Skip to main content
Guidance

Using AI in social care: a quick guide for practitioners

Published 20 November 2026

Applies to England

This practitioner guide sits alongside the joint DfE and DHSC AI in social care guidance.

This guidance:

  • should be applied in line with local policies, governance arrangements, professional standards and existing legal and regulatory requirements
  • does not replace local or national guidance, legislative requirements, statutory duties or professional standards

Six things to remember

  1. AI can enhance your practice. It can help you organise, summarise, draft and retrieve information, freeing up time for relationship-based practice and professional reflection. Your professional judgement and critical thinking remain central to good practice.
  2. You can turn AI outputs into high-quality work. AI can provide a useful starting point, helping you develop drafts more quickly. It can also be wrong, even when it sounds confident. Check, challenge and amend outputs to correct anything inaccurate, incomplete, misleading or overly confident, and make the final version your own. Effective prompting will help strengthen your initial draft.
  3. Get more value by testing and refining. AI can help generate ideas, structure information and offer alternative wording, but it does not think or understand context, nuance and inference as a practitioner does. Questioning and testing its output can help you produce something more useful and better grounded in the circumstances – AI won’t get offended if you challenge or test its outputs.
  4. Use approved, secure tools with confidence. Organisation-approved AI tools can support effective practice while helping to protect people’s information. Use them in line with local and national policies and guidance. Keep personal, confidential or case-specific information out of AI tools that are not approved and protected by your organisation. If you are in doubt, check.
  5. Keep professional judgement at the centre. AI can help you work more efficiently and strengthen recording, reflection and analysis. You remain responsible for your casework, records and decisions, and should be able to explain and justify the final actions you take.
  6. Build your skill set. Prompting is one of the most important skills required for interacting with many AI tools. Be clear what you want the tool to do and not do. Factor AI into your professional development.

AI should support practitioners and relationship-based, person-centred care – not replace professional judgement: Think first, then draft.

What AI can help with

AI should support practice, not replace it. You do not need to be an AI expert to use it.

AI may help you to:

  • save time on routine tasks by producing first drafts of notes, summaries, letters and records for you to review, test, amend and add to with your professional reflections and observations
  • record conversations more efficiently by transcribing or summarising meetings, visits and supervisions
  • find key information quickly by searching, organising and summarising long records, guidance or other documents
  • prepare for visits, meetings and assessments by bringing together relevant information and suggesting questions or topics to explore
  • strengthen your analysis by highlighting themes, changes, connections and gaps in information, helping you explore different perspectives and lines of enquiry
  • help identify emerging needs, trends and recurring issues that may support earlier intervention, service improvements and better outcomes
  • reflect on your practice by helping you test your thinking, consider other perspectives and prepare for supervision or case discussion
  • communicate more clearly by simplifying complex information and adapting the tone, length or format for different people
  • make information more accessible by translating text, explaining unfamiliar terms or presenting information in different formats
  • better reflect people’s voice by ensuring the voice of the individual and family is reflected in a person’s records, including strengths, wishes, experiences and desired outcomes
  • support joined-up working by drafting handovers, summarising agreed actions and making information easier for colleagues to understand
  • support organisational learning by helping teams identify themes, share learning and make better use of information gathered through practice

Prompting

Most AI products that you are likely to use to support your practice rely on prompts to generate and shape output. Clearer instructions result in better outputs to use as a first draft.

The GCSE framework can help improve the quality of your prompts, and the quality of AI’s outputs. It stands for:

  • Goal – clearly define what you want the AI to accomplish
  • Context – provide background information so the AI understands the purpose and audience
  • Source – specify where the AI should retrieve information from
  • Expectation – describe the desired format, tone, style or length of the output

Prompting example 1

Goal

Create a chronology of significant events from this child’s case records.

Context

I am a children’s social worker preparing for an initial child protection conference. The chronology will be shared with the family, conference chair and professionals in attendance to help inform discussions about support the child may need.

Source

Use only the case notes, alerts and historic assessments from the child’s file. Include the source and the date in the output, and flag any gaps or conflicting information.

Expectation

Present the document as a list of key events in chronological order, using factual and neutral language. Do not assume connections that are not present or make any judgements. Flag any queries, inconsistencies or gaps for me to review and explore further.

Prompting example 2

Goal

Draft an agenda for a multi-agency meeting about a child I’m working with.

Context

I’m a children’s social worker. Attendees will include the child’s parents, their school, the school nurse and their housing officer.

Source

Use only case notes from the last 6 weeks, the child’s most recent assessment and actions from the previous meeting.

Expectation

Give me a one-page agenda with timings for a 45-minute meeting, starting with the family’s views and ending with agreed actions and the next meeting date.

Prompting example 3

Goal

Quality assure my completed Care Act assessment before submission, focusing on evidence, consistency and Care Act reasoning rather than rewriting.

Context

I am an adult social worker seeking a final review of the assessment’s accuracy and defensibility.

Source

Use the assessment as the sole source of information about the person. Do not invent facts or fill gaps.

Expectation

Identify up to 5 material issues, prioritised by impact on accuracy or meaning.

For each issue, reference the relevant section and explain what needs checking. Distinguish between fact, opinion and uncertainty.

Prompting example 4

Goal

Help me use Prospera Tedam’s MANDELA model to support a social work student’s learning from a challenging home visit, linking reflection, anti-oppressive practice and professional development.

Context

I am a Practice Educator in adult social care working with a student who prefers visual learning.

Process

Use only information I provide and Tedam’s published MANDELA model (please identify the source used).

Expectation

After the final question, stop questioning and organise the discussion around the relevant MANDELA elements.

Clearly distinguish where relevant, connect the discussion to the Professional Capabilities Framework (PCF) and Key Skills and Behaviours, using verified references.

What AI should not do

AI offers a range of opportunities to support you in your practice, but there are some activities where its use is not appropriate and professional judgement must remain central.

AI should not be used to do the following.

Independently make decisions that directly affect individuals

AI should not make decisions about what interventions or support individuals, children or families supported by social care receive, particularly where a decision affects their rights, care, support or wellbeing.

Determine risk, need or safeguarding thresholds

AI cannot fully understand context, nuance or significance of the information you or your organisation hold. These judgements should remain with practitioners.

Complete assessments or care plans

AI may help you organise information or prepare a draft, but the analysis, conclusions and recommendations should reflect your own professional judgement.

Replace professional judgement

AI can help you test and develop your thinking, but it cannot understand the full context of a person’s circumstances. Your analysis, judgement and accountability should remain your own.

AI may support professional judgement, but it must not replace it

Responsibility and accountability remain with you and your organisation.

Not everything AI can do are things that AI should do. You may want to stop and think:

  • is this something I want to give up?
  • am I able to justify the decision I am responsible for?

Reviewing AI output is professional work

Checking AI-assisted work is not just proofreading. Practitioners should test whether content is accurate, evidenced, balanced and reflective of their own judgement.

You may need to add context, observations, uncertainty, professional curiosity, body language, tone, what was not said, and the significance of what was observed.

Case records, care planning and related documentation should reflect the wider engagement, not just a summary of a transcript. Where appropriate, you should consider allowing the child, family or person drawing on care to see the transcript or summary produced for them, so their voice and needs are accurately captured.

AI outputs should be considered a first draft AI may also be more effective when you develop your own critical reflections and judgements before using AI to generate a first draft, so you can interrogate the draft from an informed place. Think first, then draft.

Transparency and trust

Trust is fundamental to effective social care. Used appropriately, AI can help improve the quality, consistency and accessibility of information, supporting better engagement, communication and outcomes for people who draw on care and support.

Organisations should provide clear and accessible guidance or policies on transparency, the choices people have about the use of AI in practice, and how objections should be managed.

Practitioners should be able to explain what tool is being used, why it is being used, how outputs will be reviewed and used, how objections should be managed, and what organisational guidance applies. In multi-agency settings, AI use should be agreed before meetings where organisations may have different policies.

Clear local governance matters

Practitioners do not need to be technical experts, but you should know:

  • what the tool is for
  • what information can be entered
  • how the output will be checked and used
  • where accountability sits

Organisations should provide clear policies, approved-use boundaries, training, supervision, escalation routes and monitoring arrangements.

Quick final check Before using an AI-supported output, ask yourself:

  • have I checked it?
  • have I challenged it?
  • have I added my professional judgement?
  • would I be able to explain and justify the final document and decisions?

Overview of ethics and risks

This is a quick reference to the main risks and mitigations outlined in the non-statutory guidance. It should be read alongside local policy and governance arrangements.

Risk or challenge What to watch for Practical mitigation
Professional judgement and human oversight AI may encourage over-reliance on the AI tools, leading to weaker critical thinking or automation bias (where AI outputs are given undue trust or importance). Use AI to support, not replace, judgement. Reflect first, treat outputs as drafts, and add analysis, context and professional judgement.
Over-reliance and deskilling Practitioners may rely too heavily on AI and lose or fail to develop analytical skills. Use AI to strengthen practice, not replace professional curiosity, analysis or supervision.
Accountability and governance Outputs may be accepted without challenge or appropriate review, with no oversight, accountability or responsibility for decisions made. Practitioners remain accountable. Organisations should clarify supervision, audit and escalation arrangements.
Accuracy and hallucination AI may produce plausible but incorrect, incomplete or misleading content. Check against source information, test conclusions and do not copy AI text into records unchecked.
Bias, fairness and equity AI may reflect or amplify bias and create unfair or unequal impacts. Check for bias or unfairness. Organisations should test and monitor tools for unequal impact.
Transparency and explainability People may not understand when AI is used, what it does or how outputs are reviewed. Be clear when and why AI is used, what the tool does, how outputs are reviewed and how data is handled.
Trust Using AI without appropriate transparency can damage trust and raise data protection concerns. Follow local policy and explain the tool, purpose and output use. Be clear how objections should be managed.
Privacy, security and data protection Personal or sensitive information may be shared, stored or reused by third-party tools. Use approved tools only and do not enter personal or case-specific information into public AI tools.
Multi-disciplinary settings Partner organisations may have different AI policies. Agree AI use before multi-agency meetings and follow relevant organisational policies.
Legal risk Unchecked outputs could affect statutory decision-making or create different outcomes. Read, check and amend outputs. Do not use AI to make decisions. Be able to justify reasoning.
AI drift Tool performance may worsen as models, data, policy or practice contexts change. Continue checking outputs. Organisations should update and monitor tools and data inputs.

Checklist for using AI safely

What you need to consider before, during and after using AI to support practice.

If you are unsure about any of these checks, pause and seek advice. Do not use the AI output until any concerns have been resolved in line with local policy and AI governance arrangements.

Before using AI

Check whether the tool is approved for the task

Follow your organisation’s AI usage policy – for example can it be used to process personal data or analyse case records.

Use AI only for tasks that are appropriate

Use it in line with your organisation’s policies and procedures. For example, use AI to draft a briefing or identify themes in feedback, not to make decisions about support or risk.

Check what information can be entered

Do not enter personal, confidential or case-specific information into publicly available AI tools like ChatGPT, Claude, Gemini.

Be clear what you want AI to help with

This could be drafting, summarising, organising information, identifying gaps or supporting reflection. For example using AI to:

  • identify themes in consultation responses
  • prepare a draft briefing
  • summarise feedback from people using services

Develop your own professional view first

This should include your:

  • critical reflections
  • judgement
  • understanding of the person’s circumstances

Consider transparency requirements

Do this in line with local policy, data protection law and safeguarding responsibilities.

Be ready to explain what tool is being used and why

For example if you’re using a transcription tool to record and summarise a conversation, be clear that you will then review and edit this to reflect the discussion.

Prompting AI

Give clear instructions

Be clear about the task, context, intended outcome and format.

Set boundaries

These could be:

  • do not make decisions
  • do not assess risk
  • do not determine actions
  • base the response only on the information provided, or acknowledge uncertainty

For example, ask AI to identify patterns in service engagement, but not suggest actions, next steps or outcomes.

Ask AI to support thinking or drafting, not to make professional judgements

AI should not make judgements about risk, need, eligibility, safeguarding, care planning or intervention.

Avoid unclear prompts

Avoid prompts that invite unsupported conclusions, recommendations or invented information

Use a simple structure

This could be GCSE:

  • goal
  • context
  • source
  • expectation

Reviewing AI outputs

Develop your own critical reflections and judgements

Do this before using AI to generate a first draft, so you can interrogate the draft from an informed place.

Treat every AI output as a first draft

AI outputs should not be a final record, assessment, analysis or decision.

Check the output against the original source information

Do this before saving, sharing or using it. For example, check names, dates, chronologies and risk-related content.

Look for inaccuracies

This could be hallucinations or missing information, including confident but wrong content.

Check for bias

Check for bias, unfairness or inappropriate language.

Add what AI cannot capture

For example add:

  • context
  • observations
  • tone
  • body language
  • uncertainty
  • professional curiosity
  • what was not said
  • the person’s voice

Be clear about your own professional judgement

Make sure your own professional judgement is clear in the final version,

Critically review outputs and do not accept them at face value. Even if it appears accurate, ensure that context and nuance are included. Ask yourself whether you could justify your decision using the amended output.

Before adding anything to records or using it professionally

Review outputs

Do this against the original information and intended purpose. With transcription tools, make sure the final record reflects wider engagement and is not just a summary of a conversation or transcript.

For example, compare the AI summary with the source data and your own reflections to make sure it is accurate and reflects the case context.

Confirm the final version is accurate

Make sure that it is evidenced, balanced and appropriate for its purpose. For example, check whether findings are supported by the evidence and correct any inaccuracies.

Do not copy AI-generated text into case records unchecked by a practitioner

For example, make sure outputs meet organisational and professional standards before saving, sharing or using them. The final document should preserve the individual’s and professional’s voice.

Take extra care where outputs may be used in statutory, legal or high-consequence contexts.

Consider whether the person drawing on care should see the transcript or summary

This is so their voice and needs are accurately captured.

Stop and seek advice

Do this if:

  • a person, family, carer or partner organisation objects to AI being used, or you are unclear how objections should be managed
  • you are working in a multi-agency setting where organisations may have different AI policies
  • you feel uncomfortable with or are unclear about whether the use of AI is safe and responsible

Give feedback to managers or technology teams

Do this if the output:

  • suggests, makes or appears to influence a decision about risk, need, eligibility, safeguarding, care planning or intervention
  • appears inaccurate, biased, unsafe, inappropriate or unclear

You need to report concerns about unsafe, inaccurate, biased or inappropriate outputs through local governance arrangements. For example, raise concerns if the tool repeatedly produces inaccurate or misleading information.