Voxpopme Virtual Session: How to Prompt Your Way to Qualitative Insight

How to Prompt Your Way to Qualitative Insight 

Editor's note: This article is an automated speech-to-text transcription, edited lightly for clarity.     

Voxpopme was one of eight sessions a part of the Quirk’s Virtual Sessions – AI and Innovation series on January 30, 2025. The company’s session gave tips on promoting AI to enhance qualitative research findings. 

Betsy Shaak, head of product at Voxpopme, explained how to write prompts for AI, tailoring the prompts to your research objectives and gave real life examples.

Session transcript

Joe Rydholm 

Hi everybody and welcome to our presentation, “How to Prompt Your Way to Qualitative Insight.”  

I’m Quirk’s Editor Joe Rydholm – thanks for joining us today. Just a quick reminder that you can use the chat tab if you’d like to interact with other attendees during today’s discussion. And you can use the Q&A tab to submit questions to the presenters and we will get to as many as we have time for during the Q&A portion. 

Our session today is presented by Voxpopme. Enjoy!

Betsy Shaak

Hey everybody. Thank you for joining. We're going to get started here in a second. 

Welcome to the Quirk’s AI and Innovation series. I'm so excited to be the first presenter up. This is such an honor. We're going to get started here in a second, but just so you know, there are sections of this that are prerecorded because of some time differences. We're here if you have questions, you can ask them in the chat, of course, and we'll get back to you. Just wanted to be honest with that as we go along.

All right, so let's get started.

Hello, my name is Betsy. I'm the head of product at Voxpopme. 

For those of you who don't know Voxpopme, we are an AI powered qualitative research platform. I spend a lot of my time learning about AI and we use AI to help insight teams scale conversations with consumers and add a layer of depth and human stories to their research. 

I spend a lot of time in AI platforms. I'm developing an AI platform myself. So, definitely feel comfortable talking to you guys a little bit about prompting; prompting best practices, how prompts could look versus what you need to get started, where you can use prompting within your research cycle.

Let’s get to it.

Before we jump right into AI for research. The concept of AI and much of the underlying tech has been around for decades.  

In November 2022 when OpenAI rolled out ChatGPT total game changer and the velocity of the innovation has just skyrocketed. We all know this, we all see AI in our lives. 

A lot of people have AI mandates in their company. If you have a mandate that you have to use AI in your research, I'd love to hear about that in the chat. But you guys know this, it's part of our lives now.  

We need to start learning how to utilize it but there are a lot of schools out there. There's a lot of options and prompting best practices can actually really help you. 

So, qualitative research has never been more accessible, which has led to unprecedented scale. I have seen a lot of requests for increased ends of research. So, it used to be like, 'oh, we were going to go to 10 people and now it might be 50 or 60.’ 

‘Oh, we're going to do five IDIs. Okay, well now it might be five IDIs plus five AI moderated IDIs, and then we also have some asynchronous IDIs that we want to do.’ It's just a lot of scaling that's happening in the background, which is really exciting.

This is totally a blessing and totally a curse because tons of tools are giving us new and creative ways to connect with consumers, but you also have tons of tools giving you new and creative ways to connect with your consumers.

There are a lot of tools, a lot of information out there. It just feels like a little bit of a barrier to human connection, and that's what we wanted to solve at Voxpopme, keeping the human part of this, the forefront of this AI journey. 

We all know that since the launch of ChatGPT in November of 2022, we've seen AI integrated into a bunch of areas of our lives, including research. Now within less than minutes, AI can summarize, segment and find patterns. It's really great at understanding language, which is what qual is all about.

There's so many exciting applications, but utilizing that magic is not as easy if you don't have your wand. And in our case, good prompts are your wand.

Good news is that learning to talk to an AI is not so different from the skills that you already have as a market researcher. Everybody here has probably been on an IDI and had a respondent who's just a little bit closed lipped or has to ask the right questions and probe. These are all skills that are fantastic for interacting with large language models.

Why should you care? 

When you learn how to appropriately prompt, you become more efficient, more accurate and more actionable. So, you might get a job done faster, you might get a job done more accurately, like with less bias, for example. And you can go from idea to output in a much quicker cycle. 

Let's just master the basics. Some intro level tips. 

Number one: clear and concise language. 

One of my favorite stories that I've heard from one of our contacts with the team from Microsoft, is she would treat AI tools like they were an intern starting at her job. 

So, if you just imagine yourself, you have an intern fresh out of college, maybe you're still in college and you want to give them a project. Well, you have to give them more context. You have to tell them what to do, a little bit more than you would somebody else. And you have to maybe speak in a way that is clearer and more concise.  

‘Hey, based off of this data, I want you to look at this. Look at this, provide me with this and then help me get to this point.’ 

That's the same way that you need to talk to an AI. 

If you want to, you can also ask any large language model to act like a prompt engineer and say, ‘hey, I'm trying to do this. Write me a prompt that helps me do this very efficiently.’ That's one of my favorites because it skips some of the steps. 

One of my new favorites right now is I have the voice to talk with ChatGPT on my phone, and I use it to act like a prompts engineer. I'll just talk to it like a human being, and it will respond back to me and say, ‘okay, based off the conversation we had, create me a prompt that helps me do this with my dataset.” 

Then of course, there are a lot of complexities that you can add in here. You could ask for role, task, requirement, instructions, all of these different pieces, but really it doesn't have to be all that complicated.  

Here's an example of a very complicated prompt. This was one that our co-presenter at Quirk’s last year helped me write. It has all of these pieces, role, task, requirements and when you do all of these pieces, that's fantastic and I have the instructions at the end, but it does not need to be that complicated. It can quite literally start with just format and tone. 

‘Present me,’ ‘I'm trying to do this,’ or ‘present this data to me like that.’ ‘I'm trying to present this to this group, this stakeholder, the CMO.’ ‘Please keep a professional use marketing language.’ 

You can really dial this back. It doesn't need to be so complicated. And quite frankly, when we talk to people who are utilizing AI tools and we go in and we give these AI workshops all the time, most of the time people don't really grasp prompting until they've done it a little bit and you kind of learn as you go, right? 

It's called conversational AI because you're having a conversation back and forth. It doesn't have to be a provide everything all at once and get everything all at once. You can ask it to tweak things, you can ask it to reformat things.

I almost always find myself asking, present this in bullet points because I love bullet points. You can ask it to say it in a different way. All of these things are things that you can try. It doesn't have to be so complicated. 

You can actually really simplify prompting by just taking a step back and thinking, ‘what am I actually trying to do here?’ 

It's very easy to get lost in segments and quotas and all of these things, but in the end, what are you trying to get your stakeholder and try that as a prompt.

Where can you use prompts to go faster? What stages of their research journey are the best market research teams prompting? 

We can kind of push this into some buckets. We have the capture phase, analyze phase and sharing phase. 

So, just as an example, capturing might be creating your project brief. I definitely have been seeing more teams utilizing prompting for that. It is really helpful for that project brief if you have a template because you can put that template into a large language model and then even speak to it and say, ‘hey, this is what we're trying to do.’ Give it all the context it needs, and it will fill out that template for you. 

You could even ask an LLM or a purpose-built AI like Voxpopme to do something like write a discussion guide. ‘I want to talk to 20 millennials about their experience buying avocados,’ just as an example. You can absolutely create a very in-depth discussion guide based off time restrictions as well. 

Then, of course, there are a bunch of things you can do like ‘help write questions in screeners.’ 

One of my favorite use cases right now is I have a question, what grocery stores have you shopped at in the last year? Well, I do not know all the names of grocery stores in every region in the United States, so I want to generate a list. 

Well, that's something that the AI tools can do quite well. Whether that's something that's built for market researchers or just a generic LLM. Both of those things can grab that list for you pretty well.

That's the capture side of it. 

Also, I should mention on the capture side, of course, now you can do AI moderation. 

AI moderation is a little bit of a scary topic. It doesn't perform as well as human moderators by any means. But we're seeing clients who utilize Voxpopme’s AI moderation as going between an unmoderated video survey and a full IDI. 

AI moderation is really great for adding additional complexity and additional insight into a project. It definitely helps you get a little bit more depth. That's on the capture side. 

On the analog side, clear prompting is obviously going to be super important for qualitative data sets, minimizing that ambiguity, but reducing the risk of bias. 

If I have done 10 IDIs, for example, I've talked to consumers for 10 hours. I'll go back to my avocado toast example about their experience and why they love avocado toast. I might have just remembered something from two and then the second interview and then the ninth interview and the 10th interview. 

But sometimes trends are really sparsed out. So, for example, if you have an insight that pops up in one, four and eight, you might not catch that, but AI is going to do a really good job of leveling up those insights and showing you a trend. 

Obviously clear crafted prompts, optimize those algorithms in the background, but also it accelerates the pace of research, right? Because you, if you can prompt what you need and you can explain what you need to a large language model in a really effective way, and again, this happens a lot by trial and error, but there's also best practices. You can go much, much faster and you can scale your research.

So, what you used to have to do for five IDIs, you can now do in the same amount of time for 10 or 15. And as these qualitative data sets get larger and larger, we're seeing more usage of those AI tool sets.

Lastly, for the share side, within research, your insights are really only as good as your ability to share them out to your organization. You can do the most wonderful brief and the most fantastic research, but if it doesn't get presented to anybody, it kind of just sighs in the corner, and we don't want to do that. We want to share it out to our organization and help get stakeholder buy-in. 

Some of the ways that I've seen some market research teams utilize AI and these prompting habits is AI generated reports.

Reports with a lens can be really, really helpful because as a researcher, we definitely think in one vein. Just as an example, if you're a market researcher and you've done all of this research on our avocado toast example, and then you are supposed to hand that off to a marketing team, obviously your marketing team is going to talk differently. They're going to have different priorities. 

These are all things that the AI can help you reformat in a marketing report that has that lens on it for that team. So, the report lands a little bit better when framed in a way that those people appreciate. 

Also, downloadable PowerPoints for sharing. Then creating show reels or a highlight reel of all the things that people said in your study can be accelerated by some good prompting. 

So, not only could you ask a chat, something like pull me all examples of people talking about putting honey on avocado toast and just quickly create a generated show reel based off of that.

Also, it's really at good pulling unexpected findings, which is really interesting. So, you say, ‘this is what I expected to come out of this. Find me things that I didn't expect and make me a reel of that.’ I think that that is a good sweetener on top of a piece of research. 

Then lastly, on that share piece, I'll also mention if you have a qualitative data set and you can prompt it effectively or you have it available to prompt, you can look at things that maybe somebody else ran or you ran in the past and get pieces of the questions that you're trying to answer. 

An example of that, we had a client, they had a yearly deck that they had to fill out and everything was based off of one pillar of their business. 

Okay, well, why don't you go into your data set that you have now and effectively prompt, ‘okay, I'm trying to do this. I have this document, by the way, here is the document. I want to find examples of consumers talking about this thing that relates to this pillar.’ That's a great prompt and it is really not that complex.

So, discussion guides, screener questions, survey open-ends, question best practices, screener answers, writing briefs. There's a lot of ways that you can utilize prompting within that create step of your research.  

Then of course, in the chat and the analysis side, you can definitely use it to write reports. You can pull out themes and patterns. You can count the number of mentions of a certain topic or topics similar to that. 

So, if you have your avocado toast example, again, you might have avocado toast, but then you might also have jelly on toast. Those two things are relatively similar, but you can also count dimensions of both of those. 

I definitely can make recommendations. So, based off this consumer feedback, you might want to think about changing the lid to this package because a lot of people mentioned that they were frustrated with the lid, and they had to go and ask somebody else to open it for them.

Then also identify outliers. 

So, that's those unexpected findings, key takeaways and of course that repository of all of that information that you can query. 

And then again, just to review. On the deliverable side, you can see here creating a show reel but also replacing a clip with a similar clip. Maybe somebody has something going on in the background or they have audio issues, and you want to replace it. 

You can also add narrative. So, add a slide explaining what's happening. Written reports, and then also downloadable slides.

So, what is the future, right? Because this all sounds like a lot, but a lot of teams that we're talking to when we're doing AI workshops or things like this, they're not necessarily utilizing all these things in the different phases of research.

I think honestly, the future is probably going to look a lot like, AI in various steps of the research process and an increase in things like AI moderation and other AI tools. I think we're going to see an explosion of asynchronous IDIs.

IDIs are fantastic, and I do think that there's no way around them really, but you could scale them.

Just for an example, I'm a researcher based in the United States. I really want to talk to somebody based in Germany. Just trying to get that scheduled is a pain in the butt. So, maybe you do one of those and then you have your AI moderator learn from that one and do four or five of them. 

I think that that's kind of where we're going and that the human element of that is still going to be very integral to it, but the AI is definitely going to add additional depth to it.

Second one is the rise of those qualitative repositories are going to be so important.

Why rerun data? 

I really hope that in the next five years or so, there is a really big explosion in the reduction of that research waste, that stuff that just sits in the corner.

Lastly, turn data into more deliverables. I think we're just going to see an explosion of ways to share research. If anybody here utilizes tools like Notebook LM, there's things like that that can create a podcast based off your research.

Also, I think we're going to be able to bring the consumer into meetings a lot more easily. There's definitely some work that's going on in market research that just makes sharing the customer voice so much easier than it was before. It is definitely easier to talk to consumers than it was even two or three years ago.

And not only that but consumers at scale. You or I can launch a project, you go to a hundred people and ask them about grocery buying habits and get a response, in what, 24 hours?

This actually isn't my slide. This is a slide from Kelly at Insight Farm who presented with me recently, but she had some really great ways that she learned about prompting as somebody who wasn't spending all of her time building an AI qualitative platform. So, these were some of her recommendations for really in-depth stuff.

Honestly, I think the average researcher just going into these data sets and following those best practices that we talked about but also being really explanatory to the large language models, it's going to help a lot. But if you're really into this idea of, ‘I really want to learn about how to build the most intricate prompts,’ or maybe you want to learn a little bit about how large language models work. These are some great opportunities and options for you to learn more. 

So, key takeaways, why does it matter?

Qualitative research is more scalable now than ever. 

Effective prompting is the difference between a researcher who takes five hours versus a researcher that takes one hour to get insights. You can fast track nearly every stage of research.

There are definitely parts of this that need to be human led and need to be human reviewed. Don't copy and paste something, just don't do it. Don't send something to your boss without any background. 

Prompting is a learnable skill. You have the foundations now, honestly.  

You can take courses and dive deeper into this, but really the best way to do this is to take a qualitative dataset and play with it. Ideally, it's something that you conducted yourself and you can develop trust with that dataset. 

But if you want to learn how to enhance AI in your research, we do have these AI workshops. So, think of this presentation but made especially for your team and utilizing your team's data. They're honestly incredibly helpful for every phase of the research, and we love doing them. We learn a lot from them. Everybody leaves out with a new understanding of what it's like to use AI as a market researcher. 

If you want to, you can scan that QR code. You can also find me on LinkedIn, and we'd love to connect.

All right. Thank you, guys, so much for the time today. If you have any questions, I'm sure we'll already be in the chat going as we go through. But yeah, we're here for you. 

Thank you so much!