The phone is for the room you are standing in. Voice carries the brief. The camera carries the object. Typing is a fallback, not the point.
Why this matters
A desk thread cannot see the breaker panel, the whiteboard, the pallet label, or the way a room is actually laid out. People then describe the thing badly and argue with a model that never saw it.
The Grok apps on iOS and Android can talk, and they can look. Voice mode answers out loud. Live camera shares the view while you talk. If live camera is gated on your plan, a photo you take in this session is the honest substitute. A stock image from last year is not.
This is not a demo of a talking mascot. You will leave with a field note you could hand to someone who was not there.
Teach
Install the official Grok app (xAI) on iPhone or Android. Sign in with the same account you use on grok.com. If you skip that, A4’s next lesson will fall apart.
Voice is a briefing, not a podcast. Start with the job, then the constraint, then talk. Dead air is fine. So is “wait.” So is “that label, not the other one.”
A spoken shape that works:
Job: I need a six-line field note I can text the shop.
You are looking at [the thing].
Tell me what you can actually see. Then tell me what you cannot see.
Do not invent a brand, a rating, or a part number you cannot read.
When I say “write it up,” give me six lines I can copy.
Speak the way you would speak to a new hire on the first day. Short sentences. Point at the thing. Correct it out loud: “No, the left tag. The faded one.”
Live camera is a narrow question. Point, then ask one thing. “What does this warning label require before I energize.” “Read the four lines on this packing slip.” “Is this a top view I could sketch.” Do not sweep the room like a tourist and ask “what should I do with my life.”
If live camera is not available, take a photo in the app and ask the same narrow question. The photo must be from this session, of this object. The test is: would the answer change if you pointed at a different object? If not, you are not using the camera.
Privacy is part of the skill. Do not put these in frame or on the mic:
- Badges, faces you do not have permission to capture, kids
- Other people’s screens, lock boxes, passwords, full account numbers
- Plates, interiors of homes, or a client site if the contract would frown
- Medical marks, school records, anything you would not leave in a shared album
If you catch something in frame, stop, delete the photo if you can, and start again on a safer object. A coffee mug is legal and still a bad exercise. Pick a labeled object, a whiteboard you own, a tool wall, a breaker legend, a bookshelf you arranged, a handwritten list, a piece of equipment in your shop.
Get the words back. Voice disappears unless you capture it. After the look, ask Grok to write the field note in the thread (or dictate: “write that in the chat”). You need text you can paste into this course and into a message. If the app offers a transcript or a recap, use it. If it does not, the “write it up” ask is the recap.
Voice on a noisy lot: get closer, face away from the wind, and repeat numbers. If Grok hears “fourteen” and you said “forty,” that is on you to catch. Read numbers back.
Desktop voice exists on some grok.com sessions. It is not this lesson. This lesson is the phone, in a real room.
Worked example
A tech stands in front of a shop breaker panel she is allowed to photograph. She opens Grok on the phone, same account as the desk, starts voice, raises the camera.
She says: “Job: six-line note for the apprentice who will reset this after lunch. You can see the panel. Read the labels you can actually read. Do not guess amperage you cannot see. Tell me if any breaker is in a position you can see. When I say write it up, put it in the chat.”
Grok reads three labels, misses a fourth in glare. She tilts the phone: “The bottom left. Read it again.” It reads it. She says “write it up.” She gets six lines, including “bottom-left label was glare; confirmed on second look” and “amperage on the main not readable from this angle.”
That last line is the win. The model admitted what it could not see. A typed chat would have invented 200 A because that is a common main.
She does not film the office manager’s monitor on the way out.
Your turn
This exercise is on the phone. Laptop does not count, even if it has a webcam. Find an object you are allowed to show: a labeled panel, a whiteboard, a tool, a control screen you own, a packed shelf, a handwritten list, a machine placard. Not your face. Not someone else’s ID.
If Voice or Live Camera is locked on your plan, still open the Grok mobile app. Take a new photo in the thread and speak or type the same brief. You must include something only the lens could have provided (a word on a label, a count of objects, a color of a tag).
Exercise a4-l01-e1
| Field | Value |
|---|
| whereToDo | mobile |
| time | 18 min |
Copy-ready prompt (say this; then point the camera or attach the new photo)
I am doing Grok Mastery A4, voice and camera.
Job: a 6-line field note I can send to someone who is not in this room.
Look at the object I am showing you.
Rules:
- Describe only what you can see.
- Read any text you can actually read. Quote it.
- If you cannot read a number or a word, say "not readable."
- Do not invent a brand, rating, or part number.
- Ask me one clarifying question if the job is unclear.
When I say "write it up," put in the chat:
1) the 6-line field note
2) a See/Can't-see list: 3 things you saw, 2 things you could not see
3) How I showed you: live camera, still photo, or voice only
4) Session stamp: today's date and "Field note written by Grok from this phone session."
After Grok looks, answer its question if it asked one. Then say write it up. Confirm the field note landed as text in the thread. If voice did not drop text into the chat, type: Write it up in this thread now and send it.
Artifact to paste
- The 6-line field note.
- The See/Can't-see list.
- The How I showed you line and the session stamp.
- One sentence from you: the object (generic is fine: “shop panel,” “kitchen breaker legend,” “whiteboard”) and whether you used voice, camera, photo, or a mix.
Rubric
- Pass: Stamp is present. See/Can't-see includes at least one quoted word or a count that had to come from the lens. Grok marked something not readable or not seen (honesty about the frame). You used the mobile app. The object is not a stock download and not a course screenshot.
- Fail: A note you could have written from this lesson alone. “Voice only” with no object and no quoted visible text. A celebrity photo or a random image search. No stamp. You did this on grok.com with no phone. Faces, badges, or secrets in the description.
Common failure
The common failure is the demo sweep. You walk the house, narrate your feelings, and never ask for a field note. You will remember it as “I tried voice.” You will have nothing.
The other failure is letting Grok fill glare with confidence. If the label is unreadable, the right output is “not readable,” not a typical value for that class of device. If the note contains a number you cannot find on the object, strike it before you send the note to anyone.
What’s next
You can work in the room. Next: get that thread onto the desk, keep one account in sync, and hand a clean note to a human who is not inside Grok.
