Artificial intelligence in healthcare is moving beyond answering general questions.
A recent development from OpenAI offers a glimpse of what could come next: AI that can work with authorised information from a patient’s electronic health record and understand the clinical context surrounding a question.
OpenAI has announced that healthcare organisations can connect Epic electronic health records (EHRs) with ChatGPT for Healthcare, allowing authorised users to work with relevant health-record information within their existing workflows.
For those working in UK general practice, the development raises an important question:
Could appropriately governed AI eventually understand not only what a patient or clinician is asking, but the relevant clinical context behind that question?
The answer could have significant implications for the future of AI in NHS primary care.
Why patient context matters for healthcare AI
One of the fundamental limitations of generative AI in healthcare has been context.
An AI system may be able to explain a medical condition, describe a medication or provide general healthcare information. But without access to an individual’s authorised clinical information, it does not necessarily understand the patient behind the question.
Connecting AI with an electronic health record changes that equation.
Instead of considering a question in isolation, appropriately authorised healthcare AI could potentially work with information such as medication history, test results, previous consultations, referrals and other relevant clinical information.
A clinician might, for example, want to understand:
- What has changed since this patient’s previous appointment?
- Have there been recent medication changes?
- Are there results that require review?
- Is a referral or follow-up already underway?
The significance isn’t simply that AI may be able to retrieve information faster.
It is that AI could increasingly operate with context.
That distinction could become important as healthcare organisations explore how generative AI can safely support clinicians and patients.
What has OpenAI actually announced?
It is important to distinguish the current development from the broader possibilities it creates.
OpenAI’s announcement concerns ChatGPT for Healthcare and connected healthcare data sources, including Epic. It does not mean that ordinary consumer ChatGPT can simply access a person’s medical records.
Nor does it mean ChatGPT has been connected to GP clinical systems across the NHS.
There has been no announced integration between OpenAI and EMIS or SystmOne comparable to the Epic development discussed here.
That distinction matters.
The immediate significance for UK primary care is therefore not that this functionality has suddenly arrived in GP practices. It is that the technology demonstrates how generative AI and authorised clinical information can potentially be brought together.
That gives NHS primary care an interesting model to consider.
Could AI eventually connect with GP patient records?
Technically, the concept of authorised systems accessing GP-record information is not new to the NHS.
GP Connect already enables GP practices and authorised clinical staff to share and view GP clinical information between approved IT systems in real time for direct patient care. NHS England describes role-based access controls and defined purposes for this access.
GP Connect functionality can also make structured and coded GP-record information machine-readable by other authorised systems.
This doesn’t mean generative AI can simply be connected to GP records.
It does, however, show that the NHS already has infrastructure and governance mechanisms for appropriately controlled digital access to clinical information.
The more interesting long-term question is therefore:
What could happen if appropriately governed healthcare AI were eventually able to work with that clinical context?
What could record-aware AI look like in NHS primary care?
Consider a patient asking a digital healthcare assistant about their diabetes.
Without access to their clinical record, the AI could provide general information about diabetes.
But imagine an appropriately authorised system that could understand relevant information such as the patient’s current medication, recent blood results and whether a review was outstanding.
The conversation could potentially become much more contextual.
Or consider a patient asking what is happening with a referral.
Rather than simply explaining the referral process, a future system with the appropriate permissions and integrations might recognise that a referral has already been initiated and help the patient understand the appropriate next step.
For clinicians, the possibilities could be different again.
Healthcare AI could potentially help bring together relevant changes in a patient’s record ahead of a consultation, allowing the healthcare professional to review the underlying information more efficiently.
These examples are possibilities rather than descriptions of functionality currently available through GP Chatbot.
But they demonstrate why the combination of AI and electronic health records is potentially so significant.
AI supporting clinicians is different from AI advising patients
There is also an important distinction that shouldn’t be overlooked.
An AI tool helping a healthcare professional find and summarise relevant information is not necessarily the same proposition as an AI system independently interpreting clinical information directly for a patient.
The clinical risks, responsibilities and safeguards can be very different.
Healthcare professionals bring judgement, experience and an understanding of the wider patient situation that cannot simply be assumed from access to data.
That is why the future of AI in general practice should not be framed simply as giving an AI system more information.
The question is how that information can be used safely, appropriately and with clear human responsibility.
Governance and clinical safety will be fundamental
A patient record is fundamentally different from the public information available on a GP practice website.
Any future AI system interacting with GP clinical information would need to address issues including:
Patient authentication and authorisation: Who is requesting the information, and are they entitled to access it?
Information governance: What is the lawful basis for processing the information, and how is patient confidentiality protected?
Data minimisation: Does the AI need access to the entire record, or only the information necessary for a specific task?
Clinical safety: What happens if information is misunderstood, incomplete or presented inappropriately?
Cybersecurity: How is highly sensitive clinical information protected?
Auditability: Can organisations establish what information was accessed, why it was accessed and how it was used?
Human oversight: Where does AI assistance end and professional clinical judgement begin?
These aren’t secondary considerations. They are central to whether record-aware healthcare AI could be deployed responsibly.
The NHS already treats access to patient information as a tightly controlled area. For example, NHS England states that GP Connect Access Record supports authorised access for direct care, while current Update Record functionality is restricted to specifically assured use cases.
From generic AI to contextual healthcare AI
This may ultimately be the bigger development to watch.
The first generation of healthcare chatbots largely focused on providing information or answering predefined questions.
Generative AI made those conversations substantially more flexible.
Access to authorised healthcare context could represent another significant step.
The progression might look something like this:
Information → Conversation → Context → Appropriate Action
Each step potentially makes AI more useful.
But each step also increases the importance of governance, safety and accountability.
For NHS primary care, the goal shouldn’t simply be to make AI more powerful. It should be to determine where AI can genuinely improve the patient and staff experience while maintaining appropriate clinical and information-governance safeguards.
Where does GP Chatbot and AiLeen fit today?
At GP Chatbot, our current work with AiLeen, our AI-powered digital receptionist, focuses on helping patients navigate primary care through natural conversation.
Patients can ask questions about their practice, understand available services, find relevant information and navigate towards appropriate pathways.
Importantly, AiLeen does not currently access or interpret a patient’s clinical record.
That distinction should remain clear.
However, developments such as OpenAI’s work with electronic health records raise a much broader question about where conversational healthcare AI could eventually go.
Could an AI-powered digital interface one day connect the conversation happening at the digital front door with appropriately authorised clinical context?
Could it understand not simply what the patient is asking, but within clearly governed boundaries, the context relevant to that request?
And could that help create a more connected experience between patients, practice teams and healthcare information?
These are not capabilities we are claiming today.
They are questions worth asking about tomorrow.
The future of the GP chatbot may be much more than a chatbot
AI in primary care is developing quickly.
The important conversation is no longer simply whether artificial intelligence can generate convincing healthcare information.
Increasingly, the question is whether AI can understand the right information, in the right context, for the right person. while operating within appropriate clinical and governance boundaries.
For NHS general practice, that creates both considerable opportunity and considerable responsibility.
The next generation of healthcare AI won’t simply need to understand the question.
It will need to understand the context safely, securely and responsibly.
And that is why developments connecting AI with electronic health records are worth watching closely.
Because the future of the GP chatbot may ultimately be much more than a chatbot.