How to use AI on your phone properly
The phone wins at the camera, at voice and at short gaps. How to use each, and the button that means the opposite of what it does in a browser.
At five past eight on a Wednesday, Marta photographs a parking sign in a town she does not know and asks whether she can leave the car there until six. The answer arrives before her bus does.
A phone is the worse device for most of what people do with AI. Typing is slower, the screen holds about a paragraph at a time, and you are usually halfway through doing something else. There are three things it does better than any laptop, and the rest of this page is about those, and about the places where the app behaves differently from the website. We will follow Marta through the rest of her day: laptop at work, phone everywhere else.
Three things the phone is better at
The camera is the largest of them. Photographing something confusing and asking what it means is a question you can only really ask with a phone in your hand, and 46 of our 59 models can read an image, including 14 of the 22 available on the free plan. On a laptop the same question begins with finding a file.
Voice is the second. Speaking is much faster than thumbs, and it is the only input that keeps working when your hands are carrying shopping.
The third has no common name, so we will call it the gap: four minutes at a bus stop, the wait for a kettle. Answers are generated on our servers rather than inside the app, so an answer still being written survives you locking the screen and opening something else. Come back and the app reconnects to the same stream and carries on.
Point the camera first
Go back to that parking sign, four lines of restrictions on it and a bus due. The photograph was half of what Marta sent. The other half was the decision she actually needed out of it.
That phrasing is the whole technique. A photograph with no question attached gets you a description of the photograph. A photograph with the actual decision attached gets you the decision, and usually the line of small print it turned on.
The attach button opens a sheet with Camera, Photos and Files side by side, so a picture taken for one question never has to land in your camera roll. If your iPhone is set to High Efficiency, which is the default, its photos are HEIC files, and the app now re-encodes those to JPEG as you pick them, before anything uploads. That last one ships inside the app rather than over the air, so a photo that still lands as HEIC means the store update has not reached you. A free account sends one image per message, up to 10MB, three a day, and only on a model that reads images.
One thing worth knowing before you photograph something personal. We do not train on your conversations. What the provider behind a given model does with them varies by model, and the models where we know of a difference carry a notice on their info card in the picker on the website. Most models carry no notice at all, and the phone app does not show them, so nothing you can read on the phone will settle this for you.
The gap, and why it is longer than it looks
Four minutes is enough for one real question, which is a smaller unit than it sounds. The ordering matters more than the length: send the slow question first, put the phone back in your pocket, and spend the rest of the wait on quick ones while the long answer builds behind you.
Nothing will tell you when it lands, because the app sends no notifications at all, so a finished answer sits silently in the thread until you next open it. On the bus Marta asks for a summary of a twelve-page brief she attached at her desk, then reads it two lines at a time between stops. She arrives with the shape of it and a question for the client.
What is locked, and what is only hidden
This is the part most guides leave out, so here it is plainly. In the phone app all three tool controls are always available, and two of them are locked: tap web search or image generation on a free account and you get an upgrade prompt rather than a quiet failure. On the website those same two, and the reasoning one, are usually not there at all. Intent Detection is on by default, and while it is on the app reads each message and decides for itself about searching, images and reasoning, so the buttons are not drawn. Switch it off in Settings, then Tool Preferences, and all three come back. A free account gets two searches in any 30 days either way, and a search the app chose to run on its own still spends one of them.
The file picker offers everything on your phone, including Word, Excel and PowerPoint files, their OpenDocument equivalents, and text or code files. On a free account, choosing one of those produces a message telling you document or code uploads are a Pro feature, and the file is dropped after you have picked it. Images and PDFs are the two kinds a free account can send, and PDFs only on a model that accepts them.
Memories are paid on every platform, so that one is not a phone restriction. If you are arriving from another assistant, the import flow has a path written for the app rather than the browser. Plans and what each one includes are on the pricing section of our homepage.
Voice sits behind a gate we are in the middle of moving, so we are not going to tell you here which plan reaches it. Open the composer and look at the microphone: if it is there and it opens, you have it.
About that chat you started at work
Nothing had to be set up for this. Both apps read the same account, so the thread Marta left open on her work laptop is in the phone's list at lunchtime and she carries on typing at the bottom of it. What does not make the trip is anything she typed and never sent, because a half-written message belongs to the screen it was drawn on.
Talking to it while your hands are busy
Voice calls exist only in the phone app. The browser has a microphone that dictates into the message box and no call mode at all, which makes this the one thing here that is mobile-first rather than mobile-as-well.
At twenty to seven Marta walks home and talks through a difficult email she has to send in the morning. A call behaves like a conversation: you can interrupt, change your mind mid-sentence, and say that the last suggestion was too formal without typing any of it.
Each turn is written into the conversation as an ordinary message as it finishes, with a small marker showing it came from a call. So the walk leaves a readable record, sitting in the thread at a keyboard the next morning, as text. No audio is stored there, so there is nothing to play back.
Try again does not mean what it means in a browser
In the browser, the control labelled Try Again branches. It copies the conversation up to that point into a fresh thread, answers again there, and leaves the original exactly as it was.
In the phone app, the item labelled Try again regenerates in place. The answer you were unhappy with is deleted from the conversation on our servers, along with anything that came after it, and the replacement streams into the same thread. There is no undo, and because the thread is shared, the deletion reaches your laptop too.
If there is any chance you will want the first answer back, use Branch here rather than Try again. It is the branch button sitting beside the one you were about to press, it makes a second conversation, and it leaves the first alone.
Marta finds this out at nine in the evening when a good paragraph about the difficult email is replaced by a worse one. There is no side-by-side view on either platform: a branch is a separate conversation you move between, so comparing two answers means switching threads.
So is the phone the better device?
For capture, yes, and by a wide margin. Most of what is worth asking about arrives in front of you while you are standing up: a sign, a form, a label, a sentence in a contract someone has just handed you. The device already in your hand, already holding a camera, wins those outright, and a laptop cannot compete for them at any price.
For the second half of the work, the reading and the rewriting, the laptop still has it, and the phone app does not pretend otherwise. The practical shape of a day is therefore capture on the phone and finish at a desk.
Marta's sign, her bus-stop summary and her walk home all happened on the right device. The one thing she lost that day, she lost by tapping two familiar words in an app where they mean the opposite of what they mean in her browser. Learn that difference before you need to retry an answer on your phone.