadly rolled out, some access is free, and full access varies by plan (per Anthropic's help docs).

When you are not sure which to use: if you want to read the output later, use dictation. If you want to think out loud and keep moving, use voice mode.

Then: the one technique that unlocks both

Dictation rewards rambling. You do not need a clean prompt. The whole point is to say the mess out loud and then ask Claude to do something with it.

Speak the situation, then follow immediately with a structuring ask:

• "I just rambled for two minutes about whether to expand to a second market. Pull out the three real choices and the trade-off for each."
• "Find the one question I keep circling."
• "What am I actually deciding here?"

The ramble is the input. The structuring ask is what makes it useful. Most people skip the structuring ask and wonder why the output is unfocused. It is not the transcription that fails them. It is the missing instruction at the end.

Voice mode rewards treating Claude like a thinking partner that holds the thread. You do not have to remember where you left off or track the sub-points. Talk, react, push back, ask a follow-up. It is a conversation, not a search query.

Three concrete use cases

1 · Brain-dump between calls, get a decision structured.

Five minutes between calls. You have been carrying a decision for three days and every time you sit with it, something interrupts. Pull out your phone, tap the mic, and ramble for ninety seconds: "Okay, I keep going back and forth on this warehousing question, here is the situation, here is what I am worried about on each side..." Then: "Pull out the three real choices and the trade-off for each." Read the structured output on your way into the next meeting.

2 · Hands-free prep while walking.

You are walking to a meeting with your ops director. You have fifteen minutes. Open voice mode, start talking: "I am meeting my ops director in fifteen minutes about warehouse throughput. Give me five sharp questions to ask." Talk through the answers out loud. By the time you walk in, you are prepared, and you did not stop moving to get there.

3 · Dictate a draft response, refine it later.

You are not at a desk and a reply is sitting in your inbox that needs thought. Tap the mic: "I need to push back on this price increase but keep the relationship. Here is the situation: they are a two-year vendor, the increase is 18 percent, we are mid-contract..." Let it draft. Copy the draft when you are back at a desk, edit one line, send.

How you'll know it's working

You processed something you would have just carried. A decision that was sitting in your head is now structured. A reply you were avoiding has a draft. A prep session you would have skipped happened on a walk instead.

The output used your specific situation. Not generic advice about "vendor negotiations" but actual language shaped to your 18 percent, mid-contract situation. If the output feels generic, the dictation probably did not capture enough context. Go back and add the specifics.

You are doing this without looking at your phone. In voice mode, your hands and eyes are free. If you are reading a screen, you are probably using the wrong option for that moment.

When it breaks

Transcription errors on names, numbers, and jargon. Voice-to-text is excellent on common words and unreliable on your CFO's name, a specific SKU, a dollar figure said quickly, or any technical term outside mainstream vocabulary. Proofread anything load-bearing before it drives a real decision or goes into a reply. Transcription errors are especially quiet: the output reads smoothly and the error is invisible until someone acts on it.

Voice mode is wrong for anything you need to read carefully or copy precisely. A contract clause, a specific number you need exact, a long structured output you are going to paste somewhere. Switch to text and ask the same question; you will get a response you can read, copy, and verify. Voice mode is a thinking tool, not an output tool.

Long structured output does not belong in voice mode. If you need a six-bullet action plan you are going to send to your team, dictate the prompt and get a written response. Voice mode reading a six-bullet list back to you while you drive is hard to track and harder to act on.

Privacy. Voice goes through the app's servers. Do not dictate confidential figures, personnel decisions, or anything you would not want processed outside your organization without first checking the data setting on your account and your company's AI usage policy. The documents and images article covers the data-handling question in more detail. The short version: paid plans have more conservative handling than free, and your company may have its own rules.

Level up

Voice is the on-ramp, not the ceiling. The best use of voice mode is capturing the unstructured thinking that you will then turn into real work at a desk.

The next step from here is turning a messy brain-dump into a prioritized week: a workflow where everything you captured on the move gets triaged against your actual quarterly goals and handed back as a ranked plan. The voice dump is the input. That article is what you do with it.