7 Steps to Turn Raw Customer Research into Insights with ChatGPT
I love speaking with users and experts to learn from their experiences. But after 3+ conversations, my real challenge begins - structuring notes, identifying patterns, and sharing insights.
So, I used a new process in 3 projects: I am the UX Researcher, and ChatGPT is my AI Assistant.
Now, I’m not afraid of User Experience Research (UXR) synthesis, I look forward to it.
The steps:
Step 1 - What Did You Want To Find Out?
Step 2 - Whom Did You Speak With?
Step 3 - What Questions Did You Want To Get Answers To?
Step 4 - Clean Up Grammar
Step 5 - Remove Sensitive Information
Step 6 - Shift from Person-Based to Question-Based Analysis
Step 7 - Refine Themes And Categories Iteratively
I spend 40 minutes writing this. You need 4 minutes to read the post (excluding the prompts).
Related:
Customer Research Background
When I talk to customers, I love learning new things—whether from users, experts, or anyone with valuable insights. As a Product Manager, I conduct informational interviews, shadow customers, observe their daily work, and ask about their experiences. While I enjoy these discussions, a single conversation is merely an anecdote. Reliable decisions require multiple perspectives.
Once I have notes from several conversations, organizing them to identify key themes, action items, and insights can feel overwhelming. That’s when I turned to ChatGPT—but I’ve found that using it alone isn’t enough. So, I developed a structured approach that combines AI and human analysis. I’ll share the approach here.
Save Time, Improve Quality, Reduce Hesitation
An AI assistant for customer research can help you:
Reduce hesitation when reaching out to more customers
Make note-taking and takeaway creation effortless
Improve writing quality by making it easier to find relevant quotations
By streamlining these tasks, you can focus on what truly matters—understanding your customers.
Step 1 - What Did You Want To Find Out?
Write one or two sentences explaining why you conducted these interviews.
Example:
"I interviewed AI developers to understand their inference infrastructure needs, decision-making processes, challenges, and problem-solving approaches. My goal is to design a product that addresses these challenges."
Example:
"I conducted interviews with 24 families to understand how they navigate school applications and choices. I explored their decision-making processes, challenges, and key factors influencing their choices. The goal was to uncover insights that inform effective school selection strategies."
Step 2 - Whom Did You Speak With?
Create a table where each row represents one person you spoke with. If relevant, add another column with relevant details such as company type, family background, or career experience.
Step 3 - What Questions Did You Want To Get Answers To?
List the questions you asked interviewees to understand their needs, challenges, and decision-making processes.
For example:
What do you like about this product?
What are your biggest pain points?
How do you currently solve this problem?
What tools do you use for this?
Before speaking with users or customers, I prepare a list of questions. I now reuse that list during this UXR Synthesis process. Michele Hansen’s Deploy Empathy is gold for user interviews.
Each question becomes a separate column in my spreadsheet. This helps me organize responses clearly and compare answers across different interviewees.
Step 4 - Clean Up Grammar
While taking notes live during calls, you may use shortcuts or voice typing. To ensure accuracy, start by cleaning up grammar and fixing typing errors.
For example, I use voice typing or shortcuts when capturing conversation notes, then use AI to synthesize them.
Review notes per interviewee (a spreadsheet’s row) because proper nouns or accents are specific to each person. Reviewing all of one person's information together makes it easier to correct errors.
To avoid AI-generated inaccuracies, instruct AI to write "not available" or leave fields blank when necessary. Also, ask AI to include supporting quotes where possible.
Here’s a sample prompt I used. You can change it to output in your voice and style and for your project:
"I have notes from interviews with families about their school selection decisions. I'll paste convo notes of one parent at a time. Make small modifications to my writing. edit each cell of information. keep the information as a paragraph within each cell, i.e. not bullet points. i want the information back as a table. Use simple words. Repetitive words are acceptable; there's no need for variety. Focus on ease of reading. Use an active voice. Correct grammar and typos. I want a confident tone. Explain and inform about processes. Highlight challenges and provide detailed solutions. The aim is to not only inform but also enhance the reader's understanding and appreciation of the topic. Stick to the information I've provided. Do not add any new information. mark empty cells as n/a. give output in csv format with each cell in """""
Sometimes, I receive responses as voice messages (e.g., on WhatsApp). I transcribe them and then ask ChatGPT to synthesize the notes.
Here’s a sample prompt for that:
"Read these voice notes and fill in the table per question. Each response should be a paragraph within the corresponding cell—no bullet points. If there is no information for a cell, mark it as 'n/a.'
questions / table to fill
| Parent | Children Name, Age | Location | Which school do your children attend? | What other schools did you consider before choosing this school? | What appeals to you about this school over other options? | How was your experience with the application process and the wait list? | How do you manage the daily commute to school and work? How did the commute logistics affect your school choice? | What after-school childcare does your child go to? How did these options influence your school decision? | How was your after-school application process? | Are there any secondary schools you are considering for your child? How does your primary school choice align with it, for example, as a feeder school? | What is the impact of school's alumni network towards opportunities or community? | How do you evaluate a school's quality? For example, academic performance, GAAs, quality of teachers, student demographics, facilities, etc.? | You've been very helpful and I've learnt a lot. Any other advice you have for me? | follow up | curriculum, school deep-dive |
voice notes from the person to us"
Step 5 - Remove Sensitive Information
If you need to remove personally identifiable information (PII) from your insights, follow this step. I used this when speaking to families about their children's schooling decisions. However, I didn’t use it when quoting consultants who shared their experiences with me, nor for customer interviews.
Here’s a sample prompt:
"I have notes from interviews with families about their school selection decisions. I'll paste convo notes of one parent at a time. i want you to keep the names of the adults and children only in the """"parents"""" and """"children"""" columns. I want you to remove he children's names from other columns. replace it with kid1, kid2. or boy, girl, or kid aged 4.5, a kid aged 7. Edit each cell of information. keep the information as a paragraph within each cell, i.e., not bullet points. I want the information back as a table. give output in CSV format with each cell in """""
Step 6 - Shift from Person-Based to Question-Based Analysis
Earlier, we processed each interviewee’s responses together—for example, reviewing all responses from one person across all questions.
Now, we pivot. Instead of analyzing responses by person, we analyze all users' responses to a single question. This approach helps surface patterns more effectively, making it easier to identify themes across multiple conversations.
Here’s a sample prompt:
"I asked 24 parents about their experience and decisions in applying and choosing schools for their kid(s). I want you to review the responses to one question from all parents and synthesize it. i will give you one question and all answers to it. write most of this in past tense. combine the quotes, common themes, and variant views. here is the format you should use for the output:
step 1 identify Common Themes: Identify and describe up to 5 themes that emerge from the responses.
step 2 Variations/Divergent Views: Highlight any unique responses or differing opinions that provide an alternative perspective or highlight less common concerns. per theme that you identified in step 1. do step 1 before doing step 2.
step 3 Representative Quotes: Include 2-3 quotes per theme from different parents to backup or show a divergence per common theme you've identified.
quantify how many rows were showing up in each theme instead of writing ""many"" or ""significant number of"".
although i will paste a bigger part of my table, i want you to focus on: ""Q: What other schools did you consider before choosing this school?"" but to help you answer this question, i'm pasting context from other questions."
Step 7 - Refine Themes And Categories Iteratively
I reviewed ChatGPT's output to identify themes that did not make sense. For example, if one theme was "Started consulting for flexibility" and another was "Started consulting for family time," I grouped them for clarity.
UXR Synthesis Completed
At this stage, your UX research synthesis is complete. You can now share responses to each of your questions, including relevant user quotes for your audience.
Example
“Q: How do you get your clients?
I heard these methods to acquire new clients from consultants. In hindsight, I should have asked, “How did you get your previous client?” and asked follow-up questions. Starting with an emotional memory would’ve been better than an aggregate memory.
1 - Proactive Networking
Attending networking events or conferences was a standard method for 43% of consultants to find new clients. These events provided valuable opportunities to meet prospects (a.k.a prospective clients or potential clients), showcase their expertise, and establish connections. Networking events resulted in successful client acquisition by facilitating face-to-face interactions, which created lasting impressions.
He runs local meetups and conferences.
Got all clients through personal connections. e.g., He knew the founder from a dinner 8 yrs ago and kept in contact.
Very people person. Always connecting with people she meets apart from the ones she is working with.”
Alternative Tools
I've used Fathom.Video and HeyMarvin for customer research. I love HeyMarvin's ability to let me manually tag parts of any interview. Most other tools, including Fathom, provide a summary, but customer experience research requires deeper engagement—you create value by actively reviewing customer recordings.
This is where HeyMarvin stands out. It allows me to:
Mark key moments in any interview
Create AI-generated notes based on those tags
Combine and label multiple notes for better synthesis
I might use HeyMarvin for a personal project to explore its full capabilities. Apart from ChatGPT, I could use PerplexityAI for the steps in this article too.
Other tools I’ve used for helping me take notes from conversations:
Related: