Becker's Hospital Review surfaced a fascinating operating question this morning: patients at Advocate Health are linking their MyChart records to ChatGPT at a rapidly accelerating rate — and the health system is absorbing part of the API cost even though it did nothing to turn the connection on. For RCM leaders, the bigger question is not the API bill. It is what happens when the patient brings an AI agent into the financial side of healthcare.
Increase in Advocate Health patients linking MyChart to ChatGPT from July to August, according to Advocate EVP and CIO Bobbie Byrne, MD, as reported by Becker's Hospital Review.
Advocate Health did not enable the connection.
Patients did.
According to Becker's, every linked account and every request a patient sends through ChatGPT creates API calls against Advocate's Epic environment. Because Advocate is Epic-hosted, the system can see the resulting API costs.
For now, Advocate says the cost is manageable.
But the growth rate is what makes the story important.
The health system is no longer the only party deciding which digital tools sit between the patient and the medical record.
The patient can now bring their own intelligence layer.
This is the beginning of a patient-selected technology layer sitting on top of provider infrastructure. In RCM, that could become a new front door for billing questions, benefit interpretation, estimates, denials, payment options, and financial assistance.
Becker's framed the immediate issue well: who pays when the patient chooses a third-party AI experience but the provider's infrastructure has to service the requests?
That is already a meaningful economic question.
But for revenue cycle leaders, the second-order effects may be larger.
Patients already struggle to interpret:
Historically, those questions landed in patient-access teams, call centers, billing offices, payer service lines, or Google searches.
Increasingly, the first stop may be an AI assistant that already has context from the patient's medical record.
That changes the patient financial journey.
ChatGPT Health today is designed to help users understand connected medical records, lab results, medications, and other health information. It does not mean that every billing record, payer contract, or claim detail is automatically available to the model.
But the direction is clear.
As more financial and administrative data becomes accessible through patient portals, payer apps, price-transparency files, and other connected sources, the patient-side AI can become much more capable.
A patient could ask:
Why is this bill higher than the estimate?
Does this denial make sense based on what my physician documented?
Was this service supposed to be covered under my plan?
Do I qualify for the hospital's financial-assistance policy?
Is this amount consistent with the hospital's posted price and my insurer's explanation of benefits?
What should I ask the billing office before I pay?
Those are not clinical questions.
They are revenue-cycle questions.
Healthcare organizations have spent years building internal systems to manage the financial journey.
Eligibility engines.
Patient estimates.
Contact centers.
Digital billing.
Payment plans.
Denial workflows.
Financial-assistance screening.
Patient-side AI creates a parallel intelligence layer that the provider does not necessarily own.
That does not mean the provider loses the relationship.
It means the patient may arrive at the conversation better prepared, with different questions, and potentially with a machine-generated interpretation of the bill, policy, or denial already in hand.
The traditional model asks: how do we make our portal easier to understand? The emerging model asks: how do our financial workflows behave when the patient has their own AI interpreting everything we expose?
Most patient-facing financial communications were written for humans.
And many were not written especially well.
Statements can be difficult to reconcile with EOBs.
Denial language can be vague.
Estimates often contain caveats that are technically correct but practically confusing.
Financial-assistance policies may require patients to find the right page, download the right form, understand eligibility thresholds, and submit documentation correctly.
AI changes the importance of structure.
If a patient's assistant is going to interpret the provider's financial information, then clear data fields, consistent reason codes, accessible policy language, and transparent calculations become more valuable.
The audience is no longer only the patient.
It may also be the patient's software.
Providers and payers are rapidly adopting AI for denial prevention, coding, utilization management, payment integrity, and appeals.
Patient-side AI introduces another participant.
A patient receiving a denial or unexpected bill can already upload or connect information and ask an AI system to explain what happened.
As those systems gain more structured access to records and coverage information, they may become increasingly capable of helping patients identify inconsistencies or formulate better questions.
That could increase the volume and sophistication of:
The patient financial contact center may therefore see fewer basic questions over time — but harder ones.
Hospital price-transparency files have historically been difficult for ordinary consumers to use.
Advocate's own pricing page acknowledges that machine-readable files are lengthy and complex and encourages patients to seek estimates and assistance.
That is exactly the kind of complexity AI can collapse.
A patient does not need to understand a machine-readable file if an AI can interpret it, compare it with an estimate, and explain the difference.
The same is true for payer transparency data, EOBs, benefits documents, and financial-assistance rules.
Transparency mandates become more powerful when consumers have software capable of actually reading the transparency.
There is also a less glamorous but important infrastructure question.
Advocate's CIO told Becker's that the system supports patients using ChatGPT to better understand their health, but noted the disconnect between who benefits and who pays.
That is likely not the last time healthcare encounters this issue.
If patients increasingly use third-party AI agents to query provider infrastructure, somebody has to pay for:
At small scale, those costs may be immaterial.
At mass adoption, they become an operating-model question.
Should the provider absorb the cost because interoperability improves patient engagement?
Should the AI platform pay?
Should the EHR absorb it?
Should there be commercial API tiers?
Or does the cost simply become part of running a modern digital health system?
The answer is not obvious.
This may be the most strategically important implication.
Health systems have invested heavily in portals because the portal is where they want the patient relationship to live.
Appointments.
Messages.
Results.
Bills.
Payments.
Care navigation.
But if patients increasingly connect that data to a general AI assistant they already use every day, the portal can become infrastructure rather than the primary interface.
The patient may still use MyChart.
But the place where they understand MyChart could be ChatGPT.
That is a major distribution shift.
Today, a patient calls a billing office.
Tomorrow, the patient's AI may be able to ask a provider system a structured question.
The provider's AI may answer.
The patient's AI may compare that response with the payer's data.
The payer's AI may explain coverage.
The provider's system may offer a payment plan or financial-assistance workflow.
That begins to look less like a call center and more like agent-to-agent revenue cycle.
We are not fully there yet.
But the Advocate data suggests the patient adoption curve may arrive before health systems have designed the financial operating model around it.
The 7,000% growth number is eye-catching.
The more important signal is that patients are independently attaching AI to provider infrastructure at a pace the provider did not initiate.
For RCM, that has profound implications.
Patients may increasingly use AI to understand the same financial complexity that revenue-cycle teams have historically had to explain manually.
That can be good.
Better-informed patients can mean fewer confusing interactions, faster issue resolution, improved financial-assistance discovery, and potentially better trust.
But it can also mean more scrutiny of estimates, bills, denials, payment policies, and inconsistencies between provider and payer data.
The revenue cycle has spent years automating the provider side of the transaction. The next shift may be that the patient shows up with automation too.
And once that happens, the question is no longer just who pays for the API call.
It is who owns the financial conversation.
Daily coverage of payer policy, denials, deals, and AI developments — written for people who live in revenue cycle.
Source: Becker's Hospital Review — Giles Bruce, “Patients linking MyChart to ChatGPT jumped 7,000% at Advocate. Who pays?” · Additional context from OpenAI Health in ChatGPT · RevCycleAI analysis · October 6, 2026