Artificial intelligence is moving into patient billing and payment outreach as medical debt remains common among US households, but the available evidence does not show that AI is driving the debt burden or that its use is widespread across the collections industry. Providers are marketing tools that answer billing questions and contact patients, while separate AI-assisted coding systems may affect how claims are prepared. Those are distinct functions, and the evidence does not establish that every AI billing tool is a debt collector.
An Urban Institute analysis published in October 2026 found that 26.6% of adults surveyed said they or a family member owed medical debt. The analysis drew on a December 2025 survey of more than 10,000 adults. It found that 15.4% reported past-due bills or bills they could not pay. Among respondents with medical debt, 35.2% said a collection agency had contacted them. These are survey responses, not a count of individual debts or collection accounts.
AI enters patient billing
One documented example is Cedar’s Kora platform, announced in August 2026. The company describes it as a system of AI agents that handles billing questions and patient outreach by voice and text, alongside payment, insurance and financial-assistance workflows. The company also reports usage and performance figures, but those claims have not been independently evaluated in the evidence available here.
Tools like these could change how providers communicate about balances and process payments. But the supplied evidence does not establish how extensively such platforms are being used across US health care, whether they increase the amount collected, or how often patients are directed to human assistance. Nor does it independently verify vendors’ claims about improved resolution or lower costs.
That distinction matters because patient-facing outreach is not the same as the process that determines what appears on a bill. AI-assisted coding is another part of the issue: coding tools can affect how diagnoses and services are represented in claims, while outreach agents handle communication and payment workflows. The available material does not show that these functions are routinely combined in one system.
Debate over coding and costs
The Blue Cross Blue Shield Association said in September 2026 that its claims analysis estimated rising complex coding added $942 million in spending for its member companies between 2023 and 2025. The association linked the trend to increased use of AI-enabled tools. Its analysis does not, by itself, establish that AI caused the entire increase or that the coding was inaccurate.
The American Hospital Association disputed the inference that such analyses demonstrate AI is driving inappropriate billing. That is a stakeholder response, rather than an independent determination of the coding practices or their effects. The evidence therefore supports describing a contested debate about coding and costs, not a settled conclusion that AI has caused improper bills.
Separating the mechanisms also helps clarify what remains unknown. A coding tool may influence the information submitted in a claim; an outreach tool may affect how a patient is contacted about an amount already billed. The supplied evidence supports the existence of both types of activity, but does not establish a direct causal link between AI coding and the growth of household medical debt.
Accuracy and patient recourse
The Consumer Financial Protection Bureau warned in a 2024 advisory opinion that medical-debt collectors may not have timely access to providers’ billing and payment records. That can raise the risk of pursuing an amount that has already been paid or is otherwise not owed. The advisory also described consumers’ difficulties obtaining verification and documentation, including information about insurance payments and hospital financial assistance. It did not attribute these problems to AI.
That existing concern frames questions for any automated billing system: what information determines outreach, how patients can correct account data, and whether they can dispute a balance or seek assistance. The research materials do not answer how Cedar’s platform, or other tools, handle each of those situations. They also do not provide independent evidence about whether automation makes errors easier or harder to resolve.
The debt figures and the adoption of billing technology describe related pressures, but they do not establish that one is causing the other. The evidence documents substantial self-reported medical debt and the introduction of AI tools into some patient billing workflows. It leaves unresolved how broadly those systems are used, what results they produce independently, and how safeguards operate when a bill is disputed.