Can Social Workers Use ChatGPT for Case Notes?

A social worker showed me her phone last week. She had just come out of a visit and was in her car, typing into ChatGPT, summarising what had happened so she could turn it into a case note when she got home. She was not doing anything reckless. She was tired, the system was difficult, and she had three more visits before the end of the day.

The answer is yes, social workers can use ChatGPT for case notes, with caveats. Tools like ChatGPT and Copilot can help you tidy and structure what you have already observed. They cannot do the thinking for you, and most local authorities have not yet caught up with policies on what counts as safe use. Pasting service-user details into a free public tool is a UK GDPR breach. Signing an AI-written note you have not properly checked is a Social Work England problem. Both are happening every day. Both can be avoided.

What "using ChatGPT" can mean

Two practitioners can answer "yes, I use ChatGPT" and be doing very different things. One records a verbal summary of a visit, runs it through a transcription tool, strips client identifiers, asks for a structured draft, then reviews and edits every line before pasting it into the system. The other pastes raw notes straight into a public chatbot, names included, and copies whatever comes back. The first is risky if done carelessly. The second is unsafe by default.

If you are managing a team, the question is not whether your staff are using AI on case material. They are. The question is whether anyone has shown them the difference between those two practices.

A few rules I would hold to

Identifying details come out first. Names, addresses, partial postcodes, school names, dates of birth, NHS numbers. Anything that could re-identify the person. UK GDPR does not bend because you are tired.

Every fact in the draft has to trace back to something you actually observed, were told, or read. If you cannot source it, delete it. AI does not know what happened in the room. It fills gaps with plausible-sounding guesses, and those guesses are easy to miss in a draft that reads fluently.

The note has your name on it. Social Work England's professional standards apply whether you typed every word or asked a tool to. You are accountable for tone, accuracy, what is included, what is left out, and anything the tool dropped because it sounded too uncomfortable to write.

What it does well, and what it gets wrong

What ChatGPT is good at is the same thing it has been good at since most of us first tried it. Structure. Tidying rough phrasing. Suggesting headings. Paraphrasing long chronologies once the source is in front of it. For a tired practitioner, that is not nothing.

What it does badly is invent. It hallucinates dates, supervisor approvals, quotations the service user never said. In an early matched-pair study I have shared on the TESSA research page, ten out of ten generic-prompted runs missed a clear racial pattern that the team manager had already recorded in supervision. Under a structured prompt, every run surfaced it. The prompt did not make the AI more ethical. It forced the verification work the practitioner would have done in their head, if they had the time.

The errors are rarely dramatic. They are the kind that get caught in supervision two months later, when the chronology in court does not match the case file.

A safer way to do it

If you are going to use AI on case notes, build the workflow around the verification work, not around the tool. Start with no identifiers. Use a structured prompt that tells the model what voice you want, what evidence it can draw on, what reasoning it should show, what attribution to include, and where the human stays in charge. The VERA:H framework I am building through my doctoral research at Nottingham Trent University is one shape this can take. The principle matters more than the acronym. Then read every line. Check facts, names, dates, quotes. Fix what is wrong, add what is missing, sign it.

If that feels like as much work as writing the note from scratch, you are doing it right. The tool saves you on structure and language. It does not save you on professional judgement.

What employers should be doing

Most of this conversation falls on individual practitioners, which is part of the problem. If your organisation has not provided a sanctioned tool, a clear policy, and proper training, the moral crumple zone is exactly where your staff are standing. Managers reading this: your team is already doing this. The choice is whether they are doing it with structure, training, and a tool you have governed, or in private with a free chatbot.

So, the answer

Yes, with care. The practitioners who use these tools well are the ones who treat them as a drafting assistant under tight verification, not as a shortcut around the job.

If you want a free starting point, the TESSA Responsible AI Framework is built for exactly this. If you want structured AI literacy training designed for social care, the TESSA Training platform covers this kind of workflow ethically and legally, for individuals and whole teams.

---

---