Talking to ChatGPT and Claude instead of typing

Most people type their prompts to ChatGPT and Claude. That is the default, and it works. But after a while of using these tools, you start to notice something about your own typed prompts: they are shorter than the question in your head. You had a fuller thought. You typed a compressed version of it, because typing a paragraph is work and the assistant will probably figure out the rest.

It does not always figure out the rest. So you correct it, and clarify, and try again. The conversation gets there eventually, but it took three turns when one good prompt could have done it.

Speaking the prompt instead changes this — not by magic, but by removing the quiet pressure to be brief.

Typed prompts are compressed prompts

When you type, you economize. Not consciously — it is just that every word has a small cost in effort and time, and over a paragraph those costs add up, so you trim. You drop the background. You leave out the constraint that seemed obvious. You write "summarize this for a client" instead of "summarize this for a non-technical client who has not seen the project before, in about four sentences, focusing on what changed and why it matters to them."

The second version is the prompt you actually wanted. You did not type it because typing it felt like a chore. The assistant got the short version and produced something generic, and then you spent two follow-up messages steering it toward what you meant from the start.

The compression is invisible while it happens. You only see it when you compare a typed prompt to a spoken one.

Spoken prompts carry more

When you say a prompt out loud, the economizing pressure mostly disappears. Speaking a long sentence is no harder than speaking a short one. So you say the whole thought. You include the audience, the format, the length, the tone, the thing to avoid, the reason you are asking. Not because you are trying to write a better prompt — because that is simply how the thought comes out when you are not rationing words.

The result is a prompt that is longer, more natural, and more complete. And complete prompts get better answers. The assistant has the context it needs on the first try, so the first response is closer to right, and you spend fewer turns correcting it. The gain is not that you produced text faster. It is that you gave the model more to work with, and it rewarded you for it.

We explore this effect more fully in why voice changes the prompts you write. The short version: the medium shapes the message, and speaking produces a fuller message than typing.

How it feels day to day

The first time you dictate a prompt, it feels slightly strange — you are talking to a chat window. That passes within an hour. What it settles into is something closer to thinking out loud.

A typed prompt is a small writing task. You compose it, you edit it, you tidy it before sending. A spoken prompt is more like explaining a problem to a colleague. You start talking, the thought develops as you speak, and by the end you have laid out the whole situation. It is lower-friction and, oddly, lower-effort, because explaining something verbally is a skill you have had since childhood, while composing precise written instructions is not.

You do still read before you send. Dictated text wants a quick glance — a misheard word, a number that came out wrong. That review takes a couple of seconds and is worth it. But it is reviewing, not composing, and reviewing is the easy part.

Over a week, the pattern becomes plain. Your prompts get longer. Your conversations with the assistant get shorter, because each prompt does more. You stop dreading the long, careful prompt — the one with all the context — because it is no longer a typing job. You just say it.

Why the tool should work everywhere, not just in one app

ChatGPT and Claude are not one place. You use them in a browser tab, in a desktop app, sometimes in a mobile-style window, and AI assistants are also embedded inside editors, terminals, email clients, and note apps. If your dictation only worked inside one official app, you would be constantly bumping into walls.

What you want is dictation that behaves like part of the system — available in any text field, in any app, with no per-app integration. You hold a key, speak, release, and the text appears at the cursor, wherever the cursor happens to be. The browser tab with ChatGPT, the Claude desktop window, the assistant panel in your editor: all the same, because they are all just text fields.

That is the model Lispr uses. It is a small macOS app that lives in the menu bar with no window of its own. Hold the right Option key, speak, release; the transcribed text lands at your cursor in whatever app has focus. The round trip is around 200 milliseconds. Audio travels over an encrypted connection, is transcribed by the Whisper speech model, and is then discarded — nothing stored, nothing used to train anything. There is no account and no subscription; it is free in early access, and it is macOS-only.

The point of the system-wide design is exactly this case. You should be able to talk to your AI assistant the same way whether it is in a browser, an app, or buried inside another tool. The dictation should not know or care which.

The honest summary

Talking to ChatGPT and Claude instead of typing is not about speed for its own sake. It is about removing the small, constant pressure to keep prompts short — a pressure you did not know was shaping your prompts until it was gone. Speak the prompt and it comes out as the full thought: the context, the constraints, the intent. That fuller prompt is what gets you the better answer.

Try it on your next real question. Say the whole thing, the way you would explain it to a person. You will likely notice the response is closer to what you wanted, and you did not have to type a paragraph to get there.

For the broader developer angle, see voice prompting in Cursor and voice dictation for Claude Code.