AI in Social Care: What the Evidence Shows
What UK council trials and the Ada Lovelace Institute found on AI note-taking in social care: time saved, errors to watch for, and what it means for staff.
16 November 2026 6 min read
Artificial intelligence is already working in UK social care, mostly on one unglamorous job: writing things up. Councils such as Swindon and Kingston have tested AI note-taking tools that listen to a conversation and draft the record. The early numbers are encouraging, the independent evidence is thinner, and the risks are real. This article separates what councils measured, what researchers found and what companies claim, so you can judge how much weight each deserves.
Where AI is used in social care today
The most common tools are AI transcription tools, sometimes called AI scribes. A social worker records a visit or assessment with consent, and the software produces a draft record to check and edit. Community Care reported that Beam's Magic Notes is used by 85 councils, and the Guardian's report on a 2026 study found that most social workers interviewed used either Magic Notes or Microsoft Copilot.
A second group of tools helps care workers spot problems early. In a PA Media report from March 2025, carers on home visits log observations on a phone app, and software from home-care provider Cera assesses fall risk and flags signs of illness. PA reported its use across 29 of the UK's 42 integrated care systems.
What councils measured with AI note-taking
Swindon Borough Council trialled Magic Notes from April to July 2024, across 184 frontline meetings with 19 social workers and one leadership support officer. The council reported a 63% reduction in the time spent compiling assessments and logging case notes. The average Care Act assessment conversation fell from 90 minutes to 35, and the write-up from four hours to 90 minutes.
Kingston Council's case study for the Local Government Association describes a pilot that grew from three social workers to more than ten. It reports a 50 to 60% average reduction in time spent on case notes and assessments, and supervision write-ups that fell from 40 minutes to under 10. The council set an accuracy target of 85% and reports results of 96% or more. It also lists limits: the tool relies on clear audio, so it is less accurate for people with limited speech or mental capacity.
Researchers at the Alan Turing Institute evaluated the tool across Swindon, Barnet and Camden between April and September 2024, covering 91 staff. Staff rated it 4.26 out of 5 on average, and practitioners reported misspelt names, repetition and assumptions in summaries that needed editing. Community Care notes that Beam, which makes the tool, commissioned the evaluation. The Guardian reported the cost to councils at between £1.50 and £5 for each hour of transcription.
All of these figures come from the councils or the vendor. They point the same way, but none is an independent trial with a comparison group.
What independent research found
The Ada Lovelace Institute published the most detailed independent study in February 2026. It interviewed social workers and senior staff across 17 local authorities in England and Scotland between March and October 2025. The report found "readily observable time savings", and says the tools can improve the quality of recorded information and the relational side of the work. Almost all the social workers it interviewed reported meaningful benefits, though not equally.
The report is also frank about the gaps. It is an interview study, not a trial, and it says there is very little evidence on how AI transcription tools affect the experiences and outcomes of people who draw on care.
The risks to take seriously
An error in a care record can follow a person for years. The Ada Lovelace Institute warns of harmful hallucinations and misrepresentations entering statutory records. The Guardian's account of the study described a transcript that wrongly flagged suicidal ideation and a child's account turned into unrelated words. Social workers said that checking an AI draft took anywhere from two minutes to an hour, and one had received only an hour of training. Beam told the Guardian that its outputs are first drafts and that it runs automated hallucination-risk checks.
Accountability stays with the practitioner. Social Work Today's coverage says responsibility for accuracy often fell on frontline staff. The British Association of Social Workers has urged caution over data protection and asked whether saved time will simply become bigger caseloads, according to Community Care. The Institute recommends that councils record their use of AI transcription under the Algorithmic Transparency Recording Standard, and that government fund wider pilots and a What Works Centre for AI in Public Services.
Early-warning tools: promising claims, thinner proof
Cera told PA that pilots of its technology predicted falls with 97% accuracy, and that ongoing tests show a 20% reduction in falls. The pilots covered 800 patients in July 2023 and 2,254 in September 2023. The article gives no baseline or comparison group, and the figures are the company's own.
A council-led evaluation shows why that matters. Cornwall Council's NIHR-funded research team evaluated sensor-based falls prevention in care homes. Falls were already falling before the pilot, and the team concluded that any direct effect of the technology was at best modest and that it could not confidently attribute the reductions to it. The page also says the systems are not a replacement for staff.
What it means if you work in care
For people working in or joining social care, the evidence points to AI changing the paperwork more than the caring. The tools reviewed here take over writing and record-keeping, and the people using them are still responsible for checking the result. Wakefield Council lists basic digital skills among the qualities it looks for in adult social care staff, alongside sensitivity and patience.
If you are new to care, our guide for people new to care explains how to start, and you can browse current vacancies. Providers weighing up AI tools can find recruitment support on our employers page.
Frequently asked questions
How is AI used in social care? Mostly to transcribe and summarise conversations so staff spend less time on records, and increasingly to flag fall and illness risk from visit logs.
Can AI reduce paperwork for social workers? Councils report large reductions: 63% in assessment and case-note time in Swindon's trial and 50 to 60% in Kingston's pilot. Independent interviews across 17 councils also found time savings, though benefits were uneven.
Is AI safe to use in social care records? Only with careful checking. Researchers have documented false statements, accent errors and invented content in AI drafts, and the practitioner stays accountable for the record.
Will AI replace care workers? Nothing reviewed here suggests it will. The tools target paperwork and early warning, and the Cornwall evaluation page says its sensor systems are not a replacement for staff.
What does AI mean for people starting a career in social care? The hands-on work stays human, and basic digital skills are becoming part of the job. Start with our new to care guide.
Written by Kay Ismail, founder of workinsocialcare.com, who spent almost a decade working in UK social care. This article was drafted with AI assistance and every figure was checked against the sources linked above.