The pattern is now almost universal: take an existing focus, sleep or wellness app, bolt a generative-AI layer on top, and let it narrate. The promise is interpretation instead of raw data - a friendly voice that explains what your score means and tells you what to do next. The trouble is that fluent narration and accurate insight are not the same thing, and the gap between them is exactly where these features live or die.

What generative AI helps with

There is a real category of work here, and it is mostly linguistic rather than medical. Language models are good at taking structured numbers you already have and turning them into something readable. The real wins are narrow but clear:

  • Summarising your patterns.“Your focus held best on the three mornings you slept past seven hours” is a fair restatement of your own data, and it can surface a link you would have missed in a chart.
  • Drafting a plan. A first-pass schedule or a list of blocks to try is a useful starting point to edit, as long as you treat it as a draft and not a verdict.
  • Reframing.After a rough day, a calmer second reading - “one low day is noise, not a trend” - can lower the self-blame that makes the next day worse.
  • Explaining a score in plain language. Translating a weighting into a sentence is helpful, provided the underlying numbers stay visible.

What is theatre

The failure mode is confident, ungrounded advice. A model can produce a paragraph about your nervous system, your dopamine or your “recovery debt” that reads as authoritative and is, in fact, generated from very little. The calm font and the measured tone do a lot of quiet work here, lending the feel of clinical judgement to what is essentially a guess. Watch for advice that asserts more than the inputs could possibly support, that never shows its working, and that always sounds sure. A score you cannot trace back to its inputs is decoration, however soothing the wrapper.

Real, mixed or theatre

A rough way to sort the features you will meet. The split is not about whether AI is involved, but about whether the output is grounded in something you can check.

AI featureVerdictWhy
Summarising your week in plain languageRealRestating numbers you already have is what language models are good at.
Explaining how a score was builtRealTurning a weighting into a sentence helps, as long as the maths is shown underneath.
Drafting a focus plan or scheduleMixedA useful first draft to edit, not a regime to obey. You still own the calendar.
Reframing a bad day kindlyMixedCan lower the self-criticism, but it is a tone, not a treatment.
Confident health or productivity verdictsTheatreA fluent sentence about your body is not a measurement of it.
A chatbot in a calm fontTheatreSoothing typography is design, not insight.

This split is our editorial judgement, not a measured ranking, and the same feature can sit in different columns depending on how carefully an app builds it.

The privacy cost most apps hide

Here is the part the calm font does not mention. Most AI coaching is not running on your phone. To generate that friendly paragraph, the app ships your data - your focus sessions, your sleep, sometimes your typed notes - to a server, often a third-party model provider, where it is processed and may be retained. The interpretation is real, but so is the journey your most personal signals took to produce it. An on-device score that never leaves the phone cannot make that trade, which is both its limit and its point.

Illustrative share of your personal data that typically leaves the device for processing. Cloud coaching sends it out to interpret it; an on-device score keeps it local by design.

How to judge one

Before you trust an AI focus feature, ask four plain questions:

  • Where does it run? On your device, or on a server? If your data leaves the phone, the privacy of every other claim is conditional on that server.
  • Can you see the maths? A good feature shows the inputs behind the sentence. If you cannot trace the advice to numbers, treat it as opinion.
  • Does it admit uncertainty?Honest tools hedge - estimates, ranges, “this looks like”. Theatre is always sure.
  • Does it stay in its lane? Reflection and reframing are fair. Diagnosis is not, and any app that drifts toward it has overreached.

Where Soren stands

Soren is deliberately on the quiet side of this. It computes one daily 0-100 Soren Score on your iPhone from three pillars - Focus, auto-detected Sleep written to Apple Health, and Physical movement from HealthKit - with no account, no servers and no analytics. Nothing leaves the device. That means we forgo the cloud-coach paragraph, and we think that is the right trade: the score is a private estimate and a reflection, never a diagnosis, and it stays yours. While the category rushes to the cloud to sound clever, we would rather stay local and let you read your own numbers in peace.

Get Soren

Soren is iOS 17+, on the App Store. The Soren Score is an honest estimate and a reflection, not a medical or diagnostic claim. Descriptions of other apps reflect common patterns as of June 2026; if we have something wrong, tell us at hello@sorenlabs.xyz and we will fix it.

More from the Journal