Skip to main content
Social care · UK

AI for social workers: faster case notes, without losing accuracy

AI transcription tools record a home visit or meeting and draft the case notes for the social worker to check. Councils across the UK now use them, and most social workers report real time savings. The research also shows errors reaching statutory records. This guide covers how the tools work, what the evidence says, and the safeguards social workers and councils need before relying on them.

Try Klang for freeFree plan · hosted in the EU

01

What AI transcription tools do in social work

An AI transcription tool records a conversation, such as a home visit, an assessment, a supervision or a multi-agency meeting, turns the speech into text and drafts a summary or case note. The social worker then edits the draft and adds it to the case management system. The tool saves typing; it does not make professional judgements.

Several tools are in use in UK councils. Beam's Magic Notes, now renamed Beam Notes, is designed for social care. Beam says it turns recordings and documents into structured case notes and reports, and stores data on UK servers. Ada Lovelace Institute research found Microsoft Copilot in use too. MHCLG is building Local Transcribe for councils on top of the i.AI tool Minute; it was piloted with 22 councils, starts with housing and names adult and children's social care as priority areas.

02

What the evidence says

The most detailed UK study is the Ada Lovelace Institute's “Scribe and prejudice? Exploring the use of AI transcription tools in social care”, published on 11 February 2026. It is based on 39 interviews, conducted from March to October 2025, with social workers, managers and experts across 17 local authorities in England and Scotland. Its main findings:

  • Almost all social workers reported meaningful benefits, mostly time saved, although the savings were uneven.
  • Errors reached records. One social worker described a summary that indicated suicidal ideation, although the person had never spoken about it.
  • Social workers assume full responsibility for checking what the tools produce.
  • Councils mostly evaluated the tools on efficiency, not on their effect on the people drawing on care.
  • There was no consensus on when it is appropriate to use these tools in social care.

03

Your professional duties do not change

Social Work England's professional standards apply to AI-drafted notes exactly as to typed ones. Standard 3.11 requires social workers to maintain clear, accurate, legible and up-to-date records, documenting how they arrive at their decisions, and standard 5.3 forbids falsifying records. A case note that contains words nobody said is an inaccurate record, whoever produced the first draft.

BASW's initial guidance on generative AI (March 2025) makes the same point: accountability for AI output sits with the social worker, output must be checked for accuracy, data must be handled under UK GDPR, and not everyone who draws on care will want AI used with their information. Its February 2026 article on the Ada Lovelace findings warned that social workers could face professional or legal repercussions for AI transcription errors.

04

The legal basis for recording

Social care records contain health data, which is special category data under Article 9 of the UK GDPR. Article 9(2)(h) allows processing necessary for the provision of social care on the basis of domestic law, and the Data Protection Act 2018, Schedule 1, paragraph 2 sets the condition for health or social care purposes. Councils normally rely on these, rather than consent, for the record itself.

Recording the conversation is a separate question from keeping the record. People should be told clearly that the visit will be recorded and transcribed by AI, why, how long the audio is kept, and that they can say no. The Care Act 2014 requires a written record of a needs assessment to be given to the adult (section 12(3)), so the person may later read the AI-assisted record. That is a good reason to get it right.

05

Safeguards before you rely on AI case notes

Drawing on the Ada Lovelace recommendations and BASW's guidance, councils and teams can put these safeguards in place:

  • Tell people before recording, offer a non-recorded visit, and record their choice.
  • Check every draft against the recording before it enters the case record, giving particular attention to risk, safeguarding concerns, names and quotes.
  • Use a tool that links each point in the summary to the moment in the recording, so checking is quick and complete.
  • Delete the audio once the note is approved, under a written retention rule.
  • Agree where AI notes are not used, for example in some safeguarding or court work, and review it as evidence grows.
  • Evaluate accuracy and the experience of people drawing on care, not only minutes saved.
  • Record use of the tool under the Algorithmic Transparency Recording Standard, as Ada Lovelace recommends.
  • Test accuracy with the accents and dialects in your area before rolling out.

06

Questions to ask when choosing a tool

Every tool in this market promises time savings. These questions separate them on the things that matter for statutory records:

  • Where is the data hosted and processed, and which sub-processors are involved?
  • Is any customer data used to train AI models, with or without consent?
  • Can the worker see the source of every sentence in the summary?
  • Can the audio be deleted while keeping the approved text, and is deletion complete?
  • Is there a data processing agreement, an information security certification and an audit trail?
  • How does it handle accents, interpreters and conversations that switch language?

07

How Klang approaches social care

Klang's public-sector page describes how Klang took part, with Kalmar municipality, in the Swedish data protection authority IMY's regulatory sandbox on transcription in social services. IMY's report does not name the vendor. It assessed the method, in which a human reviews every summary and there is no automated decision-making, and concluded that there is a legal basis, even for sensitive personal data.

Record a visit with the Klang app on a phone, or a meeting on Teams, Zoom or Google Meet. Klang transcribes, separates the speakers and drafts the note in your format. Every point links back to the moment in the recording. You can delete the audio and keep the text, and restrict each case folder to the people who need it. Klang is hosted in the EU, is ISO/IEC 27001:2022 certified, never trains AI on customer data and runs AI providers under zero data retention. Councils can also run Klang in a private cloud or on-premise.

AI transcription tools used in UK councils

What each provider or publisher states, read on 3 November 2026. Ask every provider the questions above; this table is not a ranking.

AI transcription tools used in UK councils
ToolWho provides itWhat they state
Beam Notes (formerly Magic Notes)Beam, UKBuilt for social care; data stored on UK servers; never used to train models without consent
Microsoft CopilotMicrosoftGeneral assistant in Microsoft 365; used by some councils for transcription
Local TranscribeMHCLG, built on i.AI's MinutePiloted with 22 councils; housing first, social care a priority
KlangKlang, SwedenHosted in the EU; took part in IMY's sandbox with Kalmar social services; never trains on customer data

Copilot usage in councils is reported by the Ada Lovelace Institute study. Provider statements are not independently tested here.

For social care teams

Be with the person. Check the note against the recording.

Klang drafts the case note from the visit, and every line links to the moment it came from, so checking is part of the workflow rather than an extra job.

  • Record visits with the phone app, or meetings on Teams, Zoom and Google Meet
  • Notes in your council's format, each point linked to the recording
  • Delete audio after approval; case folders restricted by role
  • Processed in the EU · ISO/IEC 27001:2022 · never used to train AI

AI in social work questions

Are social workers allowed to use AI to write case notes?

There is no general ban. Social workers remain responsible for accurate records under Social Work England's standards, and BASW's guidance says accountability for AI output sits with them. Use only tools your employer has approved, and check every draft before it enters the record.

What is Magic Notes?

Magic Notes is an AI transcription tool for social care made by Beam, now renamed Beam Notes. Beam says it turns recordings and documents into structured case notes and reports. The Ada Lovelace Institute reports that around 100 local authorities use it, citing Beam.

Do I need consent to record a home visit?

The record itself usually rests on social care duties rather than consent. Recording the conversation still needs a clear explanation, and BASW notes that not everyone will want AI used with their information. Offer a non-recorded visit and record the person's choice.

How accurate are AI transcription tools in social work?

Accurate enough to save time for most users, according to the Ada Lovelace Institute's 2026 study, but not reliable enough to skip checking. The study found errors, including an invented reference to suicidal ideation, and weaker performance with some accents.

Can AI notes be used in court or safeguarding work?

Practice varies and the Ada Lovelace study found no consensus. Ada Lovelace recommends guidance for AI transcription in statutory processes and formal proceedings. Until your council has that guidance, follow its policy and treat AI drafts with extra care.

Sources

Read on 3 November 2026. This page is general information, not legal or professional advice.