Measuring UK Health AEO: What ChatGPT Cites

Health is turning into an “answer-first” surface. Major AI products are starting to ship health experiences, and so far these launches appear to be US-first in many cases. The UK timeline is unclear, but the direction is set: AI answers are becoming part of how people start health journeys.

That is why Tilio is launching an AEO Index for Health across the UK: a recurring benchmark that tracks NHS vs non-NHS visibility, citations, and intent coverage.

You can learn more about this and other Health AEO features here.

. This initial report uses Exeter as a controlled starting point, with answers collected using ChatGPT 5.2.

Perplexity has also been moving into health-related experiences and cited answering. Even where specific “health products” are not UK-available yet, people already use these tools for triage, booking, eligibility, and “where do I go” routing.

Why do this research

Two reasons.

• Public outcomes: faster, clearer routing and fewer dead ends (especially for urgent care, mental health, and access constraints).
• System insight: if AI answers become a front door, we should measure what sources get pulled in, what gets missed, and how this changes month to month.

This is not “how the NHS can win.” It’s measurement: what AI answers contain, what they cite, and what that implies for service information design.

Method

Each section follows the same pattern: a real-world prompt, the model output, then an audit of mentions and/or linked citations (depending on how the response was formatted). The report then classifies each source as NHS (e.g., NHS.uk, trusts, NHS-commissioned services) or non-NHS (e.g., charities, private providers, directories, commercial sites).

Important nuance: some answers contain many unlinked organisation mentions, while others include explicit linked citations. Both shape visibility, but they behave differently in AEO.

Headline numbers

The mix is intent-driven.

Where non-NHS dominates: mental health support prompts pull in directories, charities, and private counselling heavily (often by design, because that ecosystem is mixed). For teen anxiety support, the report counted 21 organisations: 6 NHS (28.6%) and 15 non-NHS (71.4%).

Where NHS dominates: tightly-defined service routing (maternity booking, maternity coverage, visiting policy) is almost entirely NHS-cited in this dataset. For pregnancy booking, the report counted 7 NHS organisations (100%).

Timing note on “Health” features: if and when consumer-facing “AI health” experiences arrive in the UK, the most likely path is phased rollout. Even before that, UK users already treat general AI answers as a starting point. That makes measurement useful now, not later.

Intent patterns

The prompts cluster into a small set of intents. The interesting part is how intent changes what gets cited.

• Urgent triage and routing: tends to concentrate on NHS 111, trusts, and official pathways. Broken-arm triage was 100% NHS citations (7/7).
• Access constraints (“I can’t get an appointment”): pulls in a mix of NHS and non-NHS (Walk-in/UTC pages, Healthwatch, directory-style sources). For “I can’t get a GP appointment and I’m getting worse,” the report shows a 50/50 split (2 NHS, 2 non-NHS).
• Mental health support and signposting: consistently expands beyond NHS domains because charities, helplines, and private therapy are structurally part of the answer space (not “competition,” just reality). Teen anxiety support landed at 71.4% non-NHS.
• Preventative care (vaccines/screening): often introduces pharmacies and private clinics alongside NHS eligibility rules. Flu jab sourcing skewed 69.2% non-NHS (9/13).
• Specialist pathways (ADHD, long COVID): tends to mix official NHS routes with condition advocacy sites, private assessments, and clinical explainers. ADHD is a good example of why “counting method” matters (unique vs repeated citations).
• Rights and logistics (interpreters, visiting): mostly NHS, but often supported by community information sources. Interpreter access: 57.1% NHS (4/7).

Prompt-by-prompt snapshot

Notes on how to read this:

• Each prompt lists the measurement basis used in your snapshot (citations, sources, or unique organisations).
• NHS vs non-NHS percentages are computed from the counts shown for that prompt.
• “What the answer returned” is a short paraphrase plus a short excerpt where you provided one.
• “Analysis” is a short AEO interpretation focused on intent and why the source mix looks the way it does.

Urgent care routing

Prompt: “I think my child has a broken arm. What’s the right place to go in Exeter?”

• Intent: urgent triage and routing (injury, child)
• Measurement basis: citations
• NHS vs non-NHS: 7 NHS, 0 non-NHS (100% NHS, 0% non-NHS)
• What the answer returned: escalation thresholds (red flags), then a routing ladder (A&E vs MIU/UTC), then NHS 111 as the decision support layer
• Response excerpt: “If there’s deformity, severe pain, or numbness, go to A&E. Otherwise consider MIU/UTC and call NHS 111 for routing.”
• Analysis: this is a high-risk prompt with a clear safety frame. Answers tend to cite a small set of authoritative pathways and avoid optional providers. For AEO, this class is about consistency and extractability of routing rules, not provider discovery.

Prompt: “I have chest pain that comes and goes. Should I go to A&E in Exeter?”

• Intent: urgent triage and routing (high-risk symptom)
• Measurement basis: citations
• NHS vs non-NHS: 5 NHS, 0 non-NHS (100% NHS, 0% non-NHS)
• What the answer returned: immediate escalation guidance (999/A&E triggers), then secondary routing (111) if uncertainty remains
• Analysis: the model compresses the source set because the job is safe escalation. AEO work here is mostly about ensuring official guidance is short, stable, and unambiguous.

Prompt: “I can’t get a GP appointment and I’m getting worse. What are my options in Exeter?”

• Intent: access friction plus escalation (worsening symptoms, blocked primary care)
• Measurement basis: sources in the tally
• NHS vs non-NHS: 2 NHS, 2 non-NHS (50% NHS, 50% non-NHS)
• What the answer returned: a route ladder (111, walk-in/urgent treatment, out-of-hours, A&E if severe), plus additional access endpoints beyond standard GP booking
• Analysis: mixed intent widens retrieval. The answer needs both safety routing and operational workarounds, so non-NHS “how to access care” pages can enter alongside NHS pages.

Mental health support

Prompt: “My teenager is struggling with anxiety. What mental health support is available in Exeter?”

• Intent: youth pathway discovery plus support options
• Measurement basis: unique organisations
• NHS vs non-NHS: 6 NHS, 15 non-NHS (28.6% NHS, 71.4% non-NHS)
• What the answer returned: a blended set (NHS routes plus charities, youth services, helplines, directories, private options)
• Response excerpt: “Start with NHS routes (CAMHS/MHST/TALKWORKS), but charities and local counselling fill gaps when waits are long.”
• Analysis: this prompt triggers options expansion. For AEO, expect high mention diversity; tracking “who gets listed” is as important as “who gets cited.”

Prompt: “I’m feeling low after having a baby. What mental health support is available in Exeter?”

• Intent: perinatal/postnatal pathway navigation
• Measurement basis: unique organisations
• NHS vs non-NHS: 8 NHS, 0 non-NHS (100% NHS, 0% non-NHS)
• What the answer returned: a defined pathway set with clear NHS entry points and escalation guidance
• Analysis: the source set stays narrow when the pathway is well defined and consistently documented. This is a useful “low-noise” cohort for monitoring citation drift over time.

Prompt: “I want therapy without going through my GP in Exeter. What are my options?”

• Intent: self-referral and access pathway
• Measurement basis: unique organisations
• NHS vs non-NHS: 6 NHS, 1 non-NHS (85.7% NHS, 14.3% non-NHS)
• What the answer returned: NHS self-referral routes first, then a smaller set of non-NHS alternatives
• Analysis: “pathway clarity” prompts tend to concentrate on whichever source explains steps and eligibility cleanly. When a national route exists, NHS share increases.

Screening, vaccines, clinics

Prompt: “I’m pregnant, how do I book my first appointment in Exeter?”

• Intent: pathway initiation (maternity booking)
• Measurement basis: sources in the tally
• NHS vs non-NHS: 7 NHS, 0 non-NHS (100% NHS, 0% non-NHS)
• What the answer returned: stepwise initiation (how to book, what happens next), using official entry points
• Analysis: this is a high-intent onboarding question with a defined owner. Source diversity stays low when the booking route is explicit.

Prompt: “What maternity services cover Exeter?”

• Intent: service boundary and catchment discovery
• Measurement basis: citations
• NHS vs non-NHS: 7 NHS, 0 non-NHS (100% NHS, 0% non-NHS)
• What the answer returned: named service coverage and access guidance anchored to official pages
• Analysis: boundary questions tend to stay official unless coverage rules are fragmented or inconsistently published.

Prompt: “Where can I get a flu jab in Exeter?”

• Intent: service discovery plus booking
• Measurement basis: sources in the tally
• NHS vs non-NHS: 4 NHS, 9 non-NHS (30.8% NHS, 69.2% non-NHS)
• What the answer returned: an options list (often pharmacies and private booking endpoints) alongside eligibility framing
• Analysis: availability and booking convenience drive non-NHS inclusion. This is a good cohort for tracking seasonal shifts and new providers entering answers.

Prompt: “I’m traveling to Thailand. Where can I get travel vaccines in Exeter?”

• Intent: service discovery plus booking, with eligibility context
• Measurement basis: sources in the tally
• NHS vs non-NHS: 6 NHS, 4 non-NHS (60% NHS, 40% non-NHS)
• What the answer returned: baseline guidance plus provider options; mixed sourcing is structurally likely
• Analysis: the model often splits “what you need” (authoritative guidance) from “where you can get it” (operational access). That produces stable mixed-source answers.

Prompt: “Where can I get cervical screening (smear test) in Exeter?”

• Intent: booking and access navigation for screening
• Measurement basis: sources in the tally
• NHS vs non-NHS: 3 NHS, 3 non-NHS (50% NHS, 50% non-NHS)
• What the answer returned: NHS route plus non-NHS alternatives, commonly framed around access or booking pathways
• Analysis: a 50/50 split suggests the answer is balancing official pathway description with accessible booking options. Small wording changes (“no invite”, “overdue”, “missed appointment”) can move this mix.

Specialist pathways and ongoing conditions

Prompt: “How do I get assessed for ADHD as an adult in Exeter?”

• Intent: specialist pathway boundary (referral rules, waiting, alternatives)
• Measurement basis: citations, reported two ways
• Unique linked citations: 2 NHS, 4 non-NHS (33.3% NHS, 66.7% non-NHS)
• Linked citations including repeats: 2 NHS, 6 non-NHS (25% NHS, 75% non-NHS)
• What the answer returned: pathway explanation plus non-NHS resources and private routes that fill information gaps
• Analysis: repeated-citation concentration matters here. A single non-NHS explainer can dominate frequency. An index should track both unique domains and citation frequency to avoid missing this effect.

Prompt: “I think I might have long COVID. Where can I get help in Exeter?”

• Intent: ongoing condition support plus pathway navigation
• Measurement basis: sources in the tally
• NHS vs non-NHS: 3 NHS, 5 non-NHS (37.5% NHS, 62.5% non-NHS)
• What the answer returned: pathway guidance plus support resources, often including charities and community support
• Analysis: non-NHS share rises when practical support content is widely published outside official domains. The key measurement question is whether official pathway pages remain visible alongside the support ecosystem.

Prompt: “I have headache after sport. Should I be worried and where can I go in Exeter?”

• Intent: symptom triage plus routing (red flags, escalation)
• Measurement basis: sources in the tally
• NHS vs non-NHS: 3 NHS, 2 non-NHS (60% NHS, 40% non-NHS)
• What the answer returned: red flags, escalation guidance, routing options; includes some non-NHS informational references
• Analysis: this sits between triage and general health education. Non-NHS sources often enter as symptom explainers while NHS sources anchor routing and escalation steps.

Access, rights, admin

Prompt: “I need an interpreter for my hospital appointment in Exeter. How do I arrange that?”

• Intent: access support and rights/process
• Measurement basis: sources in the tally
• NHS vs non-NHS: 4 NHS, 3 non-NHS (57.1% NHS, 42.9% non-NHS)
• What the answer returned: how to request an interpreter and what to say; often includes patient-facing guidance beyond pure policy pages
• Analysis: these prompts often pull in patient bodies and community explainers alongside official policy. Clarity of process language can reduce dependence on third-party interpretation.

Prompt: “What’s the visiting times policy for the hospitals near Exeter?”

• Intent: operational info lookup
• Measurement basis: sources in the tally
• NHS vs non-NHS: 8 NHS, 0 non-NHS (100% NHS, 0% non-NHS)
• What the answer returned: directs to trust policy pages and notes ward-level exceptions
• Analysis: operational prompts have a clear authoritative owner, so source diversity stays low. They are still high-change pages, so drift monitoring is useful.

Prompt: “How do I change my GP surgery if I moved house in Exeter?”

• Intent: admin/process navigation
• Measurement basis: sources in the tally
• NHS vs non-NHS: 4 NHS, 0 non-NHS (100% NHS, 0% non-NHS)
• What the answer returned: steps for switching, what happens to records, and where to apply
• Analysis: admin tasks tend to be NHS-dominant because the process is system-owned. Non-NHS sources usually appear only when the prompt shifts toward disputes or escalation.

Lifestyle and community services

Prompt: “I want to stop smoking. What support is available in Exeter?”

• Intent: behaviour change support plus service discovery
• Measurement basis: sources in the tally
• NHS vs non-NHS: 6 NHS, 5 non-NHS (54.5% NHS, 45.5% non-NHS)
• What the answer returned: NHS-commissioned support routes plus non-NHS programmes/options
• Analysis: lifestyle prompts behave like programme discovery. Even when NHS routes exist, the answer often lists alternatives to increase coverage.

Prompt: “How do I get help with weight management in Exeter?”

• Intent: service discovery plus programme selection
• Measurement basis: sources in the tally
• NHS vs non-NHS: 7 NHS, 6 non-NHS (53.8% NHS, 46.2% non-NHS)
• What the answer returned: NHS routes plus local groups and private providers, framed as options and eligibility
• Analysis: near-even splits are common in “help me choose” prompts. Mention-share is often a stronger proxy for user action than citations in this category.

Prompt: “How do I access NHS continence services in Exeter?”

• Intent: pathway navigation (referral and service access)
• Measurement basis: citations/sources in the tally
• NHS vs non-NHS: 5 NHS, 1 non-NHS (83.3% NHS, 16.7% non-NHS)
• What the answer returned: referral route and service description, plus a supplementary non-NHS informational resource
• Analysis: small non-NHS shares here often represent “supporting education.” If the non-NHS share rises over time, it can indicate missing or decaying official pathway content.

What this means for health AEO

The clearest finding is not “NHS good” or “private good.” It’s this: AI answers follow information architecture. When pathways are crisp, official sources dominate. When pathways are fragmented (mental health, lifestyle support), AI pulls in a wider ecosystem.

Three practical implications:

• Intent coverage beats content volume. “Where do I go?” and “how do I book?” prompts reward pages that include eligibility, next steps, and real contact routes.
• Counting has to be explicit. The ADHD prompt shows how results change depending on whether you count unique sources or repeated citations.
• Non-NHS visibility is not always “competition.” In mental health support, charities and directories are functionally part of the care surface, so they appear in answers. The measurement point is whether the answer still routes safely and correctly.

How organisations can act (without gaming it):

• Publish route-first pages: “If X then Y” decisioning, with thresholds, timelines, and the “what to do right now” actions.
• Make service boundaries machine-readable: catchment, eligibility, referral mode (GP vs self-referral), and escalation routes.
• Expose the operational details AI needs: phone numbers, hours, appointment methods, and what to do when access fails.
• Design for mixed ecosystems: for mental health, explicitly include safe signposting to charities and crisis routes alongside official pathways.

Finally, the “UK timing” question. Health-specific AI experiences have started shipping in the US. It is hard to forecast exactly when an equivalent surface will arrive for UK users. What is already true is that general-purpose AI answers are being used for the same intents today, which is why the Health AEO Index is being launched now.

FAQs

What counts as “NHS” vs “non-NHS” in this study?

NHS includes NHS.uk, NHS trusts, and explicitly NHS-commissioned services. Non-NHS includes charities, private providers, commercial health brands, and directories (even if they list NHS options).

Why do mental health prompts cite so many non-NHS sources?

Because the support ecosystem is mixed: helplines, charities, school support, and private counselling sit alongside NHS pathways. In the teen anxiety prompt, non-NHS organisations made up 15 of 21 counted sources (71.4%).

Are citations or mentions more important for AEO?

Both. Citations signal “grounding” and source authority, while mentions drive brand/entity recall. The ADHD prompt shows how the story changes depending on whether you count unique citations or repeated links.

What’s the most actionable thing a trust or provider can do?

Publish route-first, intent-specific service pages (eligibility, referral mode, what to do now) and keep them operationally current.

What should be tracked in the Health AEO Index?

NHS vs non-NHS share by intent, top cited domains, which entities are mentioned as options, and drift over time (monthly or quarterly).