Short answer: Raycast has no built-in dictation, but any text field in it accepts dictated text. Use a system-wide voice-to-text tool like Voice Keyboard Pro: click into the AI chat, a note, or an extension form, hold your dictation hotkey, speak, and release. Just make sure the two apps' hotkeys don't collide.
Raycast people are keyboard people. The whole pitch of the app is that reaching for the mouse is a waste of motion: summon a bar, type a few characters, and the thing you wanted happens. So suggesting you talk to your Mac instead can sound like heresy.
But look at what Raycast has become. It started as a launcher, where the only text you enter is a command name. Now it holds AI chat conversations, floating notes, snippets, and extension forms that create Jira tickets, GitHub issues, and to-dos. Those are not command names. Those are sentences and paragraphs, and typing sentences is the slow part: most adults type around 40 words per minute while everyone speaks at 130-150. The launcher philosophy, "reduce the friction between intent and result," points directly at dictating the wordy parts.
Raycast does not ship its own dictation, and macOS's built-in dictation is clumsy inside a transient floating panel. Here is the setup that works, surface by surface, plus the one configuration mistake that will drive you crazy if you skip this section.
The setup: system-wide dictation that types at your cursor
The trick is not Raycast-specific, which is exactly why it works. A system-wide dictation tool types text wherever your cursor is, and Raycast's panels are ordinary macOS text fields as far as text input is concerned. Voice Keyboard Pro lives in your menu bar; you hold a hotkey, speak, release, and the transcribed text appears at the cursor. If the cursor is in Raycast's AI chat box, that is where the words land. No extension to install, no integration to configure, no permission for Raycast to grant.
The flow, concretely:
- Summon Raycast with its hotkey (⌥ Space by default; many people rebind it to ⌘ Space).
- Open the surface you want: AI chat, a note, a snippet form, an extension's create-issue form.
- Click or tab into the text field so the cursor is blinking there.
- Hold your Voice Keyboard Pro hotkey, say the whole thought, release.
- The text appears. Edit if needed, then hit Return or ⌘ Return as usual.
The hold-to-talk model matters more in Raycast than almost anywhere else. Raycast panels are transient: they appear, do one job, and vanish. A dictation session that you start and stop with your own thumb fits that rhythm; there is no open-ended listening session to manage, and you decide exactly when the utterance ends by releasing the key.
The one gotcha: hotkey collisions
You now have two apps that live and die by global hotkeys. If you assign dictation to a key combination Raycast is already using (its main hotkey, a Quicklink hotkey, or a per-extension hotkey), one of them will silently lose. The symptom is maddening: dictation "randomly" not starting, or Raycast panels appearing when you meant to talk.
Pick a dictation hotkey that nothing else claims. A modifier-heavy combination, or a key you never use, such as right ⌘ or a function key, works well. Then test both: summon Raycast, dismiss it, and hold your dictation key in a Notes window. If both behave, you are set for good. This is the same advice we give for editor-heavy setups in dictating in Zed, where custom keybindings are similarly dense.
Surface 1: Raycast AI chat, where dictation pays off most
If you use Raycast AI (part of Raycast's Pro subscription), the chat box is the single best place to point your microphone. The reason is a pattern we have seen across every AI assistant: typed prompts are starved prompts. When context costs you a minute of typing, you skip the context, and the assistant gives you a generic answer that you then have to correct in three follow-up messages.
Spoken prompts flip the economics. Compare what people type into an AI chat with what they would say. Typed: "regex for dates." Spoken in the same ten seconds: "Write me a regular expression that matches dates in the format day slash month slash year, where the year can be two or four digits, and it needs to work in JavaScript. I'll be using it to validate a form field." The second prompt gets the right answer on the first try, because saying three sentences costs almost nothing.
Two habits make dictated prompts land better, and they are the same ones we recommend for dictating ChatGPT prompts on a Mac:
- Front-load the goal. Start with what you want ("Summarize this in three bullets…"), then give context. Rambling context-first prompts bury the instruction.
- Say the format you want. "As a numbered list," "as a one-paragraph summary," "as a table." Format instructions are cheap to speak and transform the output.
Follow-ups are where it compounds. A chat is a conversation, and conversations are a speech-native format. Holding a key and saying "That's close, but make it more concise and drop the second example" beats typing it every single time.
Surface 2: Raycast Notes
Raycast Notes gives you a floating scratchpad summoned from anywhere, which makes it the natural home for the thought that arrives while you are doing something else. That is precisely the thought most likely to die while you type it.
The voice version of the capture loop: hotkey, note open, hold, talk, release, dismiss. The whole thing takes about as long as saying the thought out loud, because that is nearly all you did. Meeting follow-ups, a bug you noticed but cannot chase right now, the sentence you want to open tomorrow's email with: speak them into the note in full sentences rather than the three-word fragments you would type, and future-you gets context instead of a riddle.
One formatting tip: if you use dictation's spoken commands, "new paragraph" keeps a brain-dump note readable. Bullet syntax and markdown symbols are faster typed; speak the prose, type the structure.
Surface 3: snippets and extension forms
Snippets. Raycast snippets expand typed keywords into saved text. Dictation helps when you are creating them: the snippet body (a support reply, a standard email paragraph, a code review preamble) is prose, and prose is faster spoken. The keyword itself you should type; it is short, deliberate, and usually not a real word.
Extension forms. This is the sleeper use case. Raycast extensions for GitHub, Jira, Linear, Todoist, and dozens of other tools put a create form one keystroke away: issue title, description, comment. The title-and-description fields are exactly the writing that engineers habitually leave at one line because typing more feels like overhead. Dictate the description while the problem is fresh: what you expected, what happened, how to reproduce it. Three spoken sentences produce the bug report your teammates wish you always wrote. It is the same argument we make for dictating in Xcode: the prose around code deserves the same speed as the code itself.
What to type instead (the honest boundary)
Dictation is a sentence engine. Raycast is full of things that are not sentences, and speaking them fights the tool:
- Launcher queries. Fuzzy-matching three typed characters ("saf" for Safari) beats saying anything. Keep launching with your fingers.
- Snippet keywords, aliases, and hotkey assignments. Short, symbolic, deliberate. Type them.
- URLs, tokens, file paths, and code. Symbols and casing make these poor dictation targets. The prose around them is the dictation target.
- Anything secret. Passwords, API keys, and door codes should never be spoken aloud, in Raycast or anywhere else, both because dictation formats them badly and because the person at the next desk has ears.
The rule that covers all of it: speak the sentences, type the syntax.
Make it accurate: your vocabulary is not in the dictionary
Raycast users' dictation is dense with words no general-purpose recognizer loves: extension names, repo names, teammates' names, product code names, camelCase identifiers. Stock dictation mangles these forever, because it has no idea your world contains a service called "Quokka-Sync."
Voice Keyboard Pro's Smart Vocabulary fixes this class of error at the source. It is a personal dictionary with replacement rules: add the proper nouns and jargon you actually use, and the transcription respects them; add replacement rules for terms you say often ("open paren jira ticket" habits need not apply; say "the sync service" and have it come out as the real name if you like). Five minutes of adding your top twenty terms removes the majority of corrections you would otherwise make daily.
Beyond vocabulary, the accuracy basics apply: speak in full phrases rather than word-at-a-time (context is what disambiguates), keep the mic reasonably close, and if you dictate into a hot open-plan office, read our guide to dictating in background noise, because other people's speech is the one noise type that genuinely competes with yours.
A note on privacy, since launcher users ask
Raycast's audience skews technical, and technical users rightly ask where dictated audio goes. Voice Keyboard Pro's answer is short: our servers store operational pings only, no audio and no transcript content. What you dictate into your AI chats, notes, and issue trackers is your business. If you are evaluating any dictation tool, that is the line to look for in the privacy policy before you adopt it at work, and your employer's policy on external tools still applies to whatever you dictate into company systems.
What a voice-first Raycast hour looks like
To make this concrete, here is how the pieces combine in an ordinary stretch of work. You finish a call and summon Raycast Notes: hold the hotkey and speak the three decisions and two action items while they are still fresh, in full sentences, then dismiss the note. One of the action items is a bug, so you open the issue-tracker extension, type the title (short, deliberate, typed is fine), then dictate the description: what you saw, what you expected, how to reproduce it. Thirty seconds of speaking, and it is a real bug report instead of a four-word placeholder.
Next you need a code review paragraph you write five times a week, so you open snippet creation, type the keyword, and dictate the body once; from now on it expands from four typed characters. Finally you hit a question you would normally under-type into the AI chat. Instead you hold the key and give it the full context in three spoken sentences, and the first answer is the usable one.
Nothing in that hour required a new integration or a change to how Raycast works. The only thing that changed is that every field wanting more than a few words got spoken into instead of typed into, and each of those moments got several times faster.
Frequently asked questions
Does Raycast have built-in voice input?
Raycast does not include its own general dictation for text fields. Whatever the app adds over time, a system-wide dictation tool covers every text field in Raycast today, and also works in the browser tab, editor, and chat app you switch to next, which a single-app feature never will.
Can I use macOS built-in dictation inside Raycast?
You can try: focus a Raycast text field and press the macOS dictation shortcut. It sometimes works, but stock dictation's session model is awkward inside a transient panel, it pauses out after silence, and its accuracy on technical vocabulary is weak with no way to teach it your terms. A hold-to-talk tool with a personal dictionary fits the launcher workflow much better.
Will dictation work in Raycast's search bar too?
Yes, text lands wherever the cursor is, including the root search. It is just rarely worth it: launcher queries are two or three characters. Save your voice for the fields that want sentences.
What about dictating on the go?
Raycast is Mac-side, but the thoughts you capture into it are not. Voice Keyboard Pro's iPhone keyboard has a built-in mic button that dictates into any iOS app, with Voice Edit to fix text by speaking the change, so the idea you catch on a walk can be spoken into your notes app and meet you back at your Mac.
The bottom line
Raycast compressed everything about using a Mac except the writing. The AI prompts, the notes, the issue descriptions, and the snippet bodies are still produced at typing speed, and that is the slowest thing left in the workflow. Point a hold-to-talk dictation hotkey at those fields and the launcher philosophy finally applies to your words too: intent to result, minus the friction. Voice Keyboard Pro has a free tier; set the hotkey, avoid the collision, and dictate your next AI prompt instead of typing it.