Every tool you have stops at the app door. The thing that actually shaped you this week, what the feed chose for you, what you asked the AI assistant and how often, lives on the other side of it, held by companies with no reason to hand it over.
Last month we wrote about switches. Meta agreed to overnight restrictions and a two hour limit for teenagers. Australia moved to make platforms let you turn the algorithm off. Both are controls handed to the person, and neither comes with any way to tell whether using them changed anything. The honest follow-up is harder. Even the measurement you do have stops short of the thing that matters.
Where the counting stops
Every screen time tool on an iPhone, ours included, reads the same thing from Apple: which app, for how long, at what hour, how many pickups. It is more than most people assume. Time of day alone turns out to be worth more than the total, because four hours spread across a day and four hours between ten and two are the same number and very different lives.
But it ends at the app door. Screen Time knows you were in YouTube for forty minutes. It does not know whether that was a lecture or a scroll. It knows a chat app was open. It has never seen a message. Apple keeps it that way deliberately, and we think that is right. The part of any app that reads your Screen Time runs in a sealed process that can draw the numbers but cannot pass them back. We tested it. The wall is real, and it is why a privacy claim on iPhone can be an architectural fact rather than a promise.
So the shape of your day is knowable. What filled it is not. And what filled it is where the real story is.
What is on the other side
Two things, and they are not the same problem.
The feed
An algorithm decided what you saw in roughly the order you saw it, every time you opened the app. It knows exactly what it showed you and why. It knows what you lingered on, what you skipped, what you came back to. Over a month that sequence is a fairly complete picture of what your attention was fed, and what it was taught to want.
You get a button that says “Why am I seeing this?”, one post at a time. And you can request a data export, which arrives days later as a folder of files built for regulatory compliance, not for understanding yourself. The information exists. The interface to see what a month of it did to you does not.
The AI assistant
This one is newer and we think more consequential. Earlier this year OpenAI and MIT published two studies on how people use ChatGPT: an automated analysis of nearly forty million conversations, and a four-week trial of close to a thousand people. The heaviest users were markedly more likely to call it a friend, and heavier use went with more emotional dependence. The researchers were careful: emotional conversations were rare overall, the effects were not shown to be causal, and nobody under eighteen was included.
Separately, an MIT Media Lab team put fifty-four people under EEG while they wrote essays, some with a language model, some with a search engine, some unaided. The model-assisted group showed the weakest brain connectivity, and afterwards had trouble quoting from essays they had written minutes earlier. Small study, not yet peer reviewed, and its authors say so. But it is asking the right question, which is not how long you used the tool. It is what the tool was doing to how you think.
Here is the part worth sitting with. OpenAI could run that study because it holds the conversations. It can see, across millions of people, where use tips from tool into crutch. You cannot see that about yourself, because the record of what you asked, how often, and what you stopped doing for yourself lives on their servers, not in your hands.
Who holds it, and why they will not show you
This is not a conspiracy. It is an incentive. A feed that showed you, plainly, what it had been feeding you would be a feed you used less. An AI assistant that told you when you had stopped thinking for yourself would be one you leaned on less. Neither company is paid for that.
They can do it when it suits them. In July, Meta said it would alert parents using Instagram's supervision tools if their teenager discusses suicide or self-harm with Meta AI. That is content awareness, applied at the exact moment the company chooses, for the purpose the company chooses.
Regulators have noticed, and the box is being prised open. The EU's Digital Services Act now gives vetted researchers access to the internal data of the largest platforms, in force since late last year. Australia's draft Digital Duty of Care, announced last month, is expected to let the eSafety Commissioner and researchers test what platforms actually do. Both are real progress.
Read who they are for, though. Researchers. Regulators. Institutions. Not you. Nothing in either regime hands a person a readable account of what their own feed did to them, or whether their own AI assistant use has crossed a line. The box is being opened for everyone except the person inside it.
What this means for us
ThirdLife exists because counting hours was never the point. The question is what the time did, and today that question can only be answered at the level of shape: when, which app, how it moved with your sleep and your days. That is what we build, and we build it honestly, which means we also say where it stops.
Where it stops is the app door. On the other side is the layer that matters most, and it will not be built by the companies who hold the data, because their incentive points the other way. It will have to be built by someone whose only incentive is the person looking at the screen.
Much of what that would take already exists. The platforms will give you your data if you ask. The models that could read it for you can run on a phone. The hard part is not access. It is building something that reads what your feed fed you and what your AI assistant became to you, and tells you plainly, without a company in the middle and without becoming a new kind of watching.
That is the problem we think is worth solving. We do not have it solved. But we are fairly sure nobody else is trying to solve it for you.
Sources: Your Brain on ChatGPT, MIT Media Lab, arXiv preprint, 2025. Early methods for studying affective use and emotional wellbeing in ChatGPT, OpenAI and MIT Media Lab, March 2025. Delegated act on data access under the Digital Services Act, European Commission, July 2025. Meta, parental supervision updates, July 2026.
Related reading: Everyone is handing out switches · Does any of my data leave my phone? · Screen time and mental health