Messages and conversations: attachments, dictation, voice chat
The official design positioning for Grok Bot conversations is one sentence: like messaging a teammate. Requests can be natural, but the result and the boundaries must be made clear—what it may change, and where it must stop and wait for you. In the previous lesson you dispatched your first task and watched it work on the cloud computer; this lesson covers the act of "talking to it" itself: typing and attachments, dictation, voice chat, and how to interject or call a halt while it works.
The whole thing takes about 10 minutes, and there is not a single command to run—every action is "talking to it."
What you'll be able to do after this lesson
- Hand over all the materials in one message: paste text, links, and images, attach local files, and use
@to mention a Bot, group, routine, or app - Use dictation instead of typing:
Cmd/Ctrl+Don desktop, Start dictation on mobile—review and edit after speaking, then send - Start a voice chat: the waveform button, Mute, Show transcript, and read the full transcript after hanging up
- Send follow-up instructions to change direction anytime while the Bot works; to make it stop immediately, one line of Stop now
- Review outgoing drafts the Bot writes for you: Send, Discard, or turn off drafts for a single Bot
Your current pain points
- You can only type back and forth, pasting materials as screenshots, and the Bot keeps asking "where is the source file"
- Speaking is clearly faster than typing, but you still type word by word; on the move, there's simply no way to assign work
- You don't dare send messages while it works in the background, afraid of interrupting; you don't know how to call a stop, so you can only wait
When to use this technique
When you need to:
- Hand text, links, images, and local files to the Bot together with your instructions
- Speak faster than you type—when your hands are busy or you're on the move, use dictation or voice chat
- Change your mind mid-task, add requirements, or make it stop immediately
And you don't want to:
- Switch back and forth between windows, copying and pasting piece by piece
- Wait until it finishes the current task before you can get a word in
- Let the emails and Slack messages it sends on your behalf go out without your review
🎒 Preparation before starting
- Completed 1.5 First Bot: you have at least one Bot on hand and have assigned a task once—every action in this lesson starts with "talking to it"
- Desktop app is logged in: this lesson's shortcuts and other entry points are based on the desktop version; voice chat is supported on both desktop and mobile apps
- You want to use dictation and voice chat: connect a microphone and allow Grok Bot access on first use
- A cup of water ☕—there are no commands to run in this lesson; every action is "talking to it"
Core idea
Grok Bot conversations rest on two underlying design principles; once you understand them, everything else is gestures:
- Like messaging a teammate. Keep requests natural, without imitating clicks step by step; but two things must be explicit—the result you want, and where the boundaries lie (what it may change, and which actions require it to stop and wait for you).
- The conversation stream is the workbench. The transcript is not just a message stream: tool activity, operations on the cloud computer, files created, its questions, approval requests, and voice memos all appear on the same timeline. You don't need to go elsewhere to "check progress."
Three input options—switch by scenario:
| Method | When it fits | How the Bot responds |
|---|---|---|
| Typing + paste/attachments | Complex instructions, sharing links and files | Text replies |
| Dictation (voice typing) | Speaking is faster than typing, and you want to edit afterwards | Speech becomes text, you edit and send; the Bot replies in text |
| Voice chat | Back-and-forth questions, chatting while walking | Speak up and it listens until you hang up |
There is a fourth timing: you can speak at any time while it works. Your new messages take priority over background work and can directly change the direction of the current turn—this is the biggest difference between Grok Bot conversations and ordinary chat tools.
Follow along
Step 1: Send your first Grok Bot conversation message: typing, pasting, and attachments
Why: everything in Grok Bot conversations starts in the composer (the input box); hand over all materials at once, and the Bot won't have to ask "where's the source."
Open a Bot from the sidebar and type a message in the composer. A single message can include all of these:
- Paste text, links, images——paste screenshots and web addresses directly
- Attach local files
@Mention a Bot, group, routine, or app; call out a saved skill by name- Use Reply to respond to a specific message; use an emoji to react
Send: Enter; new line: Shift+Enter; pressing Cmd/Ctrl+Enter anywhere in the interface also sends.
You should see: after the message is sent, the transcript will show not only your conversation but also tool activity, cloud computer operations, created files, its questions and approval requests—one after another, all on the same timeline.
💡 How much can attachments carry?
The limits on the number and size of attachments per message (different for images, documents, and videos) are fully listed in the next lesson 2.2 Files and Reviewable Results.
- Pasted a piece of text or an image, and attached a local file
Step 2: Turn speech into text with dictation: Start voice input or Cmd/Ctrl+D
Why: speaking is faster than typing, and the output of dictation is an "editable text message"—you review it, edit it, and send it as usual; the Bot replies in text, so nothing gets misheard with no way to fix it.
Desktop: select Start voice input in the composer, or press Cmd/Ctrl+D while the input box is focused. You can also hold the shortcut and speak as you press.
Mobile: on iPhone and Android, use Start dictation in the composer.
You should see: what you say turns into text in the composer in real time. Check, edit, then send as usual; the Bot replies in text just like with any message.
⚠️ Stop dictation before starting a voice chat
The official guidance is explicit: stop dictation before starting a voice chat—both features use the microphone, so don't run them at the same time.
- Dictated and sent a message using
Cmd/Ctrl+D(or Start dictation on mobile)
Step 3: Start a voice chat: the waveform button when the composer is empty
Why: a voice chat is a real-time call—the Bot speaks out loud and keeps listening to you until you hang up. Great for back-and-forth follow-ups without waiting for it to type sentence by sentence.
- Open a chat with a Bot, keeping the composer empty
- Select Start voice chat—the waveform button in the composer
- When your computer or phone asks, allow microphone access
You should see: the desktop shows Connecting…, mobile shows Calling…, then the Bot starts speaking. During the call, the app shows Voice chat, and the desktop sidebar also shows Voice chat in progress.
💡 Where did the button go?
Start voice chat only appears when the composer is empty. Can't see it even with an empty input box? Update the Grok Bot (see the pitfall reminder below). Also, voice chat has no keyboard shortcut—you can only click the button.
Step 4: Stay in control during the call: Mute, Show transcript, and hanging up
Why: real-time conversation also needs control—mute, view the transcript, adjust the voice, hang up; it's all on the call interface, no need to leave the call to find settings.
During a call you can:
- Mute / Unmute your microphone
- Show transcript to read the text transcript while listening
- When on a call with your own Bot, Show settings lets you change Voice (voice), Speed (speed), Language (language); choose Auto-detect for language and it will auto-detect based on what you say
- Hang up to end the call
The call doesn't lock you into the current chat: you can open other chats and do other things; on desktop, select Return to the voice chat to get back to the call at any time.
You should see: after hanging up, a Voice chat card appears in the conversation, noting the call duration; open it to read the full transcript. If work was left unfinished during the call, the Bot may follow up in the chat with the next step or the result; it might ask How was the call?—select Good or Bad to give feedback.
- Completed a call, and after hanging up opened the Voice chat card and read the transcript
Step 5: Send follow-up instructions while the Bot works; say Stop now to stop it
Why: when the Bot is working on the cloud computer it is not in "do not disturb" mode—your direct messages take priority over background work, and you can jump in to change direction at any time.
- Priority changes: send a new message directly, and the Bot can adjust the current turn based on it
- To end immediately: just send Stop now
You should see: the Bot changes direction after receiving the new message; it stops after receiving Stop now.
⚠️ Stop now is not an undo button
The official documentation is explicit: Stop now does not undo actions the Bot has already completed—it only prevents subsequent actions from happening. So set boundaries in advance: for which actions require approval first, see 3.1 Approvals and Auto Review.
Step 6: Review outgoing drafts: Needs approval and Disable drafts
Why: when the Bot writes an email or Slack message for you, by default it gives you a draft before sending—reviewing outgoing content is your final gatekeeping step.
- The draft card shows Needs approval: check the recipients and body, then choose Send to send it (on desktop Slack drafts the button is Send message), or Discard to discard it
- To have a Bot skip drafts and send directly from now on: open the menu on one of its drafts and choose Disable drafts for Bot—after that it sends directly, but it is still subject to Auto Review
- To get drafts back: choose Needs approval on the draft card, or just tell the Bot
You should see: after Send the message goes out and the draft card disappears; after Discard the draft is discarded.
💡 Three boundaries of drafts
- Disable drafts only takes effect for this one Bot, and only its owner can see it; this option is not offered in group chats (email/Slack drafts never appear in group chats anyway)
- Emails sent by a routine always wait as drafts: any email sent from the Bot's dedicated email address that you didn't explicitly request in the chat is always held as a draft waiting for your Send—see 2.6 Bot dedicated email address
- When the settings fail to save, the card shows
Couldn't save. Try again.—choose Retry to try again
Checkpoint ✅
- Delivered full materials in one message: pasted text/links/images and attached local files
- Used
Cmd/Ctrl+D(desktop) or Start dictation (mobile) to dictate and send a message - Completed a voice chat: started it with the waveform button, used Mute or Show transcript, and read the Voice chat card after hanging up
- Sent follow-up new instructions while the Bot was working, and also sent a Stop now
- Handled a Needs approval draft card (Send or Discard)
Pitfall reminders
❌ Can't see Start voice chat even with an empty composer
Symptom: the input box is clearly empty, but the waveform button just won't appear.
Cause: the button only shows when the composer is empty; officially, if it's empty and you still don't see it, your Grok Bot version is too old.
What to do: update your Grok Bot to the latest version—see Versions and updates for the current version and how to update. Also, voice chat has no keyboard shortcut, so don't look for one on the keyboard.
❌ Voice chat failed（Connection lost / No connection / The call ended unexpectedly）
Symptom: the call drops mid-conversation, the call ends, and an error pops up.
Cause: the network connection was interrupted. Grok Bot will first try to reconnect automatically; brief disconnections usually recover without any action on your part.
What to do: choose Dismiss to close the prompt, check your network connection, then tap Start voice chat again. For more error lookups, see Error quick reference.
❌ Microphone denied or not found
Symptom: a prompt says microphone access was denied, or that no microphone can be found.
What to do:
- Denied → allow Grok Bot to use the microphone in system settings
- Not found → plug in a microphone and try again
- To specify which microphone to use on desktop: open Settings and choose it under Microphone
❌ Wanting to voice chat in a group chat, or open two calls at once
Symptom: there's no voice chat entry point in group chats; starting one with another Bot while a call is already active pops up a confirmation dialog.
Cause: voice chat supports only one Bot at a time and isn't supported in group chats; there can only be one call at a time.
What to do:
- To chat with someone, open that Bot's direct chat and start it there
- To switch to a different Bot mid-call, choose End and Chat in the End the current voice chat? confirmation dialog to end the current call and start a new one
- On desktop you can also voice chat with a shared Bot a colleague shared with you—start it in your chat with it; see 3.5 Multi-Bot collaboration
Advanced tips (optional)
Assign tasks by voice on mobile. The iPhone/Android composer has Start dictation; Siri, Spotlight, and the Shortcuts app also offer four ready-made commands: List Bots, Send Message, Start Dictation, Start Voice Chat—saying "Send a message in Grok Bot" to Siri triggers it.
While dictation is in progress, the lock screen and the Dynamic Island Live Activity show the progress with cancel and send buttons; iPhones with an Action Button can bind the button to Grok Bot > Start Voice Chat, and pressing it again during a call hangs up. See 1.6 Mobile for details.
Listen to a Bot's voice memo. A Bot can reply in a conversation with a voice memo: Play voice memo to play, Pause voice memo to pause, and expanding the memo lets you read its transcript—so even when it's not convenient to turn on the speaker, you can still "listen" to everything it said.
Make efficient conversations a habit. The official collaboration guide gives 6 rules: state the result first, link to sources of fact, explain what is changeable and what needs approval, ask for evidence, explicitly correct persistent preferences, and switch to a different Bot for long-term tasks. Daily usage guide has a comparison table, and the following lessons in this stage will put each into practice.
Lesson summary
✅ Grok Bot conversations at their core are "like messaging a teammate": natural requests, explicit results and boundaries.
✅ Hand over materials all at once: paste text, links, and images, attach local files, and use @ mentions, Reply, and emoji reactions in a single message.
✅ Getting started with dictation and voice chat: Cmd/Ctrl+D dictates speech into text; the waveform button starts a call, and after hanging up the Voice chat card has the full transcript.
✅ Interrupting and stopping: new messages take priority while the Bot is working and can change direction; Stop now stops immediately but does not undo completed actions.
✅ Outbound content is gated: Needs approval draft cards have Send / Discard; routine emails always wait as drafts.
Key entry points recap:
| Action | Entry point | Notes |
|---|---|---|
| Send / newline | Enter / Shift+Enter | Cmd/Ctrl+Enter Send from anywhere |
| Dictation | Start voice input or Cmd/Ctrl+D | On mobile it is Start dictation; stop dictation before starting voice chat |
| Voice chat | Start voice chat when the composer is empty (waveform button) | No shortcut; one Bot at a time, not supported in group chats |
| Call controls | Mute / Show transcript / Show settings / Hang up | Show settings is only available for your own Bots; language can be Auto-detect |
| Follow-up / stop | Send a new message directly / send Stop now | New messages take priority over background work; Stop now does not undo completed actions |
| Outbound drafts | Send / Discard; disable with Disable drafts for Bot | Applies only to that Bot; routine emails always wait as drafts |
Next lesson preview
With messaging running smoothly, the next step is to complete the loop of "getting materials in and results out". Next lesson: 2.2 Files and reviewable results.
You will learn:
- How many attachments each message can carry: a complete list of count and size limits for images, documents, and videos
- After connecting Google Drive / OneDrive, how files transfer between the cloud drive and the cloud computer
- Reviewable artifacts: what the results delivered by the Bot look like, and how facts and assumptions are distinguished
- Library: where the pages, files, and image/video links produced by Bots are stored
If you want preferences stated in conversations to take lasting effect, continue with 2.4 Memory mechanisms.