Drafting care and support plans
AI can turn assessment notes into a first draft in your usual structure. The risk is a plan that reads well, sounds like everyone else's, and was never really agreed with the person.
Our rating
Try with care
Can help, but needs the right tool, a data protection check and a person reviewing every output.
What it looks like
Before
Assessment notes: "Likes to get up around 8, not before. Needs prompting with meds. Daughter visits Sundays. Loves the radio, Radio 2."
With AI
Morning routine: prefers to get up around 8am. Medication: needs prompting. Family: daughter visits on Sundays. What matters to me: listening to Radio 2.
An invented example.
A support plan is the document that tells a care worker who has never met someone how to support them well. Writing one takes a long time, and most of that time goes on formatting rather than thinking.
AI can take assessment notes and produce a first draft in your structure. The question is what gets lost on the way.
How it works
You paste or dictate the notes from the assessment conversation, along with your plan template. The tool arranges what you gave it under your headings.
Note the word arranges. It is reorganising what you supplied, not adding knowledge about the person. Everything useful in the output was in the input.
Where it helps
Formatting, not thinking. If your template has fourteen headings and the conversation did not happen in that order, the sorting is genuinely tedious work that a tool does well.
Consistency across a team. Plans written by six people in six styles are harder to use at three in the morning than plans in one structure.
More time in the conversation. If drafting takes twenty minutes instead of an hour, that is time available for the part that matters — provided it actually goes back to the person rather than to the next referral.
Where it goes wrong
Everyone starts to sound the same. These tools produce likely text, and likely text is average text. “Enjoys social interaction and values her independence” could be anyone. “Won’t answer the door before eight and will tell you so” is a person. The second one is what a new care worker needs, and it is exactly what smoothing removes.
It invents the plausible. Given notes that mention a daughter, a tool may produce “family are closely involved in her care” because that is a likely continuation. Nobody said that. Now it is in the plan.
It writes over the person’s words. A support plan is one of the few documents where the person’s own phrasing has value in itself. “I don’t like a fuss” should survive into the plan as “I don’t like a fuss”.
Co-production theatre. This is the serious one. A draft that arrives already complete is very hard to disagree with. If the first time the person sees the plan it is fully formed, you have consulted them on a finished document. See AI and self-directed support.
How to use it without those problems
- Draft from the person’s words, not from old records. The input should be what they said in the assessment.
- Keep direct quotes as quotes. Ask the tool to preserve them, then check that it did.
- Show the draft as a draft. Visibly unfinished invites correction. Polished invites a signature.
- Read it against the notes, specifically looking for sentences that are true-sounding but were never said.
- Go through it with the person, rather than sending it for approval.
- Tell them AI was involved. See telling the people you support.
How to trial it
Start with plan reviews rather than new plans — there is an existing plan to compare against, so errors are easier to spot.
Take five recent plans you wrote by hand. Draft five with the tool. Put them side by side and ask a colleague who knows the people which set tells them more. That comparison is worth more than any time measurement.
If the AI drafts are faster but less useful at three in the morning, the tool has failed even though the stopwatch says otherwise.