Your AI Playbook · Care & people-work

AI tools that move the needle in care & people-work.

Practical, free-first, vetted. For teachers, nurses, social workers, counsellors, occupational therapists, ECD practitioners, community health workers — anyone whose work runs on presence, judgement, and the trust of the people in front of them.

Built for the South African care professional. No hype, no outcome promises. The goal: by this Friday, you’ve used at least two of these tools to give yourself back time for the part of the work that actually needs you.

~15 min read 7 tools 5 workflows 12 prompts
What's in here
  1. What you do that AI can’t replace
  2. 7 AI tools worth your attention
  3. 5 workflows to try this week
  4. Prompt library (copy & adapt)
  5. Ethics & pitfalls in SA care contexts
  6. Where to go deeper

01What you do that AI can’t replace

Start here. Care work is the clearest case in the economy where AI is a helper, never a replacement. The seat that matters is yours.

Teaching, nursing, counselling, social work, therapy, ECD — these roles are full of moments AI fundamentally cannot do well, even as it gets faster at the paperwork around them. Hold these as your edge:

The rest of this playbook is about everything else — the admin, drafting, note-writing and research that eats your day and pulls you away from the part only you can do.

027 AI tools worth your attention

Free tiers first. Each comes with a link to genuinely free training, not a sales funnel.

ChatGPT (OpenAI)
Free tier + paid R350/mo

The general-purpose language model most professionals start with. Drafting parent letters, simplifying jargon, rewriting reports, brainstorming activities, structuring case notes from rough text.

Why for care work: the highest-leverage drop-in for the everyday writing tasks — parent communications, referral letters, activity ideas, plain-language explanations. Free tier covers most weekly use.

Claude (Anthropic)
Free tier + paid R450/mo

Long-document analysis, careful drafting, structured thinking. Handles up to 200k tokens (about a 500-page document) in one conversation.

Why for care work: the best free tool for “read this 30-page IEP / care plan / policy document and explain it”. Particularly strong at noticing when it’s unsure — it pushes back instead of inventing, which matters when you’re working in clinical or safeguarding territory.

Otter.ai
Free 300 min/mo + paid R150/mo

Voice-to-text transcription with speaker identification, automated summaries and action-item extraction. Runs from your phone for in-person sessions, joins Zoom/Meet/Teams as a bot.

Why for care work: the single biggest time-saver if your day ends with case notes. Speak the note aloud after the session, get a structured first draft. You stay present in the session instead of half-writing during it. Critical: never transcribe a session containing identifying client/learner detail in a free tool. See ethics section.

Perplexity
Free + Pro R350/mo

An AI-powered search engine that gives answers with cited sources. Like ChatGPT but every claim links to where it came from.

Why for care work: for any “is this current best practice?” question — intervention evidence, medication interactions, policy/guideline lookups, referral pathways, programme research. Cited answers stand up to scrutiny in supervision and case conferences.

Grammarly
Free + Pro R140/mo

Inline writing assistant in your browser, email and Word. Catches grammar, tone and clarity issues as you write.

Why for care work: reports, progress notes, parent letters and referrals carry your professional reputation. Free tier catches the small errors that quietly erode credibility over time. Pair with ChatGPT/Claude for drafting, Grammarly for the final pass.

Notion AI
Free Notion + AI R200/mo

A workspace tool with AI built in. Summarises notes, drafts inside documents, builds tables from messy text, generates action items from session minutes.

Why for care work: useful if you carry a caseload and need a private case-load tracker, supervision-prep workspace, or resource library. Skip it if you don’t already use Notion — prioritise ChatGPT and Claude first.

Microsoft Copilot
Free in Edge + paid M365 add-on

Embedded across Word, Excel, Outlook, PowerPoint and Teams. Drafts emails in Outlook, summarises Teams meetings, helps build lesson plans or care-plan templates in Word.

Why for care work: if your school, clinic or NGO runs on Microsoft 365 (most do in SA), Copilot is where the leverage already lives, with your employer’s data-protection settings around it — safer than free public tools for anything containing client/learner detail.

035 workflows to try this week

Each one is a 5-to-15-minute investment that should pay back hours within a week. Don’t try them all at once. Pick the one that maps to a task you’re actively dreading.

Workflow 1 · Communication

Warm-but-clear parent / client message 15 min → 4 min

  1. Open ChatGPT or Claude. Describe the situation in your own words — what happened, the relationship, what you need them to do or understand. Use no names or identifying detail. Refer to “the learner” / “the parent” / “the client”.
  2. Ask: “Draft a message to a parent / caregiver / client. Tone: warm, professional, direct. SA context. No Americanisms. Acknowledge their position before what I’m asking.”
  3. Review for anything that overpromises (e.g. outcomes, guarantees) or anything that reads cold.
  4. Rewrite one paragraph in your own voice. That paragraph carries the relationship.
  5. Send.
Workflow 2 · Case & session notes

Voice memo → structured case note post-session

  1. After the session, open Otter or your phone’s voice recorder. Speak the note aloud as if telling a colleague — what happened, what you observed, what you’re thinking, next steps. De-identify as you speak (“the learner” not the name).
  2. Get the transcript. Paste into ChatGPT or Claude.
  3. Prompt: “Restructure this into a session note with these sections: presentation, observations, intervention, response, plan. Keep my clinical language. Don’t add details I didn’t mention.”
  4. Read the draft against your memory of the session. AI miss-attribution and small additions happen — verify.
  5. Paste into your case-management system. Add the names and identifying detail there, where it belongs.
Workflow 3 · Planning documents

First-draft IEP / care plan / treatment plan 3 hrs → 45 min

  1. Open Claude (better at long structured drafts than ChatGPT free tier).
  2. Paste in: your template or framework, the de-identified profile of needs, the goals you’ve already discussed with the team / family.
  3. Ask: “Draft a first version of this plan. Aim for clarity over comprehensiveness. Mark any section where you’re inferring beyond what I gave you with [VERIFY].”
  4. Take the draft into your own document. Edit, restructure, cut anything inferred, add detail only you know. Treat the AI draft as a junior’s submission.
  5. Discuss with the family / multidisciplinary team. AI is never the final author of a plan that affects a person’s life.
Workflow 4 · Resource curation

Referrals, programmes, evidence lookup 1 hr → 15 min

  1. Open Perplexity. Search for the specific need — e.g. “evidence-based interventions for selective mutism in primary school children” or “SA referral pathways for adolescent eating disorders”.
  2. Read the cited sources, not just the summary. Models still hallucinate citations sometimes — if a paper’s title looks too convenient, check it exists.
  3. For SA-specific referrals, search again with “South Africa” or your province in the query. Default results lean US/UK.
  4. Cross-check against your own networks. AI surfaces options; your colleagues confirm which actually take cases right now.
  5. Build the resource list in your own document, with your own annotations on quality and fit.
Workflow 5 · Supervision & reflection

Prep for supervision in 10 minutes before each session

  1. Open ChatGPT or Claude. Describe a case you want to bring — de-identified, focused on the dynamic you’re stuck on.
  2. Prompt: “Help me prepare for clinical supervision on this case. What are the 3 most useful questions I should bring? What might I be missing? What countertransference signals are worth exploring?”
  3. Read the response as a thinking partner, not an authority. Push back — “that’s not quite it, because…” — and watch your own thinking sharpen.
  4. Bring your own questions to supervision, not the model’s. The point of the exercise is to surface yours.
  5. After supervision, jot what you actually heard — not what the model predicted. The gap is where the learning is.

04Prompt library

Twelve copy-and-paste prompts. Tweak the [italicised parts] for your situation. Every prompt assumes you’ve de-identified the content first. Never paste a name, ID number, school name or other identifying detail into a free public AI tool.

Difficult parent / caregiver message
I need to write to a [parent / caregiver / client] about [difficult topic]. Rewrite my draft below to be warm, clear and direct. Acknowledge their position before what I'm asking. SA context. Under [180] words. Flag anything that sounds defensive or like I'm overpromising. [paste your draft]
Voice memo to case note
Below is a transcript of a voice note I recorded after a session. Restructure it into a clinical/case note with these sections: 1. Presentation 2. Observations 3. Intervention / what we did 4. Response 5. Plan / next steps Keep my clinical language. Don't infer details I didn't mention. If something is ambiguous, flag [UNCLEAR]. [paste transcript]
IEP / care-plan summary for caregiver
Rewrite the following [IEP / care plan / treatment summary] in plain language at a Grade 8 reading level for a [caregiver / parent] who is not a specialist. Preserve every concrete commitment. Use warm but not childish tone. SA English. Flag any clause where plain-language rewriting loses clinical precision so I keep the original version for the formal record. [paste the document]
Referral letter
Draft a referral letter from a [role: e.g. school counsellor] to a [role: e.g. clinical psychologist] for further assessment of a [age]-year-old. Sections: reason for referral, observed concerns, what's been tried, what I'm asking the receiving professional to consider, contact details (leave placeholder). Professional SA tone. No identifying detail in this draft — I'll fill that in afterwards. Concerns: [short summary] What we've tried: [short summary]
Tough conversation rehearsal
I need to have a difficult conversation with [a parent / a client / a colleague] about [topic]. Play their role and respond realistically as I practice. Push back where they likely would, including emotional reactions. Stop after 3 of your responses so I can adjust. Don't soften their reaction. I'd rather rehearse the harder version.
Translate jargon
Take the following [diagnosis / clinical term / policy clause] and explain it in plain SA English for [a caregiver / a learner of age X / a community health worker]. Avoid talking down. Use a relatable example. Keep it under 120 words. If there are cultural or language considerations specific to SA, mention them. Term/clause: [paste]
Find evidence-based options
I'm looking for evidence-based interventions for [presenting issue] in [age group] in [setting: school / community clinic / private practice]. List the top 5 with: 1. What the intervention is 2. What evidence supports it (with citation) 3. Where it's been used in SA, if anywhere 4. Practical constraints (cost, training needed, time) Be honest where evidence is thin or US/UK-only. Don't fabricate citations.
Session activity design
Design a 30-minute session activity for a [age]-year-old who is working on [goal: e.g. emotion regulation, articulation, fine motor]. The activity should: - Use materials available in a typical SA classroom or low-resource clinic - Have a clear opening, middle and close - Include something the child can take home or remember - Have a back-up if the first attempt doesn't engage Avoid materials that aren't easily found in SA.
Progress note for medical aid / funder
Draft a progress note for [funder type: medical aid / NGO funder / Dept] covering [period] of [intervention]. Sections: goals at outset, observable changes, current status, recommended next phase with rationale. Professional, evidence-led tone. Avoid claims of guaranteed outcomes. Flag any number I should verify before submitting. Goals: [summary] Observed changes: [bullets, de-identified]
Safeguarding-check on a draft
Read the following draft message / report. Flag anything that could be: 1. Misinterpreted as blame or judgement 2. Inappropriate to put in writing given safeguarding risk 3. A premature conclusion not supported by the evidence shared 4. A breach of confidentiality if forwarded Don't rewrite it — just flag what I should reconsider. SA context (POPIA, Children's Act, mandatory reporting). [paste draft]
Supervision prep
I'm bringing a case to clinical supervision. Help me prepare: 1. What are the 3 most useful questions I should ask my supervisor? 2. What might I be avoiding looking at? 3. What countertransference / parallel-process signals would be worth exploring? 4. What would a more experienced colleague in this field probably notice that I haven't? Case summary (de-identified): [paste]
Adapt resource for low-literacy context
Rewrite the following [handout / explanation / instruction sheet] for a caregiver with [Grade 6] literacy, in a low-resource SA setting. Use: - Short sentences - Concrete examples - No specialist vocabulary - Where possible, suggest a visual element (describe it) Preserve every material safety instruction in full. Original: [paste]

05Ethics & pitfalls in SA care contexts

If you read only one section twice, make it this one. The standards you’re held to as a registered care professional do not relax because the tool is new. Most “AI gone wrong” stories in care work aren’t malicious — they’re well-meaning shortcuts taken before someone realised where the line was.

POPIA, the Children’s Act & identifying data

South Africa’s Protection of Personal Information Act treats most public AI tools as third-party processors. Pasting a client’s, learner’s or patient’s name, ID number, diagnosis, address, school, or any other identifying detail into ChatGPT, Claude, Otter or similar free tools almost always breaches POPIA unless your employer has a formal data-processing agreement with the provider. For minors, the Children’s Act adds an extra layer of consent and confidentiality. Rule of thumb: de-identify everything before it leaves your machine. Names become “the learner”. Specific schools become “the school”. Specific dates become approximate.

Framework grounded in: Mogoale, Pretorius, Mogase & Segooa (2025), SA Journal of Information Management, on AI ethics in SA professional contexts.
Professional codes & scope of practice

Whichever body you’re registered with — HPCSA (psychology, OT, speech, medicine), SACSSP (social work), SACE (education), SANC (nursing) — the ethical codes apply to AI-assisted work as fully as to anything else. AI drafts; the registered professional decides and signs. If a regulator asks how you arrived at a clinical conclusion, “ChatGPT suggested it” is not a defence. Document your own reasoning, in your own words, on every consequential decision.

Cultural bias in mental-health & psychometric content

Large language models are trained predominantly on US/UK English content. Their default frames around mental health, family structure, parenting norms, religious context and emotional expression reflect that. For SA contexts — multilingual families, township and rural realities, intergenerational households, faith-integrated worldviews — the model’s first draft is often subtly off, and in mental-health work that can do real harm. Always edit AI output through your cultural lens; never publish without that pass.

Framework grounded in: Abdurahman et al. (2024), PNAS Nexus, on cultural homogenisation in large language models — particularly acute in psychological contexts.
Hallucination in clinical & legal information

AI tools are often confidently wrong. They fabricate references, mis-state medication dosages, invent SA-specific guidelines that don’t exist, and confuse similar diagnoses. For any clinical or legal claim destined for a case file, referral, or court document, verify it independently against SA-specific guidelines (HPCSA professional board guidelines, SANC scopes, DBE policy, etc). The cost of one fabricated dosage or one invented guideline is much higher than the time saved.

Framework grounded in: Olawade et al. (2024), Journal of Medicine, Surgery and Public Health, on AI in mental-health and clinical practice — emphasising verification.
Mandatory reporting trumps AI confidentiality

If something a client, learner or patient discloses triggers mandatory reporting obligations under the Children’s Act, the Sexual Offences Act, the Mental Health Care Act, or your professional code, those obligations apply whether or not you used AI to draft your notes. AI tools are not a confidential vault. Do not enter detail you wouldn’t be comfortable defending in a complaints process or in court.

06Where to go deeper

When you’re past the basics, here’s what to add — in roughly this order of marginal value.

Free structured courses (allocate 2-4 hours each)

SA professional development & CPD

Paid tools worth considering after 1-2 months of free use

SA-relevant communities

Your next step

Pick one workflow. Use it on a real task this week.

Don’t install five tools. Don’t bookmark ten courses. Pick the single workflow that maps to the task you’ve been dreading — the case note, the parent letter, the IEP — and use it tomorrow. Notice what comes back to you on the other side.

See the full path →

References

Abdurahman, S., Atari, M., Karimi-Malekabadi, F., Xue, M. J., Trager, J., Park, P. S., Golazizian, P., Omrani, A., & Dehghani, M. (2024). Perils and opportunities in using large language models in psychological research. PNAS Nexus, 3(7), pgae245.

Mogoale, P. D., Pretorius, A., Mogase, R. C., & Segooa, M. A. (2025). Ethical considerations in artificial intelligence-driven environments for higher education. South African Journal of Information Management, 27(1), a2007.

Olawade, D. B., Wada, O. Z., Odetayo, A., David-Olawade, A. C., Asaolu, F., & Eberhardt, J. (2024). Enhancing mental health with Artificial Intelligence: Current trends and future prospects. Journal of Medicine, Surgery, and Public Health, 3, 100099.

Tool descriptions, pricing in ZAR, and free-training links are accurate at time of publication and may change. Prices shown are typical retail; enterprise pricing varies.