Updated 20 September 2026: first published on Elite Agent in May 2025. The article has been refreshed as an AI basics guide to testing an AI receptionist and checking what happens after a missed call.
My first Lucy experiment started with a simple question: could an AI take a useful message when I could not answer the phone?
In July 2024, Tara Christianson pointed me towards Lucy by Curious Thing. I customised the greeting and accent, and less than ten minutes later I had a test message from my husband, Mark.
I liked that Lucy introduced herself as AI. The test messages had transcriptions I could review, and David from Curious Thing followed up to ask how I had found the experience.
That was a promising first test. For an agency, the next question is whether the whole process works, from the caller’s first sentence to a person taking responsibility for the follow-up.
What should an AI receptionist do for my agency?
Start with one job. Taking a message while you are at an open home is a manageable example. Answering every property question, booking appointments and updating a CRM are separate capabilities to check.
Write down the minimum information the team needs, where the message should go and who will respond. Agree what the assistant should say when it cannot answer, and what happens when a caller asks for a person.
Do not promise a transfer or callback time your actual setup cannot deliver. A clear message-taking service is more useful than an ambitious greeting with no reliable process behind it.
Test the awkward calls as well as the easy one
Use fictional details while you practise. Ask a colleague to try these situations:
- A straightforward enquiry about an advertised property.
- An unclear name or number that needs confirmation.
- A question the assistant has no approved information to answer.
- A caller who interrupts or corrects an earlier detail.
- Someone who asks to speak with a person.
- A call that drops before the message is complete.
For property management, also test how the system handles an urgent-sounding issue using your agency’s approved instructions. Do not let a model invent emergency advice or claim that help has been arranged.
Use AI to prepare a test plan
Help me prepare fictional test calls for an AI receptionist.
Its approved job: [SCOPE]
Information it may use: [SOURCES]
Minimum message details: [FIELDS]
Actual handover and follow-up process: [DETAILS]
Agency-approved urgent-issue instructions: [DETAILS OR NOT SUPPLIED]
Ask about missing process details first. Then create six short test scenarios, including an unclear name, a correction, an unknown answer and a request for a person.
For each, state what a good result would contain and what a reviewer should check in the conversation, message and handover. Use invented names and clearly fictional contact details.
Do not invent integrations, booking powers, callback promises or emergency instructions. Mark anything we need to establish before a live trial.Follow the message all the way through
Listen to the test and compare it with the transcript and summary. Check names, numbers, property references and the caller’s actual request. A polished summary can still omit the one detail your colleague needs.
Then check receipt and ownership. Did the right person see it? Could they tell what to do next? Was a promised follow-up completed?
Before using real calls, confirm the recording, data-handling and caller-information arrangements with the people responsible in your agency. Use those confirmed arrangements in the greeting and process.
The early demonstrations
The first video below is the 2018 Google Duplex appointment demonstration that helped make voice AI memorable. It concerned an outbound call. My Lucy test concerned answering an incoming call.
The second video accompanied the original Lucy experiment. Both are retained as historical viewing, rather than current product instructions.
Since then, our newsroom has taken the idea further: Scoop answers our 1300 number and passes messages to Lois Laine, our chief newsroom assistant. You can arrange an interview or create a free press release with Scoop.
For your first trial, keep the question simple: does the caller’s message get accurately to someone who will act on it?
The original version appeared in Samantha McLean’s AI Powered Agents newsletter. Read the refreshed source article on AI Powered Real Estate by Samantha McLean on Substack.