Morgan Beschle is the Vice President of Product Management at RevSpring.
The way consumers search for care is changing faster than most healthcare organizations can react. Increasingly, the first touch point is AI-mediated: a stand-alone assistant such as ChatGPT, Claude, Perplexity or Google AI Mode; a voice agent answering for a call center; a conversational search on a payer’s or provider’s own site; a recommendation surfaced inside a third-party app; or a referral from inside the EHR.
Health systems and health plans have relied on their websites as the digital front door for years. That’s no longer the only way in, and for a growing share of consumers, it isn’t even the first stop. Organizations need to think beyond their websites. They need to make sure their data shows up wherever care decisions happen, and separately, make sure the site itself is structured, with schema and machine-readable provider attributes, for an AI agent to parse and act on. Getting that right gives an organization a better shot at becoming the source AI models cite and route patients toward.
Health systems need to focus on getting found, and accurately, for the service lines they’re trying to grow. For health plans, it’s about guiding members toward the highest-value options that meet their goals and preferences, keeping them in-network with correct costs and surfacing alternatives like care programs and virtual visits.
Consumers are making care decisions with AI, but payers and providers aren’t ready.
The patient journey has shifted before. Patients used to have a primary relationship with a single physician, someone who knew them and their family, reachable by one phone number. That gave way to finding care through a health system or provider website. Now, a growing number of patients start with an AI assistant instead.
When an AI model can’t confirm network status, the best clinically appropriate match, who is accepting new patients or real-time availability, it typically does one of two things. It either skips the recommendation altogether or guesses from general web content in a way that misleads the patient.
Consider a patient searching for help with ankle pain. A provider’s primary specialty may be orthopedic surgery, but their true clinical expertise and primary practice is total knee replacements. Health systems are typically able to capture and surface this level of clinical-expertise granularity, pointing patients in the right direction. Health plans generally don’t have that same data. If a payer’s directory can’t make that distinction, the patient sees a recommendation that looks right and isn’t, and the health plan loses the ability to guide members to the right care at the right cost. Multiply that error across an entire directory, and a data quality problem becomes a systemic trust issue.
A consumer on a narrow-network plan needs to know if a provider is in-network for their exact plan, not just affiliated with the payer generally, and that plan-level detail changes constantly across thousands of provider-plan combinations. If it isn’t structured for an AI model to read, the model gives the member the wrong answer, and the member loses trust in the network itself.
Closing the gap is a two-step process. First, organizations need to make their existing data machine readable. Then, they need to build their own AI-ready assets to guide consumers directly.
Step One: Make the website machine readable.
Provider and plan data typically lives scattered across EHRs, credentialing systems and benefit design systems, none of which were built for AI to read. If a healthcare website doesn’t publish structured clinical-specialty data, accurate provider bios, network status and location data in a form AI can parse, the model can hallucinate, guess from general web content or move on to a competitor’s site.
Organizations can’t control what AI model a consumer uses or when. They can control whether their own data is structured well enough to be read correctly.
Step Two: Build an AI-ready data asset.
The bigger opportunity is building a structured, curated data layer such as a Model Context Protocol (MCP) server to be the source of truth for AI agents used in call centers, on system or payer sites or by third parties. Owning that asset means owning the economics, answering high-volume, routine questions without paying per interaction through a third party and owning the quality of every recommendation, including business rules like prioritizing an employed provider over an outside referral, something no outside AI model would replicate on its own.
Ultimately, that grounding data layer is what lets AI deliver care recommendations patients and members can trust.
Accommodate the online shift.
Health systems and health plans need to take stock of how their data is stored and when it was last updated, then fix that foundation before implementing AI. A wrong recommendation means a mismatched specialist for a health system or an inaccurate cost estimate for a health plan, and either lands on the patient.
The work doesn’t stop at an organization’s own site. As more consumers route care decisions through an AI assistant, fewer will visit websites directly. That means structuring network and provider data so AI platforms can find and interpret it accurately, rather than waiting for them to find it on their own. Organizations that prioritize this now will shape what finding care looks like in the years ahead.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?
