Using AI in Adult Social Care: What the DHSC Guidance Says for Providers

Many adult social care providers are still working out whether and how to use artificial intelligence. At the end of September 2026 the Department of Health and Social Care (DHSC) published GOV.UK guidance called "Using AI in adult social care", written for providers. This article summarises what it says: the kinds of AI already in use, the cautions it gives, and the best practice areas it sets out. It is aimed at registered managers, nominated individuals, and staff who may be asked to use or review AI tools. For the regulatory view, see our guide to the CQC's principles on AI for care providers.

A note on the source: the GOV.UK page was first published on 29 September 2026 and last updated on 2 October 2026. It describes itself as an introduction to AI in adult social care and best practice for providers. It does not state whether it is statutory guidance, and the page we reviewed does not mention the CQC or set out specific legal duties beyond general references to UK GDPR. Check the current version before relying on any detail.

Where AI is already used in care

The guidance says many providers are in the early stages of exploring AI and some have not yet adopted AI tools. It lists four types already in use:

• Sensor-based technologies, including acoustic monitoring, which the guidance says can help with hands-on care in accommodation-based settings and with preventative and reactive care at home.
• Chatbots, which can answer questions and may serve as a first step in helplines, for example in mental health and reablement, and can run around the clock.
• Facial analysis, which can support pain assessment when a person cannot or will not say they are in pain. The caregiver records pain-related behaviours using the tool's framework, the tool produces an overall pain score, and the caregiver decides how to respond.
• Data collection and analytics, such as looking at incident reports for patterns. The guidance gives an example of falls clustering at the same place and time, traced to sunlight, which led to blinds being installed.

Care plans and assessments: a caution

The guidance says generative AI is being used to draft care plans and assessments, and to help with auditing, daily monitoring and data logging. It suggests this could reduce administrative time and free staff for direct care. It then adds two cautions. Evidence that AI can produce truly personalised care plans is, in its words, currently limited. And a member of staff must review AI-produced plans for accuracy, and the use must not breach data protection law.

Admin, meetings, HR and audit

The guidance also covers support functions. Tools such as Microsoft Copilot and Claude can record meetings and produce notes, action lists and next steps, but a human must review anything the AI produces. AI can turn meeting content into draft plans or strategies, and the guidance advises against entering confidential information into open tools and suggests hiding the organisation's name. For HR, it says AI can support recruitment and onboarding through automated responses, vacancy administration and scheduling. Because applicants use AI to write CVs, providers should decide whether they accept AI-generated applications and, if not, say so in job adverts. The guidance also says some providers use AI across several sites to summarise completed audit checklists and flag outstanding actions. Our guide to using AI tools with patient information covers the NHS England information governance view.

The best practice areas

The guidance sets out ten areas of best practice:

1. High-quality data. Data should be accurate, complete, well structured and organised, because the quality of the output depends on the quality of the input.
2. Data protection. Providers should check where data is processed, in particular whether it is inside the European Economic Area, which is usually found in the privacy policy, and tell people how their data is used. The guidance advises against putting personally identifiable information into free or online tools, because without a contract the provider cannot guarantee how the data is handled. It recommends asking the software supplier if unsure and points to the Digital Care Hub's Better Security, Better Care programme.
3. Ethics. It lists seven principles from the Oxford project on responsible generative AI in social care: truth, transparency, equity, trust, accessibility, humanity and responsiveness.
4. Bias. AI can reflect bias from training data, algorithm design or human data labelling, which can reinforce stereotypes, cause discrimination and distort decisions.
5. Workforce upskilling. Staff need training and confidence, and the guidance suggests appointing digital champions.
6. Human review. Always review AI outputs for accuracy, fairness and alignment with organisational standards. Providers should define the human role and write an AI policy covering responsibility, oversight, accountability, adoption approach and any assessment framework.
7. Choosing tools. Write a specification of what you want the tool to achieve, get feedback from people supported and from staff, compare benefits and risks in a table, and define measurable success outcomes.
8. Financial cost. Compare benefits and risks against the total cost, including hardware, software, training and system changes, and watch for hidden costs.
9. An AI policy. Every organisation should consider having one, even if it bans work use of AI. It should set out how roles interact with AI, which will differ for leaders and frontline staff, and include a review checklist. Staff should be consulted.
10. Continuous improvement. Collect feedback from human reviewers and, where possible, from the AI system itself, and use it to refine review processes as AI and organisational needs change.

What the guidance does not cover

The page we reviewed does not address consent to AI use, for example for monitoring or facial analysis. It does not discuss how alerts or failures from monitoring tools should be escalated, or the accuracy or bias of the pain assessment tool. Most of its examples come from case studies and provider experience rather than independent evaluation. These are gaps in the guidance, so providers will need to consider them in their own policies. Our guide to AI governance in care homes covers governance questions in more depth.

Frequently asked questions

Does a provider need an AI policy?

The guidance says every organisation should consider having one, even if it bans staff from using AI for work.

Can staff put resident information into a free AI tool?

The guidance advises against using personally identifiable information in free or online tools, because without a contract the provider cannot guarantee how the data is handled.

Does a person have to review AI care plans?

Yes. The guidance says a member of staff must review AI-produced plans for accuracy, and that human review of AI outputs should always be included.

To help your team build confidence with these tools, explore our CPD courses. This article is a summary of DHSC guidance and is not legal advice.