When NHS England published its delivery plan for recovering access to primary care in May 2023, it noted that an average-sized GP practice can receive more than 100 calls in the first hour of opening on a Monday, a peak it tied to practices on analogue lines with a fixed number of lines. Many practices have since moved to digital telephony, but the morning peak has not gone away: under the 2026/27 GP contract NHS England will now collect data on how long callers wait between 8am and 10am.
The national picture is improving, but slowly. In the GP Patient Survey 2026, run by Ipsos for NHS England, 56.9% of patients said it was easy to contact their practice on the phone, up from 52.9% in 2025 and 49.7% in 2024. That still leaves more than four in ten patients who did not say it was easy. At the same time, the 2026/27 GP contract has written several access expectations into the contract itself.
This guide is written for the Director of Patient Access at a GP federation, a primary care network or a large practice group, measured on getting through, same-day handling of urgent need and the survey scores that follow. It reads the 2026/27 contract against an AI voice line: what the contract requires, who decides clinical urgency, where an AI agent can sit and what it must never do. It deliberately cedes neighbouring questions to four posts that already own them. The wider healthcare case and the medical device question sit in our healthcare hub; appointment reminders and missed appointments sit in AI voice appointment scheduling; callback mechanics sit in the callback and virtual queue guide; and the clinical safety case for software sits in the DCB0129 guide.
This guide is shipped by the team behind Dilr Voice, a multi-agent voice AI platform that answers inbound calls, captures what the caller needs and hands over to a person with full context. Or see DATS, our five-stage method for placing AI inside an operation before any build starts.
Why is the 8am rush still the GP access problem in 2026?
The 8am rush is still the GP access problem because demand arrives in one burst while reception capacity is spread across the day. The GP Patient Survey 2026 shows phone access improving to 56.9% of patients finding it easy, but more than four in ten patients still did not say phone contact was easy, and NHS England will now collect data on call waiting time between 8am and 10am.
The survey also shows the shape of the shift. The share of patients who last tried to contact their practice online rose to 30.8% in 2026, from 22.4% in 2025 and 16.8% in 2024, and 57.6% said the practice website was easy to use for contact. Online routes are absorbing a growing share of contact, but the survey does not show the phone becoming optional, which is why any voice AI front door has to serve phone callers well. The survey received 654,714 responses between 2 January and 13 April 2026.
Volume explains why the morning matters so much. NHS England Digital recorded 35.1 million appointments across general practice in September 2026, and 45.0% of them took place on the day they were booked. In our reading, that makes the first contact of the day, by phone or online, the front door for close to half of each day's appointments. For a wider view of how morning peaks behave across sectors, see our guide to capacity planning for peak call demand.
What does the 2026/27 GP contract require of practice phone lines?
The 2026/27 GP contract, set out by NHS England on 24 February 2026, requires practices to deal with clinically urgent requests on the same day, never to ask patients to call back on another day, and to answer non-urgent requests with clear next steps by the end of the next working day. It also aligns the contract with cloud based telephony, and NHS England will collect call waiting data.
The NHS England letter on changes to the GP contract in 2026/27 lists the duties in numbered paragraphs, and five of them shape any phone line, AI or human:
| Contract change (2026/27) | What it means for the phone line |
|---|---|
| Clinically urgent requests, as determined by the practice, dealt with on the same day | The line must get urgent need in front of the practice's own decision process quickly, every morning |
| Practices must not ask patients to call back, or make contact, on another day | "Please ring back tomorrow" is not an acceptable outcome of a call, so callback design matters |
| Non-urgent requests get an appropriate response confirming next steps by the end of the next working day | The practice owes a confirmed next step by the end of the next working day, even when no appointment is offered |
| Data collected on call waiting time between 8am and 10am and during core hours | The morning queue becomes a reported number; NHS England says the aim is understanding, not performance management |
| Contract regulations aligned with existing cloud based telephony requirements | Ask early how any AI line connects to the practice's framework telephony |
Two further changes sit beside these. Online consultation systems must not cap the number of requests during core hours, so online access runs with the same parity as telephone and walk-in access, and practices must display opening times for every mode of access. The 2025/26 contract had already required practices, from 1 October 2025, to keep online consultation tools open throughout core hours of 8am to 6:30pm, subject to safeguards against urgent clinical requests submitted online by mistake, as the 2025/26 contract letter explains.
The call-back rule is not new. NHS England's 2023 recovery plan put it plainly: "It is time to consign to history patients being asked to call back on another day for an appointment." The plan says the 2023/24 contract was strengthened to include the expectation that patients should not be advised to call back another day, and the 2026/27 changes set that out specifically in the contract. For an AI line, that means an agent that ends a call with "try again tomorrow" works against a written contract term, not just good practice.
The duties bind the practice as the contractor. They do not bind a telephony supplier or an AI vendor, and buying an AI line does not move the duty to the supplier. That is the starting point for any buying conversation, and it is why we recommend starting any primary care project with an AI placement diagnostic rather than a product demonstration.
Who decides clinical urgency when an AI voice agent answers the phone?
The practice decides clinical urgency, never the AI voice agent. NHS England guidance on recording same-day appointments says it is for the GP practice to determine which patients are clinically urgent, based on the patient's presenting complaint. An AI voice line can capture that presenting complaint clearly and pass it to the practice's care navigation or duty clinician, but it must not make the urgency decision itself.
This is the single most important design rule for an AI line in general practice. The NHS England guidance on recording same day appointments is explicit that urgency is a practice decision, recorded against the General Consultation Acute national category. In September 2026 there were 6.8 million clinically urgent appointments, and 78.5% of them were dealt with on the day they were booked. Each of those urgency decisions is the practice's to make.
What an AI agent can do well is the part before the decision. It can answer calls as they arrive, take the caller's name, date of birth and reason for calling in their own words, ask the practice's own navigation questions word for word and record the answers without scoring or interpreting them, and place a structured summary in front of the team that decides. Applying the navigation protocol to those answers stays with the practice's navigator. It can hand a caller straight to a person when the practice's rules say so. On Dilr Voice, a warm transfer to a human carries the full context of the call, and every call keeps a full audit trail.
The risk to manage is drift. A navigation script that starts as "collect and route" can slowly become "assess and advise" as staff ask the tool to do more. Write the boundary into the practice's protocol and the vendor's configuration at the start, and review it at every change. Our guide to escalation and human handover covers how to design the handover itself.
Where can an AI voice line sit in Modern General Practice Access?
An AI voice line can sit at the front of Modern General Practice Access, the NHS England model of better digital telephony, online tools and care navigation. It answers and captures each request, routes it into the same workflow as online requests, and leaves assessment and response to the practice team. In our view it should complement framework cloud telephony rather than replace it.
NHS England describes Modern General Practice Access as practices using better online contact tools and telephony and changing their workflow to assess requests and navigate them to the right service or team member, the kind of workflow our DATS methodology maps before anything is built. Practices that fully implemented the approach had overall patient experience scores on average 6 percentage points higher than the national average, according to the same recovery plan. The plan also reports that moving to high-quality digital telephony raised ease of getting through scores by 30 percentage points in trials, driven by features including queuing, call-back, call routing and integration with clinical systems.
The recovery plan notes that the 2023/24 GP contract required practices to procure digital telephony through the nationally set Cloud (digital) Telephony Framework, and the 2026/27 contract aligns the regulations with those cloud based telephony requirements. The sources set no rule on where an AI line connects, but our reading is that it will usually sit alongside the framework telephony system and connect through it, rather than act as a second phone system. Confirm the route with the telephony supplier and ask any AI supplier early how calls reach it and how they return to a person. Our guide to voice AI telephony provider selection sets out the questions that matter. Dilr Voice telephony runs on Twilio or Vobiz, so the connection route is a question to settle in scoping.
A second placement is the overflow. Some practices will want the AI line only when every human call handler is busy, or only in the first two hours of the day. Those windows are usually set in the practice's telephony routing, while on the AI side business-hours configuration and after-hours routing decide what the agent does in each period. The same principle applies after core hours, which our out-of-hours call handling guide covers.
What should an AI voice agent never do on a GP phone line?
An AI voice agent on a GP phone line should never decide clinical urgency, give clinical advice, tell a caller to ring back another day, or keep a caller who describes an emergency on the line. Each of these either breaks the 2026/27 contract or moves clinical judgement onto software the practice has not assessed for that purpose. The safe design captures, routes and hands over.
Set these as written red lines in the configuration and the protocol:
- No urgency decision. The agent records the presenting complaint and the answers to the practice's questions. The practice's navigator or duty clinician decides.
- No clinical advice. The agent can read out practice information drawn from the practice's own documents, such as opening hours or how to order a repeat prescription, but it does not suggest what a symptom means or what to take.
- No call back another day. The contract says practices must not ask patients to call back, or make contact, on another day. Our design rule is that every completed call ends with a next step: a person now, a booked slot, a confirmed request in the workflow, or a stated time by which the practice will respond. The contract term applies to an AI line exactly as it applies to reception.
- No delay on an emergency. Configure the agent so that when a caller describes signs of an emergency, it stops and directs them to 999 at once. Our guide to emergency handling on an AI line sets out the recognise, stop and redirect pattern.
- No dead end for the patient who wants a person. The access plan asks practices to respect a patient's preference for a call, face to face or online, so a caller who asks for a human should reach one.
The fifth rule matters most for patients who depend on the phone. Callers with low digital confidence, hearing loss, a learning disability or limited English may have no easy alternative route, so the line has to work for them first. Dilr Voice supports 30+ languages, which helps, but language support does not replace an equality review. Our guide to voice AI accessibility and the Equality Act covers what that review should test.
The same diagnostic logic underpins our AI operating model consulting, which sets who owns the protocol, the script and the review once an AI line is live.
What clinical safety and data duties apply before an AI voice line goes live?
Before an AI voice line goes live, the practice as the deploying health organisation must meet DCB0160, the clinical risk management standard for deploying health IT, while the supplier carries DCB0129 as the manufacturer. Both are mandatory under the Health and Social Care Act 2012. The practice also needs a lawful basis for health data, a data protection impact assessment and DSPT evidence.
NHS England Digital's clinical risk management standards say DCB0160 requires a health organisation to establish a framework within which the clinical risks of deploying a new or modified health IT system are properly managed. The duty sits with the practice as the deploying organisation, whatever the supplier provides. In our view, the hazards to analyse for an AI phone line are concrete: a missed red flag, a mistranscribed name or date of birth, a dropped handover, and a caller who could not get through to a person. Our DCB0129 and DCB0160 guide explains how a hazard log and safety case are built, using a different AI system as its worked example.
The data side is just as concrete. A caller describing symptoms is sharing health data, which is special category data under UK GDPR, so the practice needs an Article 9 condition as well as a lawful basis, should complete a data protection impact assessment, and needs contracts that name every processor. Recording consent, permitted retention and an audit trail on every call are configuration questions to settle before go-live. Our UK and EU voice AI compliance guide sets out the wider rules, and the healthcare hub covers DTAC, DSPT and CQC expectations in one place.
Ask any supplier for its own evidence on each of these points rather than accepting a general assurance. Dilr Voice is hosted on Google Cloud Platform, encrypted at rest and in transit, with dedicated tenancy and regional data residency options for enterprise customers, and those facts belong in the practice's impact assessment, not in a sales slide.
How should a practice measure whether an AI voice line is helping?
A practice should measure an AI voice line against the numbers NHS England will collect under the 2026/27 GP contract: call waiting time between 8am and 10am, call waiting time across core hours, and the share of clinically urgent patients seen the same day. Add the practice's own GP Patient Survey phone score and a few call-level measures, and compare each against a baseline.
The contract letter lists five demand metrics NHS England will collect: call waiting time between 8am and 10am, call waiting time during core hours, the percentage of clinically urgent patients seen on the same day, and the percentage of non-urgent patients seen within one week and within two weeks. Those five give a practice a ready-made scorecard, and recording them before an AI line goes live gives a baseline to compare against. Setting that baseline first is a step we build into our approach.
Then add three call-level measures of your own. First, the share of AI-answered calls that reached a person, and how quickly, because a falling handover rate can mean the agent is working or that patients are being kept from staff. Second, the share of completed calls that ended without a confirmed next step, which should be driven as close to zero as possible. Third, complaints and significant events that mention the phone line. Our AI voice ROI framework shows how to turn measures like these into a business case a partnership or federation board can test.
Be honest about what the numbers can show. The GP Patient Survey runs once a year, with fieldwork from about January to March (the 2026 round ran from 2 January to 13 April), so a change made in November will not show in the survey until the next round's results. Ipsos publishes results down to individual practices, which makes the practice's own phone score a useful annual check. Weekly telephony data is the faster signal. McKinsey's State of AI found that 88% of organisations use AI in at least one function while about a third have it in production, so a measured baseline is worth setting before any pilot starts.
What is the best AI voice agent for GP practices in 2026?
The best AI voice agent for a GP practice in 2026 is the one that works through the practice's framework telephony, captures requests without deciding urgency, hands over to a person with full context, supports the languages its patients speak, and arrives with clinical safety and data evidence the practice can use. For many practices, better telephony and care navigation alone will be the right first step.
Applied honestly, those criteria favour different answers for different practices. A practice still struggling with engaged tones should first make full use of its framework telephony supplier's features, such as queuing and call-back, which the recovery plan lists among the drivers of a 30 percentage point rise in ease of getting through in trials. A practice with a strong care navigation team and a manageable queue may gain more from extra navigator hours in the first two hours than from any AI line, a trade-off our guide to blended human and AI staffing explores. An AI voice agent earns its place where the morning queue still outruns staff after those steps, where callers need several languages, or where the federation wants a consistent front door across sites.
When comparing suppliers, apply the same tests to each. Ask how calls reach the AI line from the practice's telephony, how a handover reaches a named person, what the supplier provides for the DCB0160 safety case, where data is processed, and how the agent is stopped from giving clinical advice. Other voice AI suppliers a federation may compare include PolyAI, NiCE Cognigy and Genesys, and NHS buying routes such as the NHS SBS AI framework can shape which suppliers a trust or federation buys from. Our best AI voice agent guide compares the wider market, and our enterprise AI voice agents guide explains how a production deployment is put together.
Dilr Voice fits where a practice group wants a multi-agent inbound front desk that draws practice information from its own documents through a RAG knowledge base, applies recording consent and audit rules by default, and transfers to a person with context. It does not fit a practice looking for software to triage patients, because that is the decision the contract leaves with the practice.
Frequently asked questions
Can an AI voice agent book GP appointments directly?
An AI voice agent can book GP appointments directly only where it is connected to the practice's appointment system and the practice allows it to book specific slot types. Many practices will start with the agent capturing requests into the care navigation workflow instead, because that keeps the urgency decision and the slot choice with staff while the integration and safety case are proven.
Booking is also where appointment categories matter. NHS England's guidance asks practices to map clinically urgent slots to the General Consultation Acute category, so any automated booking has to respect that mapping or the same-day data will be wrong. Confirm with any supplier which appointment systems it can connect to before relying on direct booking, and ask our team through the contact page if you want a second view.
Will patients who prefer the phone still reach a person?
Patients who prefer the phone should always be able to reach a person, and a well-designed AI voice line makes that easier rather than harder. NHS England's access plan asks practices to respect a patient's preference for a call, a face-to-face appointment or an online message. An AI line should offer a human early, transfer with context, and never leave a caller without a next step.
In practice, that means a short path to a person in the opening seconds of the call, a transfer that passes the caller's details so they are not asked again, and a tested route for callers who cannot use speech recognition easily. Configure the line so that a request for a person is always honoured, and check the rate at which it happens every week.
Want to see this in production? Try Dilr Voice live, compare the market in our best AI voice agent guide, see our approach to placing AI inside clinical operations, or browse more industry AI guides for UK teams.
Take the queue out of 8am.
30-min scoping call · No deck · Confidential. We will tell you whether an AI line belongs in your access model, and what your safety case needs before the first call.
Written by the Dilr.ai engineering team, practitioners who ship enterprise AI in production. Read more about Dilr.ai. Follow us on LinkedIn for shipping notes, or subscribe via the RSS feed.