TD Knowledge & Practice EcosystemTD Consultancy ↗

Evidence, limitations and practice readiness

Evidence Observatory

A cautious view of guidance and research relevant to AI in drug and alcohol services. Inclusion is not endorsement, and evidence from wider digital health cannot automatically be transferred to generative AI.

How to read the ratings

Green means established enough to consider within appropriate governance — not proven suitable for every service. Orange is emerging. Red remains experimental. Always read the original source.

AI governance · Guidance

UK Government AI Playbook

🟢 Established enough to consider

What was studied?: Practical principles and lifecycle guidance for safe, effective and secure AI use in government.
What did they find?: The playbook emphasises understanding limitations, lawful and secure use, meaningful human control, lifecycle management and choosing the right tool.
Limitations: It is government guidance, not drug and alcohol treatment guidance or a legal approval route.
What does this mean for UK practice?: Public-sector teams can use the principles as a governance baseline, then apply their own information, procurement and professional controls.

Read the original source: Government Digital Service, 2025 ↗

Confidentiality · Guidance

ICO guidance on AI and data protection

🟢 Established enough to consider

What was studied?: How UK data-protection principles apply across the AI lifecycle.
What did they find?: The guidance covers accountability, transparency, lawfulness, fairness, security, data minimisation and individual rights.
Limitations: The ICO says this guidance is under review following the Data (Use and Access) Act. It cannot determine whether a local use is lawful without the facts, contracts and data flow.
What does this mean for UK practice?: Organisations need a documented purpose, minimised information, defined accountability and suitable assurance before personal data is processed.

Read the original source: Information Commissioner’s Office ↗

Health information governance · Guidance

NHS England AI information-governance guidance

🟢 Established enough to consider

What was studied?: Information-governance implications of AI use in health and care settings.
What did they find?: NHS England provides separate guidance for patients and service users, health and care professionals, and information-governance professionals.
Limitations: The guidance supports lawful and safe use but does not approve a supplier, account or local workflow.
What does this mean for UK practice?: NHS-linked and health-care services should involve information-governance expertise early and explain uses clearly to affected people.

Read the original source: NHS England, updated May 2026 ↗

Procurement · Guidance

NHS Digital Technology Assessment Criteria

🟢 Established enough to consider

What was studied?: Baseline assurance criteria for digital health technologies used by commissioners and providers.
What did they find?: DTAC brings together clinical safety, data protection, technical security, interoperability and usability and accessibility considerations.
Limitations: Applicability depends on the technology and setting; completion is not proof of effectiveness.
What does this mean for UK practice?: Where relevant, procurement should connect product assurance with evidence, workflow testing, equality and ongoing monitoring.

Read the original source: NHS England DTAC ↗

Evaluation · Guidance

NICE evidence standards for digital health technologies

🟢 Established enough to consider

What was studied?: What levels and types of evidence may be needed for digital health technologies, including AI and adaptive algorithms.
What did they find?: The framework links evidence expectations to a technology’s function and potential risk.
Limitations: It is a framework for evidence planning and evaluation, not automatic approval or a substitute for regulatory requirements.
What does this mean for UK practice?: Commissioners should match evidence demands to claims, risk and intended use rather than accept a generic demonstration.

Read the original source: NICE Evidence Standards Framework ↗

Real-world evaluation · Evaluation learning

Lessons from the AI in Health and Care Award

🟢 Established enough to consider

What was studied?: Practical lessons from planning and implementing real-world evaluations of AI technologies in NHS settings.
What did they find?: NHS England highlights the value of early evaluation planning, suitable comparators, stakeholder expectations and patient and public involvement.
Limitations: Lessons come from a wider health programme and are not evidence that a particular product works in drug and alcohol services.
What does this mean for UK practice?: Define baseline, workflow, outcomes, inequalities and stakeholder involvement before a pilot, not after deployment.

Read the original source: NHS England, 2024 ↗

Transparency · Guidance

Algorithmic Transparency Recording Standard

🟢 Established enough to consider

What was studied?: A standardised way for public-sector bodies to explain how and why algorithmic tools are used.
What did they find?: The standard supports intelligible disclosure, named accountability and public understanding for relevant uses.
Limitations: Mandatory scope is specific; broader public-sector organisations need to check their position and local obligations.
What does this mean for UK practice?: Services should plan transparency early, especially where an algorithm influences decisions with public effect.

Read the original source: GOV.UK ATRS guidance, 2025 ↗

Generative AI and health · Guidance

WHO guidance on large multi-modal models

🟢 Established enough to consider

What was studied?: Ethics and governance of generative systems that work across text, images and other inputs in health.
What did they find?: The guidance describes risks including false or biased output, automation bias, privacy, cyber-security and unequal access, alongside possible uses.
Limitations: Global guidance needs translation into UK law, regulation, policy and local professional practice.
What does this mean for UK practice?: Health-related uses need transparency, expert oversight, inclusive design and evidence of benefit before routine adoption.

Read the original source: World Health Organization, 2025 ↗

AI and addiction · Review

AI in addiction: challenges and opportunities

🔴 Experimental

What was studied?: Published applications of AI across risk identification, prediction, treatment and addiction research.
What did they find?: The review describes promising applications but also major challenges around data quality, generalisability, privacy, bias and real-world validation.
Limitations: A broad review cannot establish clinical effectiveness for individual tools or UK services.
What does this mean for UK practice?: Treat prediction and personalised-care claims as research questions until independently validated in the intended setting.

Read the original source: Suva et al., 2024 ↗

AI chatbots · Systematic review

Chatbot-assisted substance-use interventions

🔴 Experimental

What was studied?: Research using chatbot technologies for prevention, assessment or treatment of alcohol, nicotine and other drug use.
What did they find?: The systematic review found a developing field with varied chatbot designs, outcomes and study quality.
Limitations: Heterogeneity, short follow-up and rapidly changing technology limit confident conclusions about effectiveness and safety.
What does this mean for UK practice?: A chatbot should not be treated as an established substitute for professional treatment, crisis response or safeguarding.

Read the original source: Lee et al., 2024 ↗

Digital therapeutics · Systematic review

Digital help for substance use

🟠 Emerging

What was studied?: Digital interventions intended to reduce substance use or related harm.
What did they find?: The review reported benefits across some digital interventions, while effects and designs varied.
Limitations: Digital interventions are not all AI, and study populations, substances, comparators and outcome measures differ.
What does this mean for UK practice?: Do not transfer evidence from a tested digital intervention to a new generative-AI tool without evaluation.

Read the original source: Bonfiglio et al., 2022 ↗

Adolescent prevention · Systematic review

AI-based adolescent substance-use prevention

🔴 Experimental

What was studied?: Emerging AI approaches intended to support prevention among adolescents.
What did they find?: The 2026 systematic review maps a developing evidence base rather than establishing a mature standard of care.
Limitations: New studies, varied methods and limited external validation require caution, particularly for consequential prediction.
What does this mean for UK practice?: Youth services should require safeguarding, equality, transparency and independent evidence before using predictive systems.

Read the original source: Atinga et al., 2026 ↗

Evidence reviewed 13 August 2026. Corrections and newer authoritative evidence are welcomed through TD Consultancy.