A complete reference for using iOS Voice Control to operate the Claude and Gemini iOS apps hands-free. Every command, every workflow, every gotcha — organized to build real intuition.
Voice Control isn't just speech-to-text. It's a persistent, always-on layer that sits between you and every pixel on screen. It can see every UI element, tap any button, scroll any view, and type into any field. When active, your voice becomes a mouse, keyboard, and touchscreen combined.
The key mental model: Imagine an invisible assistant holding your phone, watching the screen. You tell them what to tap, what to type, when to scroll. They do it instantly. That's Voice Control — it maps your words to touch gestures and keyboard input in real time.
There are two distinct modes your voice operates in:
When no text field is focused, everything you say is interpreted as a command. "Scroll down," "tap Send," "go home" — these are actions, not text. Nothing gets typed.
→ Your voice controls the phone
When a text field is focused, your speech is transcribed as text AND commands still work. Say "hello comma how are you" and it types. Say "tap Send" and it taps. The system figures out which is which.
→ Your voice types AND controls
When Voice Control is on, you'll see a blue microphone icon in the status bar. It's always processing your audio. There's no push-to-talk — it's continuous. This means:
No activation word needed. Just speak.
Continuous listening uses power. ~10-15% extra per day.
Background conversations can trigger accidental commands.
Settings → Accessibility → Voice Control
Toggle it on. iOS will download the speech model (~80MB) on first activation.
Settings → Accessibility → Accessibility Shortcut
Check "Voice Control." Now triple-click the side button to toggle it on/off instantly.
Voice Control → Language
Set to your preferred language. English (US) has the best recognition accuracy.
Say "show commands" at any time to see what's available. This is your safety net — use it liberally while learning.
Moving around the app without touching the screen. These are the commands that replace your thumb.
| Say This | What Happens | Usage |
|---|---|---|
| "Tap [label]" | Taps any visible button or element by its label e.g., "tap Send", "tap New Chat", "tap Cancel" |
●●● |
| "Tap text field" | Focuses the message input box This is your #1 command — starts every message |
●●● |
| "Show numbers" | Labels every tappable element with a number Say "tap 4" to tap element #4. Essential fallback when labels don't work. |
●●● |
| "Show grid" | Overlays a numbered grid on screen Say a grid number to zoom, then say a sub-number. For precise targeting. |
●●○ |
| "Scroll down" / "up" | Scrolls the current view Works on long AI responses. Repeat to keep scrolling. |
●●● |
| "Go back" | Navigates back (like swipe from left edge) | ●●○ |
| "Go home" | Returns to the home screen | ●○○ |
| "Open [app name]" | Launches an app "Open Claude", "Open Gemini" |
●●○ |
| "Long press [label]" | Long-presses an element Useful for context menus — copy message, share, etc. |
●○○ |
| "Swipe left" / "right" | Performs a swipe gesture | ●○○ |
| "Show names" | Shows the accessibility label of every element Use this to discover what to say for "tap [label]" |
●●○ |
| "Hide [overlay]" | Hides numbers, grid, or names "Hide numbers", "hide grid" |
●●○ |
What to say when the text field is active and you're composing a message.
| Say This | What Happens | Usage |
|---|---|---|
| (just talk) | Transcribes your speech as text When a text field is focused, natural speech becomes text automatically |
●●● |
| "Period" / "comma" | Inserts punctuation inline "Hello period How are you comma Claude question mark" |
●●● |
| "New line" | Inserts a line break (Return key) | ●●● |
| "New paragraph" | Inserts a double line break | ●●○ |
| "Question mark" | Inserts ? | ●●● |
| "Exclamation point" | Inserts ! | ●●○ |
| "Open quote" / "close quote" | Inserts " and " Useful for quoting text in prompts |
●●○ |
| "Colon" / "semicolon" | Inserts : or ; | ●●○ |
| "Dash" / "hyphen" | Inserts — or - | ●○○ |
| "Open paren" / "close paren" | Inserts ( and ) | ●○○ |
| "At sign" | Inserts @ | ●○○ |
| "Hashtag" | Inserts # | ●○○ |
| "Ampersand" | Inserts & | ●○○ |
| "Cap [word]" | Capitalizes the next word "cap claude" → "Claude" |
●●○ |
| "All caps [word]" | Types word in all caps "all caps important" → "IMPORTANT" |
●○○ |
| "Numeral [number]" | Types the digit instead of the word "numeral 5" → "5" instead of "five" |
●●○ |
| "No space [word]" | Types next word without a space before it Useful for URLs, code snippets |
●○○ |
Fixing mistakes without touching the keyboard. These make Voice Control feel like a real input method.
| Say This | What Happens | Usage |
|---|---|---|
| "Delete that" | Deletes the last dictated phrase The single most useful correction command |
●●● |
| "Undo that" | Undoes the last action | ●●● |
| "Delete [word/phrase]" | Finds and deletes specific text "Delete hello world" removes that exact text |
●●○ |
| "Select that" | Selects the last dictated phrase | ●●○ |
| "Select all" | Selects all text in the field | ●●○ |
| "Select [word] to [word]" | Selects a range of text | ●○○ |
| "Replace [x] with [y]" | Find and replace "Replace clod with Claude" — the killer command for fixing recognition errors |
●●● |
| "Correct [word]" | Shows correction suggestions for a word Offers alternatives when speech recognition guessed wrong |
●●○ |
| "Correct that" | Shows corrections for the last phrase | ●●○ |
| "Clear all" | Erases everything in the text field | ●●○ |
| "Move up" / "move down" | Moves cursor up/down a line in multiline text | ●○○ |
| "Cut that" / "Copy that" | Cuts or copies selected text | ●○○ |
| "Paste" | Pastes clipboard content | ●○○ |
Controlling Voice Control itself — the commands that help you learn and manage the system.
| Say This | What Happens | Usage |
|---|---|---|
| "Show commands" | Displays all available commands for current context Your cheat sheet. Use early and often. |
●●● |
| "Go to sleep" | Pauses Voice Control (stops listening) Saves battery, prevents accidental input |
●●● |
| "Wake up" | Resumes Voice Control from sleep | ●●● |
| "What can I say?" | Context-sensitive help — shows relevant commands | ●●○ |
Anthropic · claude.ai
The Claude app's send button may show as an arrow icon. If "tap Send" doesn't work, use "show numbers" to find the right number. The button is usually at the bottom right of the text input.
To switch Claude projects: "show numbers" → tap the hamburger/menu icon → find your project in the sidebar. The sidebar elements all get numbered too.
Google · gemini.google.com
Gemini's text field label is usually "Message Gemini." You can try "tap Message Gemini" directly. If the app UI updates, always fall back to "show numbers."
To attach an image: "show numbers" → tap the attachment/camera button → navigate the photo picker with "show numbers" and "tap [#]."
Navigation without touching the screen
Don't try to tap search results by name — they're often truncated or have accessibility labels that don't match what you see. "Show numbers" → "Tap [#]" is more reliable every time, especially while driving.
If multiple route options appear after searching, use "show numbers" to pick your preferred route, then "tap Start."
Dictate prompts, send, and listen to responses via TTS
The floating controller lets you adjust speed — 1.5× or 2× is great for Claude responses. You can also two-finger swipe down from the top of the screen as an alternative gesture instead of the voice command. To stop mid-read, tap the ✕ on the controller or use "show numbers" to tap it hands-free.
Speak Screen auto-scrolls as it reads, but it can only start from content currently loaded on screen. For very long Claude responses, "Tap Scroll" first to jump to the bottom and load the full response, then trigger Speak Screen — it will read from the top down.
The complete hands-free cycle: "Tap Text" → dictate → "Tap Send" → wait for response → "Tap Scroll" → "Speak Screen" to listen → ✕ to stop → repeat for follow-ups. Everything stays in the written record for later.
App updates can change element labels. If "Tap Text" stops working, use "show numbers" → tap the input field's number. That always works regardless of app version.
If any command doesn't work, the universal fallback is always the same:
"Show numbers" → say the number. Works on any screen, any app, every time. If numbers don't appear on the element you need, use "show grid" → say a grid number to zoom in → repeat until you can target the exact spot.
This two-step fallback should become automatic. It's the escape hatch that means you never need to touch the phone.
| Feature | Voice Control | Siri Dictation | Gboard Voice | Gemini Live Voice |
|---|---|---|---|---|
| System-wide | YES | YES | KEYBOARD ONLY | GEMINI ONLY |
| Can tap buttons | YES | NO | NO | NO |
| Can send messages | YES "tap Send" | NO manual tap | NO manual tap | YES auto-send |
| Always listening | YES | NO times out | NO times out | YES in session |
| Text editing | FULL replace, delete, select | BASIC | BASIC | N/A |
| Works in Claude app | YES | YES | IF INSTALLED | NO |
| Battery impact | Medium-High | Low | Low | Medium |
| 100% hands-free | YES | NO | NO | YES Gemini only |
Bottom line: Voice Control is the only method that gives you full hands-free control of any app. The others do dictation only — you still need your hands to navigate and send. Voice Control is the full package but trades battery life and noise sensitivity for that power.
The button's accessibility label might not be "Send." Use "show numbers" to find the actual number for the send button, then "tap [#]". This is the most common issue people hit. App updates can change labels.
When a text field is focused, Voice Control has to guess whether you're dictating or commanding. If "tap Send" gets typed instead of executed, try pausing slightly before the command, or say "tap Send" with clear emphasis. Some people find that a brief pause helps the system distinguish commands from dictation.
Voice Control processes audio continuously. Use "go to sleep" when you're not actively using it, or toggle via the triple-click shortcut. Don't leave it running all day unless you need it.
In a meeting or public space, nearby voices can trigger commands. Use "go to sleep" or toggle off with the triple-click shortcut. There's no "hey Siri"-style wake word — it listens to everyone equally.
Speech recognition sometimes hears "clod," "clawed," or "cloud" instead of "Claude." Use "replace clod with Claude" or "correct clod" to fix it. This is just a reality of the name being unusual.
When you say "show numbers," the overlay can cover text you're trying to read. Read first, then say "show numbers" when ready to interact. Say "hide numbers" to clear it.
Voice Control isn't faster than touch typing at first. It takes 2-3 sessions to build fluency. The inflection point is when "show numbers" + "tap [#]" becomes automatic — that's when you stop needing to think about it.
Voice Control and Siri are mutually exclusive listeners. When Voice Control is on, "Hey Siri" won't work. Toggle Voice Control off first if you need Siri.
Settings → Accessibility → Voice Control → Customize Commands → Create New Command
You can map any phrase to an action: insert text, run a shortcut, or perform a gesture sequence.
Custom command that: opens Claude app → taps text field. Saves two steps every time you start a new query.
Custom command that: taps the new chat button → focuses text field. One phrase instead of "show numbers" → "tap 3" → "tap text field."
Custom command mapped to tap the send button coordinates. More reliable than "tap Send" when the button label doesn't match.
Insert Text action that pastes a prompt prefix like "Explain this concisely:" — then you just dictate the rest.
Turn on Voice Control. Say "show commands" — just read what's available. Say "show numbers" on your home screen. Get a feel for how iOS labels everything. Say "hide numbers."
Say "open Claude" (or "open Gemini"). Use "show numbers" and "tap [#]" to navigate around. Open a chat, go back, open another. Don't type yet — just navigate.
Say "tap text field." Speak a simple message: "Hello comma this is a test period." Say "tap Send." If "tap Send" doesn't work, use "show numbers" to find the send button. This is the core loop — repeat it 3 times.
Deliberately dictate something, then fix it: say "delete that", say "undo that", say "replace [wrong word] with [right word]." Get comfortable editing without touching the screen.
Ask Claude or Gemini something you actually want to know. Use the full loop: dictate → send → scroll to read → follow up. If it feels slow, that's normal. Speed comes from the commands becoming automatic, which happens by the 3rd session.
The fluency threshold: Most people feel awkward for the first 1-2 sessions. By session 3, "show numbers" + "tap [#]" becomes muscle memory (voice memory?). By session 5, you stop thinking about the commands and start thinking about what you want to say. That's when Voice Control actually becomes faster than fumbling with your phone for certain use cases — cooking, driving, exercise, walking.