Archive note (September 2026): First published in March 2025. The podcast story and videos remain. The 100 personas in the experiment were simulated, and the revised prompts show how to use their responses as questions for real audience research.
By the time I wrote this in March 2025, the Elevate Podcast had been around since 2017, reaching millions of downloads for its 500+ episodes (I’d lost count).
It’s hard to believe the whole thing started quite by accident in one season of Transform, where we had a lot of video content and no time to edit it.
Our intern videographer then suggested that we start a podcast because audio alone would be easier to edit than video.
So we brainstormed a name that started with E (fitting with our other E-themed words like educate, entertain, and so forth).
Elevate was the obvious choice - we already used it in our branding, and it felt right.
We had the first episode out within a week—classic us, diving in headfirst.
Now, as I look back, I see so much I would do differently if I were starting a podcast again.
I would do more audience research on the title, tagline, SEO and other things.
I'd be more intentional about the programming.
Of course, now that AI is here, there's something in me that wants to bring together the best of Elite Agent and the best of AI-powered agents in a medium which is both video and community first.
The idea that inspired my experiment
At the time, I listened to Michael Taylor on a podcast discussing his approach to audience testing using AI. Michael is a prompt engineering expert and former marketing agency owner who’s developed a fascinating method for testing content before publishing it to real audiences.
What caught my attention was his step-by-step approach, which anyone can use.
If AI can roleplay one character, why not explore several perspectives? That can be a useful brainstorming exercise. It still gives you generated opinions, not an actual focus group or a representative sample of listeners.
Side note: If this reminds you of the time that Janet created “rebound guy” Derek and then a whole neighborhood on the NBC TV show The Good Place, there are some similarities…
The Workflow I Used for Testing Podcast Names
Here's how I applied Michael's method to test potential names for my new real estate podcast:
1. Create Diverse Personas
In the original experiment, I asked ChatGPT for 100 AI personas representing different types of real estate professionals. The revised prompt below uses eight fictional perspectives as a smaller, more focused starting point:
Help me explore possible reactions to a podcast about [topic]. My intended audience is [roles and context]. Any actual audience findings I can supply are: [findings, or none yet].
Create eight clearly fictional perspectives based on different work responsibilities, experience with the topic and listening needs. Avoid invented demographic facts or stereotypes about age, income or family status.
For each perspective, explain what they might hope to learn, what could confuse them and one question I should ask a real listener. Label every unverified assumption.
These are brainstorming lenses, not research participants. Do not claim they represent the audience or estimate how common their views are.For each persona, I included details about their work style, pain points, and what they look for in professional development content. I ran the original prompt several times to generate fictional perspectives from different country contexts; that did not establish what agents in those countries actually thought.
2. Generate Podcast Name Options
I also used AI to brainstorm several options that aligned with my vision for a community-first, forward-thinking real estate podcast and came up with the following list (some you’ll see are familiar):
The Spark Sessions
The Leadership Diaries
The Next Level Agent
The Strategy Lab
Agency Amplified
Amplified Agency
The next horizon
Real Estate Revolution
3. Test With Simple, Direct Questions
Following Michael's advice about keeping questions straightforward, I created a prompt to ask each persona:
"Which of these podcast names would you be most likely to click on and listen to?"
"Why does that name appeal to you more than the others?"
"What topic would you expect to hear discussed on a podcast with this name?"
Using the fictional perspectives above, explore possible interpretations of these podcast names:
1. The Spark Sessions
2. The Leadership Diaries
3. The Next Level Agent
4. The Strategy Lab
5. Agency Amplified
6. Amplified Agency
7. The next horizon
8. Real Estate Revolution
For each name, describe the topic it could suggest, a reason it might attract someone and a reason it might confuse or put someone off. Use only the listed names. Label all reactions as simulated hypotheses.
Do not count generated preferences as survey votes, calculate audience percentages or predict listening behaviour. Suggest three neutral questions I could ask real intended listeners to compare a shortlist.Image: Ideogram
4. Look for ideas worth testing
These were patterns in the model’s output, not findings about real age groups or job roles. One response even favoured “Real Estate Reimagined”, which wasn’t in the eight-name shortlist above. That inconsistency is a good reason to inspect the output rather than treat its confidence as evidence.
Fictional output from the original experiment:
Younger agents gravitated toward tech-focused names like "The next-level agent” (so did our team)
Experienced agents preferred "Real Estate Reimagined" as it suggested innovation without being intimidating
Broker-owners liked "The Leadership Diaries" as it implied actionable tips from other leaders
What I Learned From This Process
Here is what I would take from the experiment:
I could explore several angles quickly in this experiment.
Generated criticism can raise useful questions, but it is not inherently honest, independent or free from flattery.
The explanations gave me language to examine; they did not prove why real people would choose a name.
I could develop a shortlist before asking actual listeners for their time.
I did not price a comparable research study, so I would not claim a specific cost saving.
How You Can Apply This Method
If you’re considering a rebrand, new product launch or content series, use the generated ideas to prepare a test with real people:
Define the real audience and the intended promise.
Generate possible interpretations of the names.
Choose a short list.
Ask real listeners the same neutral questions, varying the order where practical.
Record their actual words and compare what they expected with what the podcast delivers.
A small convenience sample is still a limited sample; do not present it as the whole market.
The useful output is a better set of questions and a shortlist to test. The evidence comes from what actual listeners say and do. AI can help organise that feedback, provided you keep their words separate from its suggestions.
The next test is with people
I ended the original post by asking readers to choose from a shortlist. That poll is not currently present on this page, and I’m not claiming a winning result here. If you’re trying the exercise now, ask intended listeners what each name makes them expect, which they would investigate and why.
Happy hunting 🚀
—Sam