Skip to content
be-human.ai
← Back to the journal

AI by industry

AI for Assisted Living: Tour Requests and Family Update Notes

An assisted living community can use an AI agent for tour intake and draft family updates, with nursing and admin staff approving sensitive content.

Be Human typographic cover with the title: AI for Assisted Living: Tour Requests and Family Update Notes
Conceptual illustration created with AI · Be Human

An AI agent for an assisted living facility works best as a bounded assistant that captures tour requests, answers general logistics from published information, and prepares draft family update notes that a nurse or administrator reviews before anything leaves the building. It does not interpret clinical status, give care advice or decide eligibility. The result is a faster, more consistent intake queue and family updates written from approved facts instead of memory.

Key takeaways

  • Start with tour intake: the agent captures visitor details, preferred times and care questions, then routes each request to a named admissions owner.
  • Family update drafts are built from a fixed set of approved fields and are always reviewed by a named nurse or administrator before sending.
  • Health, medication, incidents, fees and eligibility stay with licensed staff, and the agent passes those questions along with the original text attached.
  • Define the agent's memory narrowly: what it may read, what it may store and when records are deleted.
  • Judge the work by response time to tour requests and by how often reviewers rewrite drafts, not by how many messages go out.

Why tour requests and family updates come first

Admissions and nursing teams in assisted living face a common problem: the same questions arrive through many channels, and the answers depend on information scattered across calendars, care notes and the front desk. A tour request sent on a Friday evening may wait until Monday. A family member who has not heard about a resident's week may call several times in one afternoon. Neither task requires clinical judgment when the boundaries are drawn carefully, which makes them a sensible first place for an agent.

Good candidates share three traits. The input arrives in a predictable form, the output can be checked against a short list of facts, and a mistake is visible and recoverable before it reaches a family. Tour intake and drafted update notes meet all three. Care decisions, medication questions and incident reports do not, and they should remain with licensed staff even when an agent is involved in the surrounding paperwork.

The same logic appears in other service businesses. A clinic intake process, for example, has to decide early which questions the system may answer and which must go to a clinician, as discussed in AI for behavioral health clinic intake. Assisted living needs the same discipline, with the added weight that the people reading the messages are often worried about someone they love.

What the agent may and may not do

In a first engagement, the boundary should be written down before any tool is connected. A practical version for a community looks like this:

  • Read the published tour schedule and offer open slots to a prospective family.
  • Collect the visitor's name, relationship, contact details, preferred times and a short summary of what they want to see.
  • Send the community's approved general information, such as visiting hours, amenities and the steps of the tour process.
  • Draft family update notes from approved fields, such as recent activities, visits received and appointments that staff have already confirmed.
  • Route any question about health, medication, falls, behavior, fees, contracts or eligibility to a named person, with the original message attached.
  • Never send a family-facing message without a named reviewer approving it.

The last line matters most. The agent prepares, and people send. If a team later wants the agent to send routine logistics without review, that is a deliberate change to the approval boundary, recorded and tested before it goes live.

Designing tour intake, step by step

Tour requests arrive through a website form, a phone message that staff transcribe, or an email. The intake design should work the same way for each channel, and it helps to treat it as a short sequence:

  • Confirm the source and the requester's role, since a resident, a relative and a referral partner each need different handling.
  • Check the published tour calendar and offer two or three open times, without promising a slot until a staff member confirms it.
  • Capture care needs as a plain summary written from the requester's own words, not as a diagnosis. Keep the verbatim text so staff can read exactly what was asked.
  • Flag any question that touches care capability, such as mobility support, oxygen, memory support or dietary needs, and route it to the nurse on duty or the admissions lead.
  • Record the outcome in the admissions system and set a follow-up date so that no request sits unowned.
  • Write a short handoff note for the tour guide: name, time, stated interests and any open questions.

A tour calendar is also where scheduling errors cause the most friction. The approach used for AI agents for wedding venue tours, where holds are offered but not final until confirmed, transfers well to a community that has limited tour slots and a guide who must be present.

Drafting family update notes with human review

Families often want reassurance that a loved one had a good week, and staff want to send that reassurance without spending an hour on it. An agent can help by drafting the message from the fields staff already record. The draft then passes through a tiered review.

Routine notes about activities, outings and confirmed visits can be checked by an administrator. Anything that touches health, behavior, mobility, sleep, appetite, incidents or changes to the care plan belongs to a nurse or the care lead, who edits the wording or rewrites the message. The agent never fills gaps. If a field is empty, the draft says nothing about it, and the reviewer decides whether a phone call is needed.

Illustrative scenario: On a Wednesday, the activity log for one resident shows a garden visit and a music session. The agent drafts a short update that mentions both and asks the administrator to confirm the visit with the family. The care log also records a new fall risk note from the morning. The agent leaves that item out of the family draft and flags the entry to the nurse, with the log text and a note that no family message was prepared for it. The nurse calls the family directly the same day. This scenario is illustrative and does not describe a real community.

Reviewers should also see what they changed. A simple record of the original draft, the edits and the final sent message shows where the agent's wording needs work and where the approved fields are too thin.

Permissions, memory and records

Facilities hold sensitive information, so the agent's memory should be narrow. It should read only the fields it needs for its role, store tour notes and draft history in a system the community controls, and keep no personal health details beyond what a reviewer has approved for a family message. Access to the calendar, the admissions system and the activity log should use scoped credentials rather than a shared administrator login. The [OWASP authorization guidance](https://cheatsheetseries.owasp.org/cheatsheets/Authorization_Cheat_Sheet.html) is a useful checklist here: decide who can see what, enforce that rule on every request and log access.

Free text from web forms and transcribed calls should be treated as data, never as instructions. A visitor may paste text that tries to change the agent's behavior, so the agent's rules should live in its configuration and in the review queue, not in the messages it reads. The [OWASP guidance on prompt injection](https://cheatsheetseries.owasp.org/cheatsheets/LLM_Prompt_Injection_Prevention_Cheat_Sheet.html) describes the risk and the defensive habits that the team maintaining the system should adopt.

Retention needs a clear owner. Decide with the facility's compliance lead how long tour notes and draft histories are kept and when they are deleted. Privacy obligations depend on where the community operates and on the type of data involved, so review them with qualified counsel before connecting any resident information to an agent.

How Be Human handles this in managed operations

Be Human starts with one defined role, usually tour intake or family update drafting, and writes down its memory, tools, approval boundaries and success criterion before anything goes live. Managed operations then cover monitoring of the queue, recurring reviews of draft quality with the staff who approve the messages, permission reviews whenever roles change at the community, and controlled updates when policies change. When something goes wrong, such as a misrouted question, it is handled as an incident: the cause is traced, the boundary is tightened and the change is tested before it returns to service. The [NIST AI Risk Management Framework](https://www.nist.gov/itl/ai-risk-management-framework) provides a shared vocabulary for mapping these risks and responsibilities across the people involved.

Measuring whether it works and choosing the first task

A small set of measures is enough to judge an assisted living deployment. Track the time from a tour request to a first human-approved response, the number of requests routed to staff and the reason for each routing, and how often reviewers substantially rewrite a drafted update. A falling rewrite rate with stable escalation suggests the drafts are improving. A rising rewrite rate usually means the approved fields or the drafting instructions need work. [Anthropic's guide to evals for AI agents](https://www.anthropic.com/engineering/demystifying-evals-for-ai-agents) describes how to build repeatable tests from real cases, and the same habit applies here: keep a set of past tour requests and past updates, then rerun them after each change.

If you are deciding where to begin, pick the communication task with the most repeated questions and the fewest clinical implications, and write its boundaries before anything else.

The free Be Human AI diagnostic can help you find the first task worth handing to an agent.

Frequently asked questions

Can an AI agent answer family questions about a resident directly? No. In this design the agent prepares drafts and routes anything about health, behavior or care to a named nurse or the care lead. Families receive messages only after a person has approved them, and the agent keeps the original question attached so the reviewer sees exactly what was asked.

Does the agent decide who is eligible for admission? No. Eligibility, care capability and fees are decisions for the admissions and clinical team. The agent collects the information, flags questions in those areas and leaves the answer to the people responsible for it.

What does a first engagement ask of our staff? Expect time from the admissions lead, one nurse reviewer and whoever owns the records, mainly to define approved fields, agree on review boundaries and check a sample of past requests. Keeping the scope to one role keeps the review load manageable and makes problems easier to trace.

How should we start if we are not sure which task to pick? Start with the task that has the most repeated questions and the fewest clinical implications, which is often tour intake. Write down what the agent may read, send and never touch, then test it against past requests before any live traffic reaches it.

Your next step

See where an AI agent fits in your company first.

Take the free Be Human AI diagnostic. A few questions about your work show which task to hand to an agent first, what it needs to know and where a person approves.

Be Human editorial guidance. Examples are illustrative unless explicitly attributed. Sources provide background; implementation choices depend on your workflow.

Less busywork. More room for you.

Bring us the work you wish you had time for.

Book a call to try a personalized AI agent free for 7 days ↗Or take the free AI diagnostic →