AI for Customer and Market Research: A Practical Workflow
How to Build Surveys, Synthetic Personas, and Test Your Content in One Afternoon
Research used to be the slow part. Recruitment. Fieldwork. Transcription. Coding. Synthesis. Six to twelve weeks before insight reached the strategy table. It can take such a long time to gather data.
But there are ways now that you can either complement what you're doing or speed up that process while still being the human involved. In one afternoon, maybe forty-five minutes, you can have workflows and techniques you can actually use straight away.
Demo 1: Build a Survey in Minutes
Not sure if you've ever used Google to build a survey. Generally you have to come up with the questions and then go in and manually build them, and it can take a bit of time, particularly if you're not familiar with how Google Forms works.
With this technique, we get the AI to draft the survey questions and then generate some code that we put into scripts.google.com. It will actually build the survey for us.
The workflow
Give ChatGPT a prompt with your research question, the number of questions you want, and the question type (like a five-point Likert scale)
Ask it to generate Google Apps Script code that will create this survey as a Google Form
Copy that code into scripts.google.com
Click Run
The survey appears in your Google Drive, ready to send.
No form builders. No drag-and-drop. No template hunting. That is so much faster than any survey we’ve ever built.
Beyond Surveys - The Harder Questions
Surveys are great, but they answer "what." The harder questions are "who" and "why."
There's been a lot of talk in the AI field, particularly in the marketing space, around this term called synthetic audiences. A synthetic audience is when you want to test something or see what an audience is likely to respond to, but using AI to take on the characteristics of that audience based on previous studies you've done.
It's a population of AI-generated respondents. Grounded in real data. Validated against real human benchmarks. Queried like a research panel. It returns distribution, not a single voice.
But there are different things people refer to when they're talking about synthetic audiences:
Synthetic persona: A single AI character representing a segment
Synthetic respondent: An LLM mimicking survey responses based on previous responses
Synthetic consumer: A model based on data from previous purchasing behaviours
Synthetic data: AI-generated data that originates in real data
Synthetic audience: A full population simulation, validated
Only the fifth is the rigorous version. Everything else is useful, but they’re not the same thing.
Why the Distinction Matters
There was one study where they used ChatGPT to estimate the UK's perceptions of Apple as innovative. The GPT estimate said sixty-five percent of the sample agreed. The actual figure from real survey data was twenty-nine percent. A thirty-six point error. Delivered with full confidence.
We cannot a hundred percent rely on a synthetic audience. But a synthetic persona, grounded in real testimonial data? That's the most accessible version of this methodology. This can be the entry point for a solo practitioner or small agency.
Watch the full webinar recording below where Dr Karen walks through all three demos live and build a complete synthetic persona from scratch:
Now that you've seen the workflow in action, let us walk you through how we built this for a real client.
Case Study: Real Client, Real Brief
Our client runs spiritual retreats. Her name is Natalie McIvor, the Vibrational Healer. She's trying to drive bookings for her upcoming retreats, but is now going for a more specific market. She generally has women at her retreats, and wants to speak more directly to mature women.
But we need to know more about them to do that well.
The brief
Client: Spiritual retreat facilitator
Goal: Drive bookings for upcoming retreats
Assets: Customer testimonial transcripts, AI audience research
Output: One Reel storyboard and caption, validated against persona
There are two things that are unknown. One is the audience. The other is what makes a Reel perform well in this niche. We can use AI to find out both, then create something based on research and test it with the target audience before we actually send it out into the world.
The Workflow
Step 1: Audience Research Agent
First, we use an AI agent to research the audience. The prompt is quite simple because we want it to actually go out and find information for us.
I prompted it to research women in the baby boomer category within Australia most likely to have the time, funds, and desire to attend a retreat to release past trauma in a nurturing environment.
When you run these things, it normally takes about ten minutes to really go and find that information. It came back with the age range, population size, time availability, health and wellbeing considerations. And the beauty of it is it's got all of these actual sources, and they're credible sources as well.
It contains trauma and PTSD prevalence. Financial capacity and spending priorities. Overall, a full profile of what this group is about and who would be most likely to want to go to a retreat like this.
Step 2: Build the Persona
We used Claude to create the persona. In Claude, we created a project and loaded it up with relevant documents:
The demographic and psychographic research report from the agent
Customer testimonial transcripts (we put the video testimonials through Descript, got the transcripts, and put all the transcripts from the target market into one document)
A transcript from Natalie herself talking about her practice
When we have a client, we interview them and ask them about their customers, their goals, everything they do. Then it's in their words. We get the transcripts and put that into the AI, and it's trained directly on information coming directly from our client.
The persona that came back was thorough. Meet Robyn Whitfield. Age 64. Lives in Maleny, moved up from Brisbane eight years ago. Divorced, not actively looking for another partner. Has adult children and grandchildren. She used to be a primary school teacher.
The persona also details the income, core values, background story, her fears, what she listens to, and how she makes decisions.
You can start to really get a feel for who she is as a person.
Step 3: Deep Research on Top-Performing Content
We used ChatGPT's deep research function with this prompt: What are the characteristics of the top-performing Reels that perform in the Australian wellness retreat niche?
It did an analysis from 2024 to 2026. Hooks, pacing, visual treatments, caption structures, sound choices, all included.
The beauty of it is we can get that report and put it into the AI. Then when we’re working with it to create something, it will automatically do what is working rather than just guessing.
Step 4: Create the Storyboard and Caption
With the persona built and the research loaded, we asked Claude to create one Instagram Reel concept that promotes Natalie's upcoming retreat.
It came back with a shot-by-shot direction. Cut to Natalie on the veranda. Mid-shot, soft natural light. Close-up on two older women's hands. One is writing on a piece of paper, the other is folding hers. And it was drawing directly from the testimonial transcripts.
Step 5: Test Against the Persona
Here's what we loved about Claude in this instance. When we asked whether Robyn would stop scrolling, it said yes. But when we asked what felt off, it didn't say "Oh, that was wonderful." It actually went shot by shot to critique it.
The woman on the veranda step holding a mug, looking out at the bush. That's exactly the wellness tired woman composition we've seen a hundred times. That kind of honest feedback before you create anything is exactly what you need.
What to Replicate This Week
Run one audience research agent on your top buyer segment
Build one persona grounded in real testimonial transcripts
Pressure-test one piece of upcoming creative against it
Start small. Trust the workflow. Layer in multiple personas later.
Where AI Does Not Replace Humans
We always must remember that AI can be very helpful, but we still need to drive it. Essentially, any decisions need to come from you because cultural nuance matters. If different cultures aren't reflected in the testimonials, it will just veer to what it has. Final decisions on high-stakes campaigns, anything genuinely unprecedented, or the "irrational" customer choice, these are all your responsibilities to get right.
Use AI to narrow the field. Use humans to make the call.
What audience segment have you been meaning to understand better? And how could you use these techniques to test your next piece of content before you publish it?
If you want support integrating AI strategically into your research and content workflows, here are some ways we can help:
Join Dr Karen’s AI Integration Sprint for communication leaders. Three weeks, and by the end you'll have an AI strategy for your organisation.
Join our free Facebook community where Dr Karen shares different AI tools and articles, and goes live every Friday afternoon talking about what's happened in AI this week.
Book a chat with Dr Karen to talk about how AI can work for your specific situation.
Because knowing your audience shouldn't take six to twelve weeks. Same rigour. Different workflow.
