Three technologies — built for completely different purposes — are quietly assembling the infrastructure that could save rural healthcare. Here's what's working, what's possible, and what's still standing in the way.
Start with what's true: one-third of all rural hospitals in America — 734 facilities — are at risk of closure. This week, emergency departments in Arkansas, Pennsylvania, Maryland, and New Jersey either closed or announced they will. More than 300 hospitals have already eliminated OB services. More than 450 have eliminated chemotherapy. The "One Big Beautiful Bill" is projected to remove between 10.5 and 11.8 million people from Medicaid, and states are now required to run eligibility redeterminations every six months — an administrative burden widely understood to push eligible people off the rolls.
The conventional response to all of this is a supply argument: we need more hospitals, more physicians, more federal money. That argument is correct and useless at the same time. Physician pipelines take a decade. Hospital construction takes longer. Federal healthcare politics are generational.
So let's ask a different question.
Not "how do we build new capacity?" but "what is strangling the capacity we already have?"
The answer, when you look at the data, is not what most people expect. And neither is the solution.
The signals that tell you what's actually happening on the ground.
| Rural hospitals at risk of closure | 734 (⅓ of all rural) |
| Rural census tracts with care shortage | 89% |
| Medicaid enrollees at risk of loss | 10.5 – 11.8 million |
| Adults who miss care due to transport | 21% annually |
| Cost of missed appointments to system | $150 billion/year |
| Physician time lost to admin/day | 35 – 40% of total hours |
| Hours ambient AI reclaims daily | 2 – 3 per physician |
| NEMT reduction in hospitalizations | 15% |
Look at the bottom three rows alongside the top five. The access crisis is visible in the top half of that table. The solution is hiding in the bottom half. The same week we're watching hospitals close, the data on what could stabilize them is sitting in plain sight — deployed, proven, and massively underconnected.
Here is the part of the access conversation that almost never gets said plainly.
A rural physician serving a county with one doctor per 3,000 residents is not spending 100% of their time on patients. They're spending 35 to 40% of their day on documentation, prior authorization, and administrative tasks that have nothing to do with clinical care. That's not a complaint — it's a structural measurement. The average primary care physician loses nearly an hour each day just managing AI alerts from tools meant to help them. Add documentation burden and prior auth, and you have a physician operating at roughly 60% of their potential patient-facing capacity.
Now apply ambient AI. Physicians using ambient documentation tools are reclaiming 2 to 3 hours daily and seeing 15% more patients per hour. In a solo rural practice, that's not an efficiency metric. That's the difference between a physician who can serve their community and one who burns out and leaves.
The access crisis is, in significant part, an efficiency crisis wearing a supply shortage costume. We keep trying to solve it by adding supply. The faster path is unlocking what's already there.
The efficiency unlock is one layer. But reclaiming physician hours only matters if patients can reach the care those hours enable. That's where two other technologies enter the picture — neither of which was designed for healthcare, and both of which are already deployed.
Physicians using ambient documentation tools are reclaiming 2 to 3 hours daily and seeing 15% more patients per hour. In a solo rural practice, that's the difference between a physician who can serve their community and one who burns out and leaves. The access crisis is, in significant part, an efficiency crisis wearing a supply shortage costume.
The historical failure of rural telehealth wasn't the clinical protocol or the video platform. It was bandwidth. Frontier counties, tribal lands, the Mississippi Delta — the places where access collapse is most acute are the same places where terrestrial broadband has always underperformed or simply not existed. Starlink's low-earth-orbit coverage reaches those geographies. The Health Wagon in rural Virginia deployed Starlink during Hurricane Helene to run mobile units and transmit patient data to specialists in real time. Holland Hospital is running it as failover for mission-critical systems including EHR and telehealth. The NHS is using it to connect GP practices where traditional broadband has failed for years. This is not a proof of concept. It is operational infrastructure being used right now.
21% of American adults miss medical appointments annually because they have no reliable way to get there. That costs the healthcare system an estimated $150 billion per year and generates a significant share of the preventable hospitalizations that are driving rural hospitals toward insolvency. Non-emergency medical transportation benefits in Medicare Advantage plans — funded through plan rebates averaging $2,250 per enrollee — have documented outcomes: 15% reduction in hospital admissions, 40% improvement in treatment adherence, $480 million saved per 30,000 Medicaid users. A Lyft-partnered transportation program saw physician visit utilization increase 73% among beneficiaries. AI is now being applied to routing, scheduling, and no-show prediction in NEMT — making those benefits usable rather than theoretical.
1. The physician doesn't have enough time → ambient AI reclaims it.
2. The patient can't get a telehealth signal → Starlink delivers it.
3. The patient can't physically get there → AI-optimized NEMT routes them.
This is not a moonshot. It is an assembly problem.
Understanding what's already incentivized matters before arguing for what's missing. The picture is more encouraging than the access headlines suggest.
The most significant new incentive is one almost nobody is connecting to this stack. The CMS ACCESS Model — Advancing Chronic Care with Effective, Scalable Solutions — launches July 2026 with more than 150 participating organizations already approved. It runs for 10 years. It pays on outcomes, not activities, for conditions covering two-thirds of Medicare beneficiaries: hypertension, diabetes, CKD, cardiovascular disease, depression, musculoskeletal pain. Critically, it includes an explicit payment adjustment for rural patients. These are precisely the chronic conditions where transportation barriers cause the most missed care and where the AI + Starlink + NEMT stack would have the highest measurable impact. The incentive structure exists. The delivery stack to fulfill it is not yet assembled.
The MA supplemental benefits framework gives plans the legal and financial runway to fund transportation. The evidence for the ROI is documented at scale. The FCC's Healthcare Connect Fund subsidizes provider-side connectivity for telehealth delivery. Ambient AI's efficiency gains translate into direct revenue in fee-for-service settings, creating organic adoption incentives without requiring a specific reimbursement code. The building blocks of a policy environment that rewards the stack are real.
Three things are working against the stack despite those foundations.
Thirty percent of individual Medicare Advantage plans offered non-emergency medical transportation in 2025. That figure dropped to 24% in 2026. The mechanism is the sunset of the Value-Based Insurance Design model at the end of 2025 — a policy that had encouraged supplemental benefit generosity. As VBID expired, plans facing margin pressure cut the benefits that were discretionary. Transportation was one of the first to go. The access cliff is steepening. The benefit designed to address it is shrinking. These two things are happening simultaneously and the policy conversation has barely noticed.
The FCC declined Starlink's application for nearly $900 million in Rural Digital Opportunity Funds. The Healthcare Connect Fund subsidizes provider-side connectivity — hospitals and clinics — but not patient-side. A rural patient in frontier Montana who needs to participate in a telehealth visit with their ambient-AI-assisted physician has no federal program funding their home connectivity specifically for healthcare. The stack works at the provider end. It breaks at the patient's kitchen table, which is where it matters most.
If a rural MA plan deploys this stack and reduces hospitalizations by 15% — which the NEMT evidence suggests is achievable — most of the financial benefit accrues to CMS as reduced spending, not to the plan that made the investment. STARS quality bonuses are the nearest mechanism, but they're lagged two years and indirect. A plan CFO looking at the cost of deploying ambient AI tools, NEMT benefits, and connectivity infrastructure against the return timeline has a legitimate problem. The incentive to assemble the stack exists in theory. The math doesn't close in practice.
This is where the piece earns its keep. Not a wish list — a targeted set of structural changes that would connect what already exists.
The ROI evidence is strong enough, the utilization gap is documented, and the access correlation is clear. Tying a NEMT mandate to HRSA's Health Professional Shortage Area designations creates a targeted requirement without broad cost impact. Making it discretionary means plans cut it when margins tighten — which is precisely when access barriers worsen.
The E-Rate program subsidizes broadband for schools. An equivalent program for rural patient homes in designated shortage areas — with satellite explicitly included as an eligible technology — would close the last-mile gap that Starlink's RDOF rejection left open. This doesn't require a new agency or a new bureaucracy. It requires extending an existing framework to the demand side.
The ACCESS model's outcome-aligned payment is the right blueprint and it's in Original Medicare. Extending a version of that logic to Medicare Advantage — specifically rewarding rural access improvement that can be attributed to technology-enabled care delivery — would align plan incentives with investment. Without it, the financial case for deploying the stack in MA lives entirely in STARS bonuses and enrollment growth. That's too indirect and too lagged to move capital.
Right now, ambient documentation tools, NEMT scheduling platforms, and telehealth systems don't share data. The AI doesn't know when a patient needs a ride. The NEMT system doesn't know the care plan. The physician doesn't know who couldn't get there. A data exchange requirement under TEFCA or as a condition of ACCESS model participation could mandate the connections that would make the stack function as a system rather than three adjacent point solutions.
Issues 001 and 002 argued that healthcare AI is being sabotaged by success — deployed into dysfunction, resisted by the humans whose identities it threatens, scaled before the organizational conditions for scale exist.
This week's signals suggest something different is also true, running quietly alongside the dysfunction: there are places where AI isn't threatening anyone's identity, because it's filling a void, not displacing an incumbent. Rural communities where the physician already left. Patients who already can't get there. Hospitals already operating at a loss.
In those contexts, the resistance that sabotages AI in well-resourced urban systems is largely absent. The ambient documentation tool that a burned-out ER physician in Boston views with suspicion is a lifeline to the solo practitioner in rural North Carolina who is otherwise drowning in paperwork. The telehealth platform that a specialist in a major academic center views as a downgrade is the only specialist access available to a patient in a frontier county.
The access crisis is a terrible problem. It is also, in a narrow but important sense, an opportunity: the human barriers to AI adoption that dominate the conversation in well-resourced settings are weakest precisely where the access crisis is worst.
The stack doesn't need to convince anyone it belongs. In these communities, it just needs to show up.
AI can't build a hospital. But it can give a physician back three hours a day, extend their reach two hundred miles, and make sure the patient who needs them can actually get there. That's not everything. It's enough to start.
Quick note before we start: Several of you wrote in after Week 1 — thank you. The most common question: "I set up my sheet with placeholder prices. How do I get real numbers in there?" That's exactly what we're doing today. Week 2 is where the tracker stops being a template and starts being a tool.
If you're joining us for the first time, the 13-week map is below. Week 1 built your foundation — Claude account, Google Sheet, Bubble.io app skeleton. Today we connect it to the real world.
Go to goodrx.com. No account needed. Search for each of the 3 drugs you entered in Week 1. For each one, note the lowest price in your zip code and which pharmacy offers it. You're looking for the cash price — not insurance — because that's what GoodRx optimizes. What you'll notice immediately: the price variation is shocking. The same 30-day supply of Metformin 500mg can range from $4 at Walmart to $18 at a chain pharmacy two blocks away. That gap is what this tool is built to close. Enter the real GoodRx prices into your sheet — update the Price and Pharmacy columns with what you just found. Add "GoodRx" to the Notes column so you remember the source.
Go to needymeds.org → click "Drug Pricing Tool." Look up the same drugs. NeedyMeds pulls from a different set of discount programs — manufacturer assistance, state programs, nonprofit discounts — and sometimes beats GoodRx, especially for brand-name drugs with no generic equivalent. If NeedyMeds shows a lower price for any of your drugs, add a second row for that drug with the NeedyMeds source. Your sheet is now doing something real: holding two competing prices for the same drug so you can choose.
In your Google Sheet, add an 8th column: Price Source (GoodRx, NeedyMeds, Manual, etc.). Go back and fill this in for all your existing rows. This matters more than it seems — in a few weeks when you're comparing 10 drugs across 4 sources, knowing where each price came from is what makes the data trustworthy. While you're there: delete any placeholder prices you entered in Week 1 that weren't real. Your sheet should now contain only verified, sourced prices.
Right now your "Date Checked" column is static — you typed a date. Let's make it smarter. Click on any cell in the Date Checked column. Delete what's there and type this formula exactly:
This makes the cell always show today's date. The limitation: it updates every day whether or not you've checked the price. So use it as a "last opened" signal rather than a verified check date — and note in the Source column when you actually verified a price. In Claude, ask: "How do I lock a date in Google Sheets so it only records when I actually enter data, not every time I open the file?" The answer introduces you to a slightly more advanced formula. Don't worry if it takes a minute — that conversation is your first real prompt engineering exercise with practical stakes outcome.
In Google Sheets: select your Price column → click Format → Conditional Formatting → set it to highlight the lowest value in green. Now your sheet does something visually useful: the cheapest price for each drug is immediately obvious. This is the foundation of the comparison view we'll build in Week 3. Take a screenshot. Your sheet now has real prices, real sources, auto-dating, and visual hierarchy. That is a functional tool.
If your sheet shows real prices, real sources, and the cheapest option is highlighted — you finished Week 2. Your tracker is no longer a template. It's a tool.
One question worth sitting with before next week: How different were the prices across pharmacies for the same drug? Drop your most surprising finding in the comments. The data usually shocks people — and that reaction is exactly why this app is worth building.
Next week: We build the comparison view. Instead of looking drug by drug, you'll see all your medications side by side across pharmacies — so the full picture of where to fill your prescriptions becomes obvious at a glance. Week 3 is where the tracker starts to feel like the app it's becoming.
Data sources: Commonwealth Fund Rural Hospital Funding Crisis 2026 · HRSA Health Professional Shortage Area Data April 2026 · Urban Institute Medicaid Cuts Analysis · Milliman NEMT Analysis · AHA Rural Hospitals and the AI Advantage January 2026 · Becker's ambient AI productivity data · Health Wagon Starlink deployment · CMS ACCESS Model documentation April 2026 · FCC Rural Health Care Program records · NeedyMeds.org · GoodRx.com.