The challenge
In a UK care home group, carers complete a behavior chart whenever a resident shows distress. They note what happened before, what the resident did, the consequences, the support offered and whether it worked. The group asked us for AI behavior chart analysis that could read weeks of those entries at once and show nurses the patterns.
One chart describes one moment. What the care team really needed was the story across many moments: which situations tend to upset a resident, and which responses calm them.
Finding that by reading chart after chart is slow, and patterns that run across weeks or times of day are easy to miss.
What we built: AI behavior chart analysis
The behaviors chart is our work too, scoped with the care team and built on React, Node.js, Express.js and MongoDB to run on web and mobile. Carers complete it whenever distress occurs, and each new entry schedules a chart review for the nurses.
Above the chart sits a Generate Summary button. A nurse sets the period to review, and the AI works through each entry inside it. It then answers a fixed set of 13 questions agreed with the care team, covering:
common triggers, and how the behavior is described;
recurring themes in how the resident expresses distress, and the usual outcomes;
how staff respond, which support is used again and again, and which strategies work or don't;
changes over time and links to the time of day;
takeaways for future care planning.
Before any real data was used, the care team gave us five fictional example charts with the analysis they expected. Those examples set the standard for the output.
The summary opens in a pop-up and downloads as a PDF. Entries also show day by day with a This Week button, so staff can scroll back through the raw records.
How it works day to day
Take a nurse picking up a chart review for a resident who has had a difficult few weeks. They set the date range, press Generate Summary and read what the AI found: the triggers that keep appearing, the support that tends to help and any link to the time of day.
Every summary ends with the same warning. It is AI generated, and all behavior charts must be reviewed before changing the care plan or the resident's care. So the nurse opens the day-by-day entries, checks the summary against them and decides, with the team, what to change.
The AI changes nothing. It reads and reports; nurses judge.
The result
Nurses get a single view of triggers, recurring themes, effective and ineffective support and time-of-day patterns for any period they choose, instead of piecing it together from separate entries.
Every summary carries a clear AI warning and sits beside the original charts, so the evidence is always one click away. No outcome figures exist for this feature yet, and we haven't estimated any.
What the research says
Care records are now mostly digital. As of March 2026, 83.7% of CQC-registered adult social care locations in England were estimated to have a digital social care record, up from 76.8% a year earlier (DHSC provider statistics). Once charts are typed rather than handwritten, software can read weeks of them in one pass, which is what the Generate Summary button does.
Language models can sound sure and still be wrong. NIST's profile for generative AI names confabulation as a core risk: false or mistaken content stated with confidence, often called hallucination (NIST AI 600-1, July 2024). That is why the summary answers 13 fixed questions and ends with a warning that every chart must be reviewed before care changes.
Even daily users check AI output. In Stack Overflow's 2025 Developer Survey, 46% of developers said they distrust the accuracy of AI tools, against 33% who trust it. Their top frustration, named by 66%, was answers that are almost right but not quite (Stack Overflow). The nurse's chart review builds that check into the care team's routine.
Where AI fits next
The chart does not do the following yet. The system could prompt a nurse when a new trigger appears that the care plan does not yet cover. The analysis could also focus on evening entries, to help teams track sundowning, which the care team's own example flagged as a gap.
Planning something similar?
Before you add AI analysis to distress charts, think through:
Which questions should every summary answer, and who on your clinical team agrees them?
What example charts and expected answers can you give the developers before real data is used?
What warning should sit on every AI summary, and who reviews the charts behind it?
Which date ranges matter for your chart reviews?
The chart is one module of the clinical platform we delivered for this care group. Its summary button works much like our AI activity summaries for care homes. Read about our AI automation work, or browse our healthcare software projects.
