AI for Senior Living Facilities: What the Agent Can Tell Families

AI for Senior Living Facilities: What the Agent Can Tell Families

See what an AI agent for a senior living facility can tell families about visits and routine notices, and which care records stay with staff.

•

Want to apply this in your company?

Book a call with Juan →

An AI agent in a senior living facility can answer repeated questions about visiting hours, house rules, the published menu, the activity calendar, and general notices, using texts approved by management. It should not interpret health status, write clinical notes, or give guidance on medication, because those belong to the care team. The practical rule is simple: the agent shares what is public and agreed, and routes anything about care to a person.

Key takeaways

  • The agent can share visiting rules, schedules, and routine notices, as long as management and nursing have approved the content.
  • Health status, incidents, falls, medication, and changes in condition should never be answered by an automatic message.
  • Medical records, nursing notes, and medication administration logs stay in the facility's systems, under the care team's responsibility.
  • Any message with an urgent signal or a clinical question must go to a person, with a stated time for a callback.
  • Before the agent contacts families, map the current process, limit the data it can access, and agree on a success metric.

What an agent can tell families safely

The first question management must answer is what families can receive automatically without someone checking each message. Usually these are stable, public contents that already appear in brochures, notice boards, or the house rules. Once that content is written and approved, the agent reduces the front desk call queue without inventing anything.

  • Visiting hours, permitted days, and rules for staying during a visit.
  • Documents and items the family may bring, and items that are not allowed inside.
  • The published menu and general dietary restrictions set by the facility.
  • The activity calendar, outings, and commemorative dates that are already confirmed.
  • Maintenance notices, temporary access changes, and general announcements.
  • How to book a visit or a conversation slot with the coordination team.

The key distinction is between stating a rule and interpreting a situation. Saying that visits are allowed from 10 a.m. to 4 p.m. is information. Saying that a visit is recommended because the older resident is doing better is a clinical assessment, and that is not the agent's role.

What should never go out as an automatic reply

Some topics require a professional with clinical or administrative responsibility, even when the question looks simple. Families often ask about these topics on WhatsApp because it is the fastest channel, so the agent must be ready to stop at the edge of its role.

  • Current health status, test results, vital signs, and any interpretation of symptoms.
  • Falls, incidents, fever, refusal to eat, or changes in behavior.
  • Medication: schedules, dosages, changes, and reasons for suspension.
  • Requests for discharge, transfer, hospitalization, or procedures.
  • Contract, billing, and payment questions, which go through administration.
  • Anything involving another resident, which must never be disclosed to third parties.

In these cases, the right response is short. The agent acknowledges the question, says that the responsible team will get back to the family, gives the agreed timeframe, and logs the request. It should not try to reassure the family with phrases like "everything is fine," because it has no way of knowing that.

Which care records must stay with the team

The central point is to separate two kinds of information. The first is routine communication, which the agent can manage. The second is care records, which belong to the resident's medical file and care plan. Those must be written and reviewed by qualified staff, inside the facility's own systems.

In practice, the agent should not create or change nursing notes, physician or physiotherapist entries, vital signs, medication administration, fall reports, or incident accounts. It also should not become a storage place for family conversations that contain health information. When a family reports something important through WhatsApp, for example that the older resident seems confused since yesterday, that report must reach the team as a callback request. The responsible nurse then decides whether it goes into the medical record.

Health data is sensitive personal data under Brazil's LGPD, which requires extra care with purpose, access, and storage. Check the guidance from the national data protection authority (ANPD) and, when automated decisions are involved, read our piece on LGPD automated decision-making and the right to review. The questions to put to your vendor should be specific: where conversations are stored, who can access them, how long they are kept, and how a conversation is deleted on request.

How to build the flow in seven steps

A safe flow starts with the current process, not with the tool. The sequence below follows a simple method: map, limit, pilot, and hand off.

  • Gather the real questions received over the past few weeks by phone, at the front desk, and on WhatsApp.
  • Sort each question into one of three groups: inform, schedule, or escalate.
  • Write the answers for the "inform" group and get formal approval from management and nursing.
  • Define the escalation triggers: words, topics, and urgency signals that always go to a person.
  • Define who handles escalations, during which hours, and within what agreed timeframe.
  • Test the flow with simulated questions, including difficult ones and attempts to obtain another resident's information, before exposing it to families.
  • Review conversation logs every week, focusing on incorrect answers and escalations that took too long.

If the channel is WhatsApp, observe the rules for the service window and outbound messages. Our article on the WhatsApp 24 hour customer service window explains how an agent should act outside that window. The facility should also avoid sending proactive notices without recorded permission.

Illustrative scenario: weekend visit and a health question

Illustrative scenario, not based on any real facility: a family asks on WhatsApp whether they can visit on Saturday afternoon. The agent checks the approved rule, states the permitted hours, and offers to book a slot with the front desk. A few minutes later, the same family writes that the older resident did not want lunch and asks what happened.

In the second message, the agent does not try to explain or suggest causes. It replies that the question will be passed to the responsible nurse, shares the callback time set by the facility, and creates a contact request with a short summary of the question, without copying unnecessary details. The nurse sees the request, reviews the day's care notes, calls the family, and records what was done in the medical file. The agent keeps only the log of the contact, without turning it into a clinical record.

What makes this flow reliable is not the sophistication of the model. It is the clarity about who decides each thing.

When to keep, adjust, or stop

After the pilot, management needs to decide based on logs and on a metric agreed before the start. A simple guide helps make that decision without relying on impressions.

  • Keep the setup when automatic answers are correct, escalations reach the team on time, and families get what they need without calling.
  • Adjust when many repeated questions have no approved answer, or when the team reports escalations that were poorly summarized.
  • Narrow the scope when the agent answers something outside the approved rules, even in a minor way, and treat that case as an incident.
  • Stop the pilot when there is an incorrect answer about care, exposure of another resident's information, or a failure in access to care records.

It is also worth being clear about what the agent should not become. A tool that replaces listening to families does not resolve the worry that often sits behind their messages. For a broader view of automating customer contact and what should stay with people, the article on AI customer support and what to automate offers criteria that apply well to this setting.

How we apply the method on this kind of project

In our work, the first step is to map the current communication process: who receives the message, who responds, where records are kept, and which topics already carry risk today. Next, we limit the data and the actions, so the agent only accesses what it needs to share public rules. Then we pilot a real flow with logs and a success metric agreed with management. Finally, we hand the operation to the facility's team with a clear plan to expand, adjust, or stop. When the operation is managed, Be Human delivers it. For general risk management of AI systems, the NIST AI Risk Management Framework is a useful reference for structuring governance.

Frequently asked questions

Can an AI agent tell families about a resident's health status? No. Health status, symptoms, test results, and incidents require assessment by a qualified professional. The agent should acknowledge the question, state the team's callback timeframe, and log the request for escalation, without trying to reassure the family or interpret the situation.

Where should care records be kept if the agent works through WhatsApp? Care records such as nursing notes, vital signs, administered medication, and fall reports must remain in the facility's medical record or system, under the team's control. The agent can keep only a log of the contact, with a minimal summary, and should not turn the conversation into a clinical document.

Does a senior living facility need authorization to send routine notices on WhatsApp? It depends on the type of message and how the contact was obtained, and the question should be checked with legal counsel and with the platform's rules. In general, replies within a conversation the family started follow different logic than proactive outbound messages. That is why it is advisable to record the basis for permission and review message templates before any proactive sending.

What is the safest way to start with an agent in this setting? Begin by mapping the real questions, approving in writing the answers that can be automatic, and defining the escalation triggers. Then run a small pilot with simulated tests, weekly log reviews, and a metric agreed with management. Expand the scope only when the team confirms that escalations work as intended.

Sources