The challenge
A UK care home group wanted its own team to build and update assessment forms. Forms change as practice and regulation change, and waiting for a developer every time slowed that down.
The group also knew that a do-it-yourself form is only as good as its questions. So it asked us for an AI form builder for care assessments that would help authors write them well.
Writing a good assessment question is a real skill. The wording has to be clear to a carer at the end of a long shift.
The answer options have to cover the situations staff actually face. Follow-up questions should appear only when they apply.
Get any of that wrong and the weakness spreads. A vague question produces vague answers, and vague answers produce a care record nobody can rely on.
What we built: an AI form builder for care assessments
We built the platform's form builder in 2024 with React, Node.js, Express.js and MongoDB. A new form gets a name, a category, a review schedule and a witness setting. The question builder sits inside it.
For each question, the author sets the details, the response type, any conditions and its place in the order. We connected this screen to OpenAI's ChatGPT API, so suggestions arrive while the author works:
Clearer phrasing. The AI reads the question in progress and proposes plainer wording.
Response options. It suggests answer choices that suit the question and its response type.
Conditional logic. It proposes which follow-up question should appear after a given answer.
Every suggestion is a prompt, not a change. Nothing is applied to the form unless the author chooses it.
How it works day to day
Picture a form author adding a new section to an assessment. They type a question and pick a response type.
As they work, the AI offers a tighter version of the wording and a set of answer options. It also proposes a follow-up that should only show when one particular answer is chosen.
The author keeps what fits, ignores what doesn't and reorders the questions until the section reads well. Then the form goes through the same checks as any other: the author previews it and saves it as a draft.
Staff see it only after someone deliberately publishes it. From then on, it feeds the completion screens, which show progress by section and remind staff about mandatory fields.
The result
Form authors get a second opinion on every question as they write it, without waiting for a developer or leaving the builder. They accept or ignore each suggestion, preview the form and keep it as a draft until they choose to publish.
People stay in charge of what the group asks about its residents. The AI helps with wording and structure, and a person decides what goes live. We have no usage figures for the suggestions, so we don't quote any.
What the research says
Shorter, clearer forms get finished. Baymard Institute's checkout research finds the average US checkout shows 23.48 form elements by default, against the 12 to 14 its testing points to (Baymard Institute). Suggested response options and conditional logic help an author ask a follow-up only when an earlier answer calls for it.
Most people check AI output before trusting it. Stack Overflow's 2025 Developer Survey found that 46% of respondents distrust the accuracy of AI tools. It also found that 33% trust it and only 3% trust it highly (Stack Overflow, 2025). In the builder, every suggestion is a prompt the author can accept or ignore.
AI suggestions still need a human eye. The UK Information Commissioner's Office notes that even high-quality, correct training data will not stop generative AI models from hallucinating. Such data can narrow the margin of error, though (ICO consultation response). That is why the builder offers suggestions and the author decides.
Where AI fits next
These are ideas only; the builder does not do them yet. It could flag questions that seem to confuse staff, based on how often they are left blank. The AI could also compare a new form with existing ones and point out duplicated questions for the author to resolve.
Planning something similar?
Before you let your own team use an AI form builder for care assessments, ask:
Who is allowed to create, edit and publish forms, and who checks their clinical content?
Which response types and conditions does your builder need to support?
How will authors preview a form the way a carer will see it?
What review schedule and witness rules belong on each form?
The builder belongs to the care and clinical platform behind this group's homes. Forms made with it feed AI writing help for the staff who complete them and nurse-approved AI care plan drafts. Read how we approach AI automation, or see more healthcare software projects.
