The thought arrives complete and leaves in about eight seconds. The keyboard is not the right capture device for that.
You had the idea in the shower. You had it again on the walk to the kitchen. You sat down to write it and the cursor blinks at you for ninety seconds while you stare at the wall trying to reconstruct what was, twenty minutes ago, a fully formed paragraph.
This is not a discipline problem. The gap between thinking and typing is roughly 3 to 4 times in raw speed, and the cognitive cost of moving from "I have a thought" to "I have a thought, an open app, a cursor in the right place, and the working memory to hold the sentence while I type it" is enough to lose the thought entirely.
Voice dictation closes that gap. This post is a practical guide to using it for ADHD capture: the workflow, what to actually look for in a tool, and the failure modes nobody else talks about.
TL;DR
Why Voice Beats Typing for ADHD Capture
There are three things going on, and they compound.
Speed. Sustained speech runs at 130 to 150 words per minute. Average typing speeds sit around 40 to 60. For short bursts, fast typists close some of that gap. For a 400-word brain dump, voice is done in three minutes; typing is a ten-minute commitment.
Working memory load. Typing splits attention between the thought and the mechanical act of producing it: spelling, autocorrect, the red squiggly line, the urge to fix the typo three words back. Speech is a single stream. You think the sentence, you say the sentence, the sentence exists.
Ramp-up cost. This is the one that matters most for ADHD. The thought arrives. To type it, you need to find the right app, open a new note, place a cursor, and start typing before the thought decays. That sequence is maybe twelve seconds of pure friction, and twelve seconds is enough to lose it. A good dictation setup is one hotkey: press, talk, the words appear wherever you are.
The fix is not "type harder". The fix is to make the capture interface match the speed of the thought.
What "Good" Looks Like for an ADHD Dictation Setup
Most reviews of dictation tools optimise for accuracy. Accuracy matters, but it is not the only thing, and for ADHD capture it might not even be the most important.
Here is what to actually evaluate.
One hotkey, everywhere
The tool should work in any app. Not "in our app". Not "in a special chat window". Not "paste from our window into your app". Press a hotkey, talk, words appear at the cursor. Slack DM, Notion page, browser address bar, email composer, terminal, code editor, calendar event title. All of them.
If the tool routes you through its own UI, you have rebuilt the ramp-up problem.
Fast to start, fast to finish
The latency from "I press the key" to "I can start talking" should be effectively zero. Some tools have a 1 to 2 second warm-up where the first half of your sentence gets eaten. That is fine for an email reply. It is fatal for capture, because the first half of the sentence is the half you cannot reconstruct.
Same on the back end: when you stop, the text should appear immediately, not after a 4-second cloud round trip.
It handles your mess
You will say "um". You will say "wait, scratch that". You will start a sentence, abandon it, restart. You will go off on a tangent and then come back. A tool that requires you to dictate clean copy from the top is asking you to perform a skill ADHD specifically makes harder.
Look for: filler word removal, optional light reformatting (capitalisation, paragraph breaks, "new paragraph" commands), and a one-key undo if the AI gets clever in a way you do not want.
Custom vocabulary
If you are dictating notes about your job, your meds, your kids, your projects, or your codebase, those names are going to recur. A dictation tool with no custom vocabulary turns every recurring proper noun into a phonetic guess. "Vyvanse" becomes "vie vance". "Notion" becomes "notion" with a lowercase n in the middle of sentences. You should be able to add custom terms and not fix them by hand every single time.
It does not punish a bad day
Some days the audio is great. Some days you are at 60% capacity, talking through a mask of fatigue, sniffling, sentences trailing off. The tool should still be usable. If it requires studio-clean audio to function, it is not a real-world ADHD tool.
The Capture Workflow That Actually Sticks
A lot of advice in this space ends at "get a dictation tool". The actual workflow has three parts, and the dictation tool is only one of them.
1. A single, dumb capture surface
You need exactly one place that thoughts go when you do not have time to file them properly. Not a system of seven labelled inboxes. One bucket.
The most reliable version of this is a daily note in whatever app you actually open: Apple Notes, a today.md in Obsidian, a single Notion page, a Slack DM to yourself, a Telegram "Saved Messages" chat. The important part is not the app, it is that you never have to decide where the thing goes.
Hotkey-driven dictation into this single surface is the workflow. You think the thought, hit the hotkey, speak the thought, the thought is captured. You do not file it. Filing is a separate job for later, and most thoughts do not actually need filing.
2. Capture wide, edit narrow
When you dictate, dictate wide. Half-formed thoughts, tangents, asides, the recursive "and also" sentences your brain produces. Do not edit while capturing. You are doing one job: getting the thought out of your head before it leaves.
Editing is a separate, much slower mode. Do it later, in batch, when the working memory pressure is off. A lot of ADHD productivity advice fails because it tries to combine capture and editing into one act. They are different cognitive modes and they fight each other.
3. A regular sweep, not a perfect system
Once or twice a week, open the capture surface. Triage. Most things will be either obvious actions, obvious notes, or junk that already feels less important than it did when you said it. The 80% that is junk is fine to leave or delete. The 20% that matters can be moved into a project, a calendar, or a longer note.
The goal is not a perfectly organised second brain. The goal is to never lose the 20% of thoughts that matter because the 80% of junk thoughts blocked the pipe.
What's Wrong With the Standard Recommendations
If you search "voice dictation ADHD" you get a familiar list: Willow, Wispr Flow, Dragon, Otter, Google Docs voice typing, Apple and Windows built-in dictation. They are not all wrong, but they have specific problems for this use case.
The pattern, especially for the AI-powered tools optimised for ADHD: they are good, but they assume you are comfortable uploading every thought you ever have to a third party. For a tool that lives in your shower-thought capture pipeline, that is a higher bar than most people think about.
Where Voca Fits
Voca is a desktop dictation tool that runs on Mac, Windows, and Linux. The pitch for ADHD capture specifically:
One hotkey, anywhere. Press, talk, the words appear at your cursor. Slack, Notion, Apple Notes, Obsidian, browser bar, terminal, IDE. Same behaviour everywhere.
Local processing. Your audio is not stored. It is not uploaded to be used for training. The recording happens on your machine and the transcript is what gets inserted. If your brain dumps contain anything personal, medical, work-confidential, or otherwise not-for-the-internet, this is the default you want.
Cross-platform parity. If you have a Mac at home, a Windows machine at work, and a Linux laptop, the muscle memory carries. This sounds minor and is not: switching tools across machines is one of the things that breaks an ADHD capture habit faster than almost anything else.
Custom vocabulary. Add your project names, drug names, colleague names, codebase terms once. Stop fixing them by hand.
Filler word removal and light cleanup. The "um"s and the false starts get cleaned. You can turn it off if you want raw transcripts.
It is not the right tool for everyone. If you need full hands-free computer control for accessibility reasons, Talon on Linux or Dragon on Windows are still the deeper options. If you want the most polished consumer-grade AI rewriting and do not care about local processing, Wispr Flow is fine. Voca's lane is the "press, talk, captured, private, everywhere I work" lane.
A Sample Setup
Here is a minimal daily-capture stack that works for most people.
Capture surface: one daily note. Pick the app you actually open. Apple Notes is the lowest-friction option on Mac. On Windows, OneNote or a single Notion page. On Linux, an Obsidian vault with a daily-notes plugin, or just a today.md file with a hotkey to open it.
Dictation hotkey: bind Voca (or your tool of choice) to a key you can hit without looking. The default on most tools is a function key or a hold-to-talk modifier. Make sure it does not collide with anything you use heavily.
Workflow:
Thought arrives.
Hit the hotkey wherever you are.
Talk.
Release. Text appears.
Move on.
Weekly sweep: 15 minutes, once a week, in your calendar. Open the daily notes from the past week. Triage. Done.
That's it. There is no app stack, no Zettelkasten, no PARA method. Those systems work for some people, but the entry point is the capture, and most capture systems fail at the entry point.
Common Failure Modes (Honest Section)
Voice dictation is not magic. Things that will go wrong:
You will feel weird talking to your computer for the first week. Especially in shared spaces. This passes. Most people normalise within a few days. Headset mics help a lot, both for accuracy and for not feeling like you are broadcasting.
The first few sessions will produce ugly transcripts. You will see your own filler patterns written out, and they look worse on the page than they sound. This is information, not a failure. You learn quickly to slow down at the start of a thought and let the model catch the front of the sentence.
Technical jargon will get butchered until you train the vocabulary. Add terms as you notice them. Five minutes of vocabulary setup pays back for months.
Background noise matters more than you think. A loud cafe will degrade any tool. A quiet room with a decent mic gets you 95%+ accuracy on most modern Whisper-class models. AirPods are passable. A wired headset is better.
You will occasionally lose a thought because the hotkey misfired or the app stole focus. Build a habit of saying the thought once before you stop talking, like a verbal autosave. "OK so the idea is... [thought]... yeah, the idea is [thought]." Looks silly. Works.
FAQ
Does voice dictation actually help with ADHD?
It helps with one specific ADHD difficulty: the gap between having a thought and capturing it before working memory drops it. Speech is roughly 3 to 4 times faster than typing and has almost no ramp-up cost. That makes voice a better capture interface than a keyboard for most people with ADHD. It does not address focus, time blindness, task initiation, or any of the other ADHD-shaped problems. It is one tool, narrowly useful.
Will dictation tools train on my brain dumps?
It depends on the tool. Cloud-based tools (Willow, Wispr Flow, Aqua Voice, Otter, Google) upload your audio for processing, and their terms of service vary on retention and training. Read them. Local-first tools like Voca process audio on your device and do not store or transmit it. If your brain dumps contain personal or work-confidential content, this is worth caring about.
Is built-in Apple or Windows dictation good enough?
For occasional, short, low-stakes dictation, yes. For an ADHD capture workflow you use dozens of times a day, the built-in tools tend to be inconsistent on Windows, weak on custom vocabulary on both platforms, and slow enough that the first part of your sentence often gets dropped. Most people who try voice capture seriously upgrade to a dedicated tool within a couple of weeks.
What about Linux?
Most of the popular ADHD dictation recommendations skip Linux entirely. Voca runs on Linux at parity with Mac and Windows. Talon also supports Linux but is positioned for accessibility power users, with a steep learning curve. Beyond that, you are looking at open-source projects like Whisper-based scripts that you wire into your window manager yourself.
How do I stop the AI from "fixing" my words?
Most tools have a toggle for the cleanup pass. If the model keeps autocorrecting a name, a slang term, or a deliberately weird sentence into something else, add the term to your custom vocabulary and turn the aggressive rewrite mode off. Look for "raw transcript" or "no rewrite" mode in your tool's settings.
Does it work for code or technical writing?
Partially. Voice is excellent for prose: prompts, design docs, commit messages, PR descriptions, comments, README sections, emails, Slack messages. It is weaker for raw code, where you spend more time saying punctuation than words. Most developers who dictate use voice for the prose around the code and keep typing the code itself. There is a separate post on dictating in Cursor and Claude Code that goes deeper on this.
Will I still need to type?
Yes. Voice is a capture and prose tool. You will keep typing for code, for editing, for short fields, for anything where speaking out loud is awkward. The shift is from typing as the default to typing as one of two inputs, picked based on what you are doing.
The Point
The point of voice dictation for ADHD is not to be 4x faster at writing emails, although it will do that. The point is to close the gap between thought and capture so that the 20% of thoughts that actually matter do not get lost to friction.
Pick the lowest-friction capture surface you have, point a dictation hotkey at it, and stop trying to remember the thing you thought of in the shower.
Voca runs locally on Mac, Windows, and Linux. Your audio is not stored or used for training. Try it at usevoca.net.