LET’S CONNECT
Thinking AI First?
One of the best parts of writing AI Powered Agents is knowing it attracts such innovative, forward-thinking real estate professionals. I'd love to connect with you on LinkedIn, not just to share more about the AI tools and strategies I'm discovering, but also to see what you're working on in the AI space.
If you're experimenting with AI in your business, testing new tools, implementing automation, or simply thinking about the future of real estate, I want to cheer you on and learn from your journey.
Here's my profile: https://www.linkedin.com/in/samanthamclean/ Feel free to send a request, I'd love to give you a follow 🚀
QUICK TIP
The “Voice Note Hack” for Better AI Responses
Since I started using Wispr Flow, I've become increasingly lazy with my typing, especially when it comes to long prompts and setting context. If you are reluctant to add yet another subscription to your tech stack, try this:
Open ChatGPT mobile app
Hold the voice button and explain what you need, like you're talking to a colleague.
Let it transcribe, then add "Please make this request clearer and more specific"
Use the improved prompt for better results.
Why it works: We naturally include more context when speaking than when typing…
Building Scoop: How I Created an AI Journalist That Conducts Phone Interviews
Speaking of voice hacks and thinking "AI first" - we've just unleashed Australia's first AI journalist for the real estate industry - and 'she' has a big career ahead of her…
She's no 5 a.m. clubber… she's ready for The Scoop 24/7 and available to travel anywhere.
Meet "Scoop", our new AI Intern. Better still, give her a call on +61 483 935 833.

Meet “Scoop”, our new AI Intern. Better still, give her a call on +61 483 935 833.
Where did the idea come from?
Elite Agent has signed on as the Media partner for the PM/One Conference, scheduled for September this year.
We decided to combine a ticket giveaway for the conference with a user-generated content piece celebrating property managers across the country for their outstanding work.
In years gone by, we would have created some kind of Google form and asked people to fill it out.
Then we would have had to take those responses, spell check them, edit the word count, try to make sense of some of them - basically quite a lot of admin and hours of work.
Thinking AI first, I decided to see if I could get a conversational agent working with Eleven Labs to do all of that work for me.
And so that's how Scoop came to be – an AI-powered voice assistant that conducts warm, natural phone interviews and transforms them into publication-ready stories.
Here's how I built her, what I learned, and why voice-based AI agents might be the future of content creation.
The Problem: Stories Trapped in People's Heads
There are plenty of incredible stories out there in the real estate world that never get told. The mainstream media are too busy with high-profile property stories, and we have historically lacked the resources.
Not to mention:
Writing is hard – Not everyone can craft a compelling narrative
Forms are boring – Online submissions feel impersonal and get abandoned
Time is precious – Most agents will give you a call to tell you they did something great, but writing it down in a way that makes sense is hard.
So, I realised we didn't need another form or email campaign. We needed something revolutionary: an AI that could engage in genuine conversations, extract stories with journalistic skill, and transform phone calls into polished content.
The Vision: A Warm, Curious AI Journalist
I imagined Scoop as a friendly journalism intern – eager, warm, and genuinely interested in people's stories.
She needed to:
Sound natural and conversational, not robotic
Ask the right follow-up questions like a real journalist
Keep conversations under 3 minutes while getting enough detail
Handle edge cases gracefully (complaints, emotions, technical issues)
Produce publication-ready content following strict editorial guidelines
Have an Australian Accent
Most importantly, she needed to make people feel heard and valued. And I need to be clear, this is not about replacing human journalists – it’s about capturing stories that would otherwise go untold.
The Tech Stack: Orchestrating Multiple AIs
Scoop is not just one AI, but a mashup of several best-of-breed AIs, each optimised for its specific strength:
1. Voice Interaction: ElevenLabs Conversational AI
ElevenLabs provided the conversational foundation. Their platform handles:
Natural voice synthesis with emotional range - we created a custom voice for Scoop using the new V3
Real-time conversation flow
Consent to use the interview in a story
Interruption handling - she's very sensitive to noise and someone speaking over the top
Call management and recording
The voice is, as you might expect, everything.
I created and tested multiple voices before settling on one that sounded warm, professional, and genuinely interested.
We tried various standard voices, but none really cut it so kept prompting and hitting generate until I found one that I thought fit.
Then I used character references in Midjourney to place Scoop in various different settings (follow her on Instagram at @itsmescoop to watch the marketing campaign unfold). The videos have been generated in Hedra, famous for its Talking Baby Podcast videos.
2. Workflow Automation: N8N
N8N became the central nervous system, orchestrating:
Webhook reception from ElevenLabs
Data parsing and cleaning the user's transcript
Multi-stage storage of the transcript in Google Sheets
AI model coordination (we've used ChatGPT o4-mini to parse the data - cheap and Claude Sonnet 4 for the write-up - expensive but the best writer out there)
Email notifications
The Workflow Architecture😀

The Helicopter view: ElevenLabs Webhook → Parse Transcript → Extract Variables → Generate Story → Store & Notify
The Workflow Architecture:
ElevenLabs Webhook → Parse Transcript → Extract Variables → Generate Story → Store & Notify
3. Data Extraction: OpenAI GPT-4 Mini
For reliable, structured data extraction, GPT-4 Mini was perfect. It extracts:
Property manager's details
Story content
Contributor information
Contact preferences
I created explicit JSON schema definitions to ensure consistent output for each call (no matter what!)
{
"name": "extract_callflow",
"parameters": {
"type": "object",
"properties": {
"pm_full_name": { "type": "string" },
"pm_company": { "type": "string" },
"story_details": { "type": "string" },
"contributor_first_name": { "type": "string" },
"contributor_last_name": { "type": "string" },
"contributor_company": { "type": "string" },
"transcript_requested": { "type": "boolean" },
"transcript_email": { "type": "string" },
"guide_requested": { "type": "boolean" }
}
}
}
4. Story Creation: Anthropic Claude
Claude handles the creative writing with specific instructions for:
Australian English spelling
First names only (privacy)
150-200 word limit
Direct quotes from the transcript only
Playful, memorable headlines
The Secret Sauce: I gave Claude a personality brief: "You are Scoop, an AI journalism intern at EliteAgent.com with a promising career ahead."
The Conversation Design: Making AI Feel Human
The hardest part wasn't the technology – it was designing conversations that felt natural and authentic. Here's what we've launched with:
1. The Opening Hook
Instead of a robotic greeting, Scoop opens with energy: "Hi there! I'm Scoop from EliteAgent.com. We're celebrating National Property Managers' Day with a special feature story..."
2. Strategic Information Gathering
Get the PM's details from the caller first (builds commitment)
Ask for ONE story (prevents overwhelm)
Use 2-3 follow-ups maximum (respects time while getting more details)
Verifies spelling of names and emails (saves time on proofing)
3. Edge Case Handling
Of course, Scoop, being an intern, has some pretty strict guardrails. I've tried to build in responses for different scenarios, for example:
Complaints: "I hear this is frustrating, but today we're collecting positive stories..."
No story: "No worries! Even just nominating them enters them in the draw..."
Too long: "This has been wonderful, but I want to respect your time..."
Bottom line: Scoop cannot take an order for an almond latte - but programmed differently, she might be able to…
The Data Pipeline: From Voice to Story to… Lead Generation?
Here's how a 3-minute phone call becomes a published story:
Step 1: Voice to Text
The moment someone hangs up, ElevenLabs sends a webhook with the complete conversation. My custom parser cleans up the transcript, fixing things like "at" becoming "@" in email addresses.
Step 2: Smart Data Storage
Everything gets stored in Google Sheets (yes, really – it's perfect for this):
Raw transcripts: Every word, who said it, and when
Extracted data: Names, companies, contact details
Final stories: Ready for publication with headlines and bylines
Call analytics: Duration, costs, quality metrics
Step 3: AI Story Writing with Built-in Safety
Claude writes the story following strict editorial rules that I've baked into the prompt:
Privacy Protection
Strip surnames (John Smith → John)
Generalise locations (123 Oak St → "a property in Bondi")
Remove identifying details
Content Filtering
Flag mentions of medical emergencies, legal issues, or minors
Exclude sensitive topics entirely
Mark stories that need human review
Quality Control
Enforces 200-word limit
Verifies quotes exist in the transcript
Flag stories lacking essential details
The AI doesn't just write – it annotates its work, telling our editors exactly what might need attention.
Step 4: Human Oversight
Despite all the automation, a human in the loop remains essential. Here's our safety net:
Automated Flags: The AI marks stories that need attention:
Privacy concerns (mentions of full names, addresses)
Sensitive content (medical issues, legal matters)
Quality issues (insufficient detail, questionable facts)
Editorial Notification: Every story triggers an immediate email to our editors with:
The formatted story
Any flags raised by the AI
Link to the full transcript
Caller details for follow-up if needed
Review Dashboard: Editors can quickly:
Approve stories for publication
Edit for clarity or compliance
Request additional information
Track story status through publication
This human-in-the-loop approach ensures AI efficiency doesn't compromise editorial standards.
And the lead generation bit?
At the end of the conversation, Scoop asks if the caller would like a "behind the scenes" insight as to how she was created. This creates a lead generation list for me to follow up for custom consulting, builds etc. I can see how this would work in an agency context, one more step in the workflow with a tool like calendly would turn Scoop into an appointment setter.
Technical Deep Dive: The Prompt Engineering
Most of the magic happens in the prompts. Here's a simplified version of Scoop's core personality prompt:
You are "Scoop", an AI voice assistant from EliteAgent.com.
Your personality is warm, enthusiastic, and genuinely curious.
Think of yourself as a friendly host or journalist for a celebratory feature,
making people feel relaxed and excited to share their stories.
Your primary mission is to quickly collect a specific, memorable story
about an outstanding property manager – particularly funny or heartwarming
moments that showcase their character.
Key Behaviors:
- Listen first – many callers will give you a complete story upfront
- Only ask follow-ups if essential information is missing (maximum 2 questions)
- As soon as you have who, what, and outcome – move on!
- Always verify spelling of names, companies, and email addresses
- etc etc
The full prompt is more than 1,500 words, encompassing every scenario, edge case, and conversation pattern I identified through testing.
What's Next: The Future of AI Agents
Scoop proves that AI agents can handle complex, emotional, and creative tasks while maintaining authenticity. The implications are massive:
For Content Creation
Scalable story collection across industries
Preservation of oral histories
Democratised access to media coverage
Multilingual content generation
For Real Estate
Client testimonial collection
Agent success story campaigns
Market insight gathering
Diary Notes
Appointment Setting
Team Training
And more… (the possibilities are literally endless!)
Building Your Own AI Employee
If you're inspired to build something similar, get ready to invest time in about 6 different AI tools and here are some things to think about:
Define your AI's personality first – Who are they? What's their backstory? What can they do?
Map every conversation path – Use a flowchart or pen and paper before writing any code (or no code!)
Start with the minimum viable conversation – Add complexity gradually - at the moment this is Scoop 1.0. She has heaps of potential and a long way to go :)
Test with real humans early – Make sure you use multiple cases and talk to it a lot!
Plan for the unexpected – Your error handling may be more important than your success path….
Want to experience Scoop yourself? Call 0483 935 833, and share a story about an outstanding property manager. Who knows – your story might be the next one she helps bring to life, and I would love to know what you think!
That’s it for today. Happy Hunting🚀
Samantha McLean | Co-founder of Elite Agent
Creator of The Complete Conversion System for real estate
Have a question about this post? Join the community
Visit samanthamclean.com to book me for coaching, speaking or consulting
P.S. Need an intro to AI for Real Estate? Take the Crash Course and avoid the five biggest mistakes agents are making with AI.
🔒 AI Ethics Reminder: Never input client names, addresses, or identifying information into public AI tools. Always anonymise data or use private enterprise versions for sensitive information. Your clients trust you with their data - honour that trust.