People no longer exclusively begin a health journey at a search engine (Google) or a provider website. Increasingly, patients are opening an AI chatbot and asking. Rock Health's 2025 Consumer Adoption of Digital Health Survey put a number on it: 32% of consumers have now used an AI chatbot for health questions, double the 16% reported a year earlier. Among those users, 64% turn to these tools for health questions weekly or more often, demonstrating habit formation.
Most importantly, patients are starting to take direction from these experiences: 81% of AI chat users reported taking at least one relevant action after a health-related chatbot conversation and 40% consulted a provider. Others changed or made an appointment, tried a new health behavior, or adjusted a medication. People are forming health intent inside a chat window and then carrying that intent somewhere else to act on it.
The question for every health system marketing and digital executive is whether they are in the room when that moment arrives. Right now, they are not.
General-purpose assistants dominate these conversations, with 23% of consumers using ChatGPT for health questions and 15% using Gemini. Provider and payer sponsored chat experiences remain in the single digits, at 5% and 4% respectively. The patient is establishing a pattern and forming intent inside ChatGPT, Claude, Copilot, and Gemini, while the health system is left hoping patients will come to their portal.
Where the experience breaks today
The breaking point is not the moment the patient asks a good question. The models handle them with precision. The break occurs when the patient is ready to do something and the model has nowhere to send them. Today, the models can't close the loop.
Consider a few common experiences:
- A patient inputs her records into the LLM, asks the assistant about a recent lab result, and learns she is overdue for a follow-up with her primary care doctor. The AI assistant gives her sound advice: Book a visit with your PCP. Then the conversation stops. She is left to open a separate portal, remember a password she set up two years ago, search a provider directory, guess at the right department, and find a slot that works. Each of those steps creates friction and provides an opportunity to give up, and many people do.
- A patient recovering from a procedure asks how to refill a prescription that is out of stock at their local pharmacy. The AI assistant can reason through the question and suggest he message his care team or move the prescription to another pharmacy. But it cannot actually open a message to his provider or move the order. It can only tell him to go do it himself, through a login it cannot access.
- A patient researching a health worry concludes they should get a specific lab test. The AI assistant can explain why the test makes sense and what the results would mean. It cannot show the patient where to get it, what it costs, or how to book it. The patient is left to start a new search from scratch, often landing on whatever option ranks highest rather than the one connected to their own health system.
In every one of these cases, the AI chat experience did its job. It understood the need and directed the user to the right next step. The experience failed in completing the handoff. The patient was handed back to a phone tree or a portal login to act, and by then, the friction has set in. The intent that formed in the conversation leaks away, and when it does land somewhere, it often lands at a more convenient competitor option rather than the patient's own health system.
The unsolved problem is the data and action layer
In the first half of 2026, every major AI platform raced to connect a patient's records into the experience. OpenAI launched ChatGPT Health with b.well. Anthropic launched Claude for Healthcare with HealthEx the same month. Microsoft followed with Copilot Health and Google relaunched its Fitbit app as Google Health, a Gemini-powered wellness coach that pulls in wearable data and medical record summaries. Each has solved the same half of the problem, bringing data into the conversation. None solved the other half, getting the patient to act.
The hard part is not the AI. The models are already good at interpreting a lab result or suggesting a next step, and are getting better. The hard part is connecting everything that comes out of the conversation. Return to the patient who is overdue for a follow-up. For that insight to become a booked appointment, several things must be known. The conversational AI has to know who the patient is with enough assurance to act on their behalf. It has to have their consent, captured and auditable. It has to reach into the right health system or ancillary services provider and see real availability. And it has to write the booking back into the system of record.
That is a data and identity problem, not a model problem, and it is exactly what has kept health systems out of these experiences. Health systems control the inventory every AI conversational experience needs, including appointments, programs, third party services (lab, pharmacies, PT / OT / imaging) and clinicians, and they will not expose it to a consumer AI without infrastructure they trust for identity, consent, and governance. The organizations that close that gap will be the ones that bring a health system's own identity and consent framework into the experience, rather than asking the patient to hand their data to a technology company.
Where Praia fits
Praia is the patient identity, context, and orchestration layer for health systems. It already connects scheduling, care programs, benefits, third party service partners, and virtual care across a health system's digital footprint, sitting alongside the EHR rather than replacing it. That is the layer the consumer AI experiences are missing. It is also not theoretical: Praia already powers more than 6 million patient accounts and a 67% monthly retention rate at one of the country's largest health systems, so an AI handoff would build on a layer that already works at scale, not start from scratch.
Praia serves as the rails that connect a specific health system into a consumer AI conversation, carrying the patient's verified identity and consent, surfacing real inventory, and completing the action inside the system of record. The AI identifies the need. Praia executes the action. The health system maintains the patient relationship and governance over its own data. The result is a closed loop journey: a patient asks a question, the assistant identifies a real care need, and the patient books with their own health system without leaving the conversation.
We are working through this now with a small group of early adopters who want to be present at the moment of intent rather than downstream of it. The window is open before these patterns harden and before someone else becomes the default path from an AI conversation into care. If your organization is thinking about how to be present in that moment, we would like to talk.