32% of US adults now use an AI chatbot for health information — up from 16% a year earlier, according to Rock Health’s 2025 Consumer Adoption Survey. 64% use it weekly or more often. The most common use cases are not wellness — they are diagnosis from symptoms (56%), treatment exploration (59%), drug research (55%), and lab interpretation (41%). These are clinical decision activities that, until very recently, required a clinician.

If you are building a digital health product, this is the data point that should reorganise your thinking: your real competitor is no longer the incumbent care pathway, the legacy condition-management app, or the other DHT in your category. It is the consumer’s free default. That competitor has zero customer acquisition cost, no evidence requirements, no DTAC submission, and a UX your eight-figure raise cannot match. Worse, your evaluators use it personally. The evidence architecture you are building against — NICE ESF, DTAC, MHRA SaMD — was designed for a world where patients enter healthcare through a regulated front door. They no longer do.

The new comparator, in numbers

Each of the top use cases has an equivalent in the UK system — and almost all of them used to be the entry point for either an NHS pathway or a consumer DHT.

The use-case profile matters because it tells you where the substitution risk is concentrated. AI chatbots are strongest exactly where most consumer-facing DHTs have historically positioned: information retrieval, symptom triage, condition education, and self-management coaching. They are weakest — for now — where defensibility actually lives: clinically integrated workflows, longitudinal data capture, regulated diagnostic outputs, and anything that needs to write back into an NHS system. If your product’s wedge is better information, you are now competing with a zero-cost incumbent that 32% of your potential users are already habituated to.

Why the existing evidence architecture struggles with this

The NICE Evidence Standards Framework — 21 standards across five groups, classifying DHTs into tiers A, B and C — was designed to give NHS evaluators a consistent way to assess whether a DHT is likely to deliver benefits. It assumes a clear comparator: standard care, the current pathway, or the existing tool. The economic standards (15 and 16) ask companies to demonstrate budget impact and value relative to that comparator.

That assumption is now strained. If 56% of consumers reach for a chatbot to interpret symptoms before they touch the NHS, the relevant comparator for a symptom-triage DHT is no longer NHS 111 — it is ChatGPT. That comparison is awkward in three ways. First, the chatbot is unregulated and outside the ESF’s scope, so there is no agreed evidence base to benchmark against. Second, the chatbot’s apparent performance — fluent, fast, free — sets a UX expectation your DHT will be measured against, even informally. Third, your evaluators are themselves chatbot users; their reference point for what good looks like has shifted, even if their procurement framework hasn’t.

The framework is aware of the gap. A 2025 Imperial College / JMIR analysis argued explicitly that the ESF — grounded in static, predefined evaluation methodologies — does not sufficiently accommodate AI tools that evolve through machine learning and real-world data. NICE has updated the framework to include adaptive algorithms, but the framework still assumes the technology is being commissioned. The category that is reshaping consumer health behaviour fastest is the one nobody is commissioning.

The cohort that breaks your go-to-market playbook

Conventional digital health diffusion theory has worried-well early adopters leading and high-need patients lagging by years. The Rock Health data shows the opposite is happening with consumer AI. Among users with four or more chronic conditions, AI chatbot engagement is markedly more intensive — 52% use it to manage ongoing conditions (vs 25% of users with no conditions), 54% to interpret labs (vs 38%), 40% to find providers (vs 28%), and 37% for mental health support (vs 23%). This is the patient your DHT most wants to reach, and they are already AI-native.

This inverts a standard DHT GTM assumption. The polymorbid patient — historically the slowest to adopt new digital tools, the costliest to onboard, and the hardest to retain — is now arriving with a baseline expectation of conversational, on-demand, multi-purpose health information. Your product is not their first digital health interaction; it is their tenth that week. Pricing models, onboarding flows, and evidence narratives built around introducing digital tools to vulnerable populations need a serious rewrite.

What this means for how you build, evidence, and position

Three implications worth taking seriously, in roughly the order they hit your roadmap.

1. Your comparator is now the chatbot, not the pathway

If your product overlaps with any top-five chatbot use case — symptom triage, treatment information, drug lookup, lab interpretation, condition education — your ESF Standard 5 (“effectiveness”) and Standard 15 (“value”) narratives need to address the AI comparator explicitly. “Better than current care” is no longer the relevant claim. “Better than what 32% of your patients are already doing” is. That is a higher bar, but it is also a more credible narrative to a commissioner who, privately, is using the same tools.

2. Wedge selection just got harder, and more important

The defensible wedges are the ones AI chatbots structurally cannot occupy: integration with NHS clinical systems (DTAC compliance, GP Connect, FHIR), regulated diagnostic outputs (Class IIa+ where SaMD applies), longitudinal data capture from connected devices, and anything that requires write-access to a patient record. Use cases that depend on information retrieval alone — even with proprietary content — are increasingly commodity. Building a pure information product in 2026 is building against a permanent zero-cost competitor.

3. Position to commissioners assuming they are already AI users

Healthcare professionals are using consumer AI, often informally, often outside their organisation’s policies. Your evaluators have personal context for what good feels like — and a private suspicion that consumer tools are eating territory their formal procurement processes are still pretending belongs to regulated DHTs. Your positioning can either ignore this (and feel out of touch) or address it directly. The DHTs that will win in 2026–27 commissioning conversations are the ones that can articulate, in plain terms, why they are clinically defensible in ways the chatbot is not — and why that distinction is worth paying for.

The bar just moved

Healthcare evidence frameworks were built on the assumption that the commissioned product is the patient’s primary digital touchpoint for the use case it addresses. That assumption is breaking. For DHT founders, the strategic implication is not how do we beat ChatGPT — it is how do we build something a chatbot structurally cannot replicate, and how do we evidence that delta. The 32% number will keep rising. Your evidence strategy, your wedge, and your positioning need to assume it.

If you are working through how this affects your own evidence narrative or market access plan, get in touch— this is the kind of strategic problem Healthonomix exists to help with.