What we actually know — and how sure we are
Five load-bearing claims, each with its source and a confidence rating. Select a row for the nuance behind the number.
What the data is telling leadership to do
Each observed behavior carries a strategic consequence. The bridge from “interesting” to “act on this.”
AI is the engine. Each patient’s own data is the fuel.
Patients now arrive carrying their own longitudinal data — wearable streams, genetic risk, lab panels, a portable record — and a new expectation: that someone will actually read it. The patient who has tracked their resting heart rate for two years no longer accepts “your labs look normal.” They expect their data referenced, decoded, and acted on. AI is what makes that expectation answerable at scale; the system that can’t meet it looks blind.
A genome, a CGM trace, or a 100-biomarker panel is meaningless without translation. The patient-facing job is comprehension: turning raw streams into something a person can act on.
The data already flows toward the clinic — 59% of wearable owners have discussed their data with a provider — but most systems can’t ingest it. The provider-facing job is leverage: making individual data usable inside the encounter.
The same survey base (Rock Health, Dec 2025) anchors the wearable figures; records, genetic, and lab figures are separately sourced and confidence-rated below. Market-size dollar forecasts vary widely across vendors and are deliberately omitted — the behavior and direction are the evidence, not anyone’s revenue projection.
Six segments, six different risks
Adoption isn’t uniform. The same technology produces a prepared patient and a care-avoiding one. Select a segment.
How AI reshapes each step of the patient journey
Seven stages, the traditional path beside the AI-shaped one — and the concrete thing a health system could build at each step. Select a stage.
The traditional and AI-shaped columns are structured inference built on the survey evidence above, not measured findings. Figures shown are sourced; the stage framing is judgment.
Jobs migrate to AI in inverse proportion to stakes
Each patient job moves at a different speed, scored across clinical risk, trust required, data needed, regulation, and integration. Select a job.
Adoption compounds on a flywheel made of trust
Change compounds. Each low-stakes success earns the trust that licenses the next job up. Tap a stage.
The wheel turns for whoever delivers the wins. Every successful interaction builds trust in some AI — and with 74% of patients on general-purpose tools and just 5% on provider bots, it’s the platforms’ wheels spinning today, not yours. The asymmetry is the danger: trust compounds slowly but collapses fast, so one visible high-stakes failure can spin the whole wheel backward across every job at once.
Four worlds, two axes — and where we are now
Cross two uncertainties — who owns the trusted agent, and whether trust converges with usage. Today’s baseline is marked; the labels are a rough read on each world’s near-term (18-month) likelihood, not a forecast. Select a quadrant to inhabit it.
Trust convergence accelerates whichever way the wheel is already turning
The same force is the incumbent’s best case or its worst — depending only on whose wheel was turning when it hit.
Best case.
Convergence resolves the usage–trust contradiction. Deeper jobs — triage, provider matching — migrate sooner, and onto your surface. The patient routes through the system’s AI because they’ve come to trust it.
Worst case.
The same convergence accelerates the delegated world. The patient’s own agent now trusts itself to run the journey end to end — and treats your system as an interchangeable supplier. The trust got earned; it just didn’t get earned by you.
Which moves survive which worlds — and where to start
Each move scored against all four futures, then classified and sequenced. No-regret moves pay off almost everywhere; hedges protect the downside; options wait on a signal. Priority, owner, and effort turn the analysis into a fundable plan.
Machine-legible content & data. It is the cheapest of the eight, it is no-regret in every world, and it is the precondition that lets every external AI — and every later move — find and trust you. Run clinician AI-readiness as the parallel process track.
Owner and effort are rough order-of-magnitude estimates for a mid-size system; actual scope depends on existing infrastructure, vendor choices, and integration debt. Treat them as a starting point for planning, not a quote.
What to do across the 1–3 year horizon
Sequenced to the trust flywheel: become visible to AI, earn low-stakes trust, then prepare for agentic delegation.
Become visible to AI
- Structured content for conditions, providers, services, locations
- Clear authorship and medical-review dates
- Schema and entity optimization so AIs cite you correctly
- Real-time availability where possible
- Measure AI referral traffic and answer visibility
Earn low-stakes trust
- AI-supported post-visit comprehension
- AI-assisted care navigation and appointment prep
- Honest triage — willing to route patients away
- Safe, fast escalation to a human
- Clinician workflows for AI-informed patients
Prepare for delegation
- APIs for availability, services, insurance, access points
- Patient-permissioned data exchange
- Agent-readable quality, cost, and access signals
- Governance for external-agent interactions
- A plan for when the patient’s AI is the interface
Four indicators that would change the picture
The scariest uncertainty — when the delegated world arrives — stays monitorable through a handful of leading signals. These are its preconditions.
Data portability
Already accelerating: TEFCA went from ~10M records exchanged to ~500M in a year. Watch for it reaching the patient’s own agent.
Trust convergence
The accuracy and privacy gaps narrowing — usage and trust closing in survey data.
Agents that act
Personal AI agents gaining persistent health memory and permission to act, not just advise.
Device-data integration
Health systems turning on wearable feeds (HealthKit, Health Connect) — closing the gap between data patients bring and data providers can use.
Whose AI earns the trust that compounds will decide who owns the patient journey — and that goes to whoever becomes both trusted by humans and legible to machines.
A note on method. Adoption, trust, privacy, and access figures are drawn from named 2025–26 surveys (Rock Health, KFF, West Health–Gallup) and rated for confidence in the evidence section. The journey-migration structure, segments, scenarios, flywheel, and strategy grid are structured reasoning built on those facts. The two most disputable claims — that trust stays split in the near term, and that the flywheel is asymmetric — are flagged as assumptions, not findings. Scores in the migration ladder and strategy grid are calibrated judgment, not measured values.