Research Theater vs. Research That Matter

Episode Timestamps

​00:00 – Guest introduction: Divya Kulshreshtha and background

01:13 – How AI is changing day-to-day user research and decision-making

04:50 – Responsible research with AI + “garbage in, garbage out”

​08:15 – Designer guardrails: partnering with researchers and avoiding synthetic users as “insights”

10:09 – Rapid vs deep research: how to decide what you need ​

14:35 – Why it’s hard to “sell” research internally (even with easy user access) ​

17:56 – The “picnic” metaphor for getting stakeholder buy-in ​

21:46 – “ChatGPT said so”: treating AI outputs as hypotheses + being devil’s advocate

​24:07: – Remote research: when it works and what gets lost vs in-person ​

27:23 – AI as the researcher vs AI for analyzing recordings/observations ​

29:55 – Avoiding “research as a checkbox” in an AI-fast world ​

34:59 – Mindset shift: stay curious and keep asking “why” ​

40:23 – Closing

Main Insights From The Conversation

“With AI, obviously you can put a lot of garbage inside the system and you can expect to put out a lot of garbage out of it, right? You can call it an insight, but it may not be on point to the reality of the user”.​

For Divya Kulshreshtha, Principal Designer at Naukri, this garbage-in-garbage-out principle isn’t just a cautionary tale—it’s the foundation of responsible research in an AI-accelerated world. With a decade transitioning from leading pan-India research for Metronic and Samsung to full-stack product design roles at Paytm, Udaan, Capillary, and Freshworks, Divya has seen the entire journey from discovery to business impact.​

His unique vantage point—researcher turned designer turned strategic thinker—offers rare perspective on how AI is transforming every step of the research-to-product pipeline. From using NotebookLM to synthesize interviews, to deploying Gong for sales conversation analysis, to watching stakeholders say “ChatGPT told me so” when defending questionable decisions, Divya navigates the messy reality of democratized AI tools colliding with the sacred craft of understanding users.​

In this conversation, he unpacks the guardrails needed when ingesting interviews into AI systems, why synthetic users build hypotheses but not answers, how researchers have become managers of insights rather than sole generators, the art of “selling the picnic” to get stakeholder buy-in, and why remote research struggles with the human connection that makes truth emerge. He challenges the checkbox mentality that treats research as performative theater and instead advocates for “focused research” where meaning defines value, not speed.​​

His core message? AI is everywhere in the research journey now—from recruiting participants to synthesizing findings to accelerating execution—but the human touch remains essential for assimilation, empathy, and ensuring insights do justice to users’ lived realities. And the most powerful use of AI isn’t getting answers—it’s getting reflective questions that help you think more deeply about what you’re building and why.​​

The Journey: From Research to Full-Stack Design

Divya introduces his transitional career path:​

“I’ve seen the journey from starting from discovery point of view for research, and then trying to convert that research insights into some actionable ideas, and then as a product designer trying to create some sort of artifacts out of it, and then as a product and business thinker as well trying to see if the market can have a value for it or not, right?”.​

“So I see the whole journey end to end. So I definitely see AI has changed every bit of my process basically”.​

How AI Has Changed Every Step of Research

When asked how AI has transformed the role of user research, Divya breaks down the process:​

The Basic Research Journey

“User research starts from problem discovery, I would say. And then the synthesis part—how do you conduct the research itself, you conduct the interactions with the users and stuff like that, and then how you synthesize those insights and then you present them into a certain doc, like a certain presentable format, right?”.​

“I think each of these parts AI has changed”.​​

NotebookLM for Synthesis

“Like in my own experience itself, we are using NotebookLM quite extensively to sort of, once we’re done with the interviews part, I can use NotebookLM to sort of consume some of those inputs and then start adding strategy point of view and all that and try to convert it to actionable ideas, right?”.​​

AI for Recruiting and Scheduling

“I think AI also is kind of changing quite a lot because AI can now reach out to the users on your behalf and try to schedule some of these calls, interviews and stuff like that, do all that part of the journey”.​​

The Human Touch Remains Essential

“So I think the whole part of the journey—let’s think of it like that step-by-step journey—and I think every aspect you will see that AI has played a role in many ways, right? And it has some parts it has made very much productive for us to automate the aspect of scheduling the calls, recruiting the right people and stuff like that”.​

“And certain aspects, still the human is in the control part of it, right? They’re still responsible for the assimilation of those inputs, the insights part of it, right? The human touch is still required because NotebookLM or some of these tools cannot still do that kind of justice to the insight as well, that right?”.​​

Still a Long Way to Go

“So I would say, I mean, there’s still a long way to go. I think it’s starting out in this small way. I think we have to see from these step-by-step journey of a researcher’s life, and you will see that AI is there everywhere. But I think the end-to-end packaging still a human has to do so it becomes consumable. That’s how you think about it, maybe”.​​

Guardrails: Garbage In, Garbage Out

When asked about responsible research with AI, Divya introduces the critical framework:​

“I think that’s a very relevant question today because with AI, obviously you can put a lot of garbage inside the system and you can expect to put out a lot of garbage out of it, right? So you can call it as an insight, but it may not be on point to the reality of the user, right?”.​​

The Fundamental Question

“So I think definitely there is a very relevant question to ask right now: how do we put the guardrails at the input source itself so that whatever we are trying to ask and generate insights upon, it’s doing justice to the insights, right? Justice to the users’ day-to-day life scenarios and stuff like that”.​​

NotebookLM Guardrails in Practice

“So for example, I’ll give a simple my own work-life example, right? So in NotebookLM we are trying to closely put guardrails in terms of who’s conducting interviews and who’s ingesting some of those interviews into the system before we can generate insights”.​​

“So a simple action of putting a process around it, putting guardrails around it, having a responsible person who understands the empathy of what a user would be doing—so a researcher’s empathy will be very relevant here at that point—that guard actually helps itself to actually you don’t put garbage inside the system so that you don’t confuse the outputs and you don’t generate something which like hallucinations more than the real user’s life, right?”.​​

The Synthetic User Trap

Divya addresses one of the hottest topics in research today:​

“For example, synthetic users is a hot topic right now, right? So you don’t want to talk to synthetic users so much. You want an evolved version of your user’s understanding, but you don’t want to go in a complete virtualization of your experience, right? You don’t want some random person to answer random thoughts basically, right?”.​​

The Control Environment

“That’s how I think about it, right? So if you don’t put garbage inside, you don’t get garbage outside. So a researcher’s job has always been—put the right emphasis on the right users, recruiting the right people, conducting the interviews with the right stakeholders in the right manner, right? Empathy, collecting the right way of the observation that is there on the ground”.​​

“Then translate that into synthesis in a control environment where AI is sort of helpful—Dovetail can help you do that, right now, because you have a lot of these interviews getting simulated in Dovetail in a single manner”.​​

Gong for Sales Conversations

“And there’s another tool that we have been using called Gong, where you have all the sales conversations being put together in one single view. And any designer like me or a researcher like me can actually put in a query and understand what is a user talking about from that context, all right?”.​​

The Sanity Check

“So we put the guardrails where you ensure that these sources have their own sanities there. You don’t violate the sanity of those sources. Then you can probably get the right output, and then the AI can be in a way very helpful for you as well”.​​

Researcher as Manager of Insights

When asked if AI helps with sense-making while researchers manage insights, Divya shifts to his designer hat:​

“Would you say that AI’s role there is to help you in sense-making of all the different feedback coming in, and then the researcher’s role is kind of like a manager of the insights and making sure things are—they make sense for anyone who’s trying to consume the insights?”.​

Designer Guardrails

“For example, in my practice, I ensure that whenever I’m using any AI tool to generate insights, I always ask my research partner to actually enable that process for me. I don’t do it on my own. I don’t use synthetic users as such directly to generate insights for me”.​​

“I actually, what I do in terms of my own process as a designer, I would say I will put the guardrails of who am I talking to”.​

Starting from Existing Knowledge

“So for example, in a recent project, I asked my research partner, ‘Can you tell me what we’ve done already in this particular area? What knowledge do we have right now? Let’s start from that. And whatever is missing, can you help me set up a NotebookLM where I can actually use this that you’ve created for yourself? Can I use that as part of my own inquisitions and my own queries basically?'”.​​

Talk to the Right Stakeholders

“So that’s how I would say—you talk to the right stakeholders. That’s the simplest thing you can do in the context. Second is obviously you can plan to do your own research and all that, but that takes an effort”.​

“And I’ll, in terms of—because I’ve worked in so many organizations in the career—so I know it’s not very easy. It’s not very easy to coordinate some of these calls, get those people into the room and get their perspectives in place. It’s such a difficult task to accomplish”.​

The Easiest Starting Point

“So I’d say the easiest way is to talk to your research partner or research thinker in the team that you have, equivalent of that. That could be one starting point. And then if you don’t have it, then build your own practice, build your own approach, and try to do justice to the users by talking to them at least once, recording those conversations, using AI to synthesize those conversations. That is what you can do”.​​

Transcripts, Not Synthetic Conversations

“Like I use a lot of my NotebookLM analysis is actually happening on the transcript of those people, not in terms of how I’m talking to those people. That’s the way we can control it, maybe”.​​

Fast vs. Deep: When to Do What Kind of Research

Divya outlines his framework for deciding research depth:​

“I think see, if I see from my work’s perspective, right, I always—I think you need to understand what you already know about the space. So before you can start executing a plan for research, whether it’s a quick research or a long-term research, whatever, right, deeper research—so you need to understand, like, what do we already know about that space, about that particular user?”.​

Starting Point: Query Your Repository

“As I said, I think the starting point for me always is whatever repository I have for my user knowledge—like for example, Dovetail is there in my system or NotebookLM elements are there for my system—can I start querying that up and then summarize the knowledge first of all? That’s the starting point”.​​

“I think that is one way definitely will give you a quick understanding of, ‘Hey, I know this much only this space, and I’m trying to create a very delta experience for the user, and this is a big question I’m trying to solve. So let me go very fundamental about it. So let me just then do a deeper research or a deep dive in that context.’ That is one easy way to come to that stage of knowledge”.​​

Where You Are in the Journey Matters

“The other scenario that I see is where you are in the journey also matters. Because I’ve seen the whole journey design process part of it, end to end, like the whole product delivery model that we have in India and across design teams in the world as well”.​

Early vs. Late Stage

“You need to know: I am very much early in the space, I know very less about the space, so I need a deep dive, I need explorative research, I need ethnographic studies, I need some of those kind of knowledge to be there to build some perspectives”.​

“If I’m very late in terms of—like for example I’ve already built some concepts, hypotheses are already in place, AI has helped me to analyze a lot of synthetic data to actually be able to hypothesize what is happening to the end user and stuff like that—so maybe I’ve built some ideas, I want to test it out quickly. So can I do this thing unmoderately? Can I test out some of those ideas, right? So that’s a very small study that you can do actually in that context”.​​

The Matrix Approach

“So where you are in the process matters and how much knowledge do you already have in that space matters. I think these two things can actually be a good metrics to think about. I mean, researchers are very good at synthesizing in terms of matrices, right? X and Y can give you a plan in place basically, like where do I see the research going basically, right?”.​

GPT as Sounding Board

“And obviously a lot of times I think AI can actually be like a good sounding board as well. I use, personally for example, GPT as a lot of my sounding board for decisions basically I do, right? I put up my thoughts across to that tool, and then it just gives me a reflective question. I would keep asking not for answers but reflective questions. That’s another thing I do. It’s not giving me solutions but it is giving counter thoughts”.​​

Selling the Picnic: Getting Stakeholder Buy-In

When asked about the difficulty of getting organizational alignment for research, Divya gets refreshingly honest:​

“If I answer this thing very honestly, right—and yes, please—because I’ve been part of a lot of enterprise organizations in the recent times, like with Freshworks and with Naukri as well in the recent times—even if the access is easy, it is still difficult to actually align people to come to that point and we should do user research. It’s not very easy”.​

The Reality Check

“I’m sure anybody who’s listening to the podcast also as a researcher will know that it’s difficult to sell user research in the organization as well. It is very idealistic for me to say that, ‘Yeah, I want to do all the studies and X and Y,’ and all. I had to my academy training so I also had the same feeling. But it is very difficult”.​

The Consultant Bubble

“And interestingly, because in my career when I started my journey, I was a consultant. So I was getting a lot of research projects hands-on to me directly. That was something in my mind also was like an ideal world I have put together that, ‘Okay, it’s very easy to do research or it’s very—everybody’s aligned to do research in the organizations, right?'”.​

“But what I was actually getting projects to work upon are a very handful of companies in the country who were intentionally spending into research. So it’s not very easy to get the alignment of people to actually conduct research in the first place”.​

What Worked at Capillary

“But that being said, how do you do that, right? Something I’ve tried in Capillary because I was the first researcher there. I tended to build research function as a design function as well there, in my small area that I was focusing on basically”.​

“So what we started doing was a lot of immersion of the leadership can actually happen, right? I started engaging a lot of my leaders in terms of, ‘Hey, we as a company are focusing on customers, shoppers. Let us understand what are the different mental models of these shoppers,’ as simple as that”.​

Workshops, Not Documents

“I started doing workshops with them. A lot of these immersion sessions that you do with the leadership can actually bring people together to understand the value of what they’re trying to deliver, right?”.​

“And especially researchers—I was just overhearing one of my friends who was, I mean, sharing his research knowledge to another product team member, and he was like, ‘Yeah, it’s too long for me to even consume.’ And that keyword was there. I knew that, you know what has happened is that that guy is not sold so much”.​

Four Sentences Nobody Needs

“Even if the researcher has put a heart and soul in terms of putting four good sentences of thoughts and all that, that guy doesn’t need that four sentences. He just needs one good idea to sell to the other guy basically. And this is how in organization, if I answer from a very practical point of view, the reality is it’s very difficult to sell”.​

How to Sell It

Divya outlines his approach:​

“Now how do you sell it and how do you bring it together? In my mind, I think if you think of it, I think start immersing your leaders. Start bring them together. Get them to empathize to the whole context and make them aware that if you don’t do this investment, somebody else is doing this investment. That also helps product people understand it from that lens very strongly”.​

The Competitive Angle

“‘Hey, we are a company, UX Army may be a good platform, but UserTesting platform is doing 10x amount of effort in the user research part. They’re doing faster iteration of cycle of product development also. Let us invest in this space and let us disrupt the whole space,’ like that, right?”.​

“So commercial selling to your stakeholders quite helps quite a lot. Once that has happened, then I think it’s easy for a researcher to plan the whole thing, right? Then you can decide when you want to do research, how you want to do research. It can happen actually”.​

The Picnic Metaphor

Divya introduces a memorable analogy:​

“A good metaphor for this one actually, the way I put it to my friends also, is that it’s like a picnic. You want people to go, but you need to sell the idea of a picnic to everybody, right? Unless you sell that idea and then the benefit of it or the kind of experience that we’ll get together, the whole experience—if you’re able to sell your friends a picnic idea or a trip idea or road trip idea—you can sell research to your team members. And that’s the same dynamics works, it works, right?”.​

“So you need to identify what’s the USPs or the value adds for this, for that person. Similarly, answer the same question in your context. That’s it. That’s how it can actually make sense”.​

Remote Research: The Human Connection Problem

When asked about remote versus in-person research, Divya identifies the core challenge:​

“I think access to certain people, certain roles are very difficult to have, then remote research cannot actually play a role. But some places, like for example in my context, I’m talking to a lot of recruiters who are hiring across India for different kind of roles. It’s easy for me to reach them with the channel, right?”.​

COVID’s Legacy

“So that, I think, another way—I think COVID made us all comfortable with this idea of remote research and all that. But personally in my experience transitioning from the in-person research to remote research, I’ve seen what I’ve seen is people don’t relate to you so much when you’re on a video call basically, right?”.​

You’re Just a Graphic

“You are not a human being. You are just a graphic which is moving, right? So obviously that bias is going to come in the answers as well. That’s why I think the value of day-to-day life actually—day-to-day scenarios, day-to-day studies, diary studies—are actually relevant because you want the person to be able to be comfortable in the context and share the real truth. That’s the challenge with remote research: that the real truth may not even come out unless the person is made comfortable, and a lot of things can happen that way, right?”.​

The Moderator Solution

“So I believe there’s a value in this remote research when you actually—for example, maybe having a moderator in the context. For example, a lot of times in my previous experience, how people—what they used to do is they will have a moderator who travels to that particular place, empathizes the people, sensitizes them to the idea and everything, and then they come up with a warm gesture to the video call and then you can talk to them. You can enable those processes. That’s one way I think we can think of unlocking the power of it because the access to the user is really difficult”.​

The Work-from-Office Analogy

“But I would—I have personally felt that it’s always good to be in the context. I mean, why are people asking everybody to come to office now? Work from office is the real trend right now, right? Four, three years back, it was work from home was a trend. Everybody was saying work will completely go virtual, right?”.​

“But now people realize, yeah, no—the collaboration is not happening so much, the human gesture is not there, the connection is not happening, X and Y thoughts, right? So now the whole trend is towards work from office again, which I think is another example to think about this remote research. There’s a value to it, but people realize, okay, the balance has to be made somewhere. That’s how I think about it”.​

AI Moderation: Observation vs. Interaction

When asked about AI conducting moderated sessions, Divya draws an important distinction:​

“I think I’ve seen it from two lenses actually, this question, right? So I’ve seen AI being used to sort of conduct the whole session itself where somebody is, an interview is happening with an AI, right? And a lot of people have used this thing as a product space to actually build a lot of SaaS tools where you can actually conduct interviews virtually and stuff like that, right? That’s one space I see”.​​

The More Relevant Use Case

“The other space around that is—I also think that’s the more relevant space in this context—is you let the people use your product in your—in their own environments. In the old times we used to generate heat maps out of it and used to see recordings, recording dumps we used to get actually, right, of these users using our product”.​

“Now a friend of mine from my team actually has been very smart in terms of building an AI agent which analyzes that and understands where the gaps are, where the person is pausing so much, where is the flow which is taking a lot of time actually, that way, right? So all the basic analysis, it’s already doing and serving it as an insight to us”.​​

Why AI Interviews Won’t Work

“Now that I think is still a relevant use case in my mind. I think the first one I think will not work because it’s difficult for a person to interact with a virtual human being on a camera. But the second level of complexity is that the person itself is not real. That again, you’ll have a lot of possibility of people not saying the real truth. They’re not going deep enough. Everything staying on the surface part of it, everything being very shallow”.​​

Observation Over Interrogation

“And the other side of the story is more relevant and more applicable and a more very profitable use case where the whole hard work of seeing recordings, building insights—all that can actually be automated very easily. That I think still sees a potential”.​​

“So I think that’s where I think AI should go. Unmoderated should be more about observation, day-to-day life studies, observation happening through AI, rather than having AI conduct interviews and sessions. I think that’s not going to be so much fruitful in my understanding so far”.​​

Focused Research, Not Fast Research

When asked about research becoming a performative checkbox, Divya reframes the problem:​

“I mean, that’s a question I think I struggle myself as well, if I be very honest about it, right? I mean, as a researcher trying to design, I know that I’m not doing justice at certain points when I’m actually quickly taking decisions to design products, right?”.​

“But I think I’m sure people who would have gone through the transition would have seen that experience themselves also, that sometimes you don’t have the time or the luxury of the time as well”.​

Not Fast, But Focused

“So what I feel is that fast then becomes—I would say I position more on that—not like fast research, but like focused research. So if you treat your research activity like not a fast activity but a focused activity, then it’s not feeling like a checkbox. Then it’s like a meaningful check, like it’s a meaningful action that you have done”.​

Adding Meaning

“So if you add more meaning to your actions—like in terms of why am I talking to synthetic users, for example, for XYZ decision-making, right, for validating my ideas and stuff like that—if there’s a meaning behind that can actually still be a good way to not make it like a checkbox but make it like a thoughtful way of approaching”.​​

“Anything that you add meaning to it will become important to you, right? And then it’s not like a checkbox. It’s like a meaningful action that you have done. That’s number one”.​

User-First Is More Relevant Than Ever

“Number two, obviously is the part where you actually sensitize yourself with the aspect of research, the value of it and the credibility of end users’ perspective and all that. I think now everybody I’ve seen in organizations across organizations, they want to be user-first. User-centered design was a movement that happened in 2000s and all that, but now everybody realizes that, okay, things are moving so fast that you want the user focus to actually guide you in the right direction”.​

“Because right now 10 different LLMs in the market and 10 different sort of ways you can engage with the user and build some concepts and products and ideas and all—what is differentiating between these products, I’m sure, is the way somebody is looking at the things more thoughtfully. The user aspect comes in very strongly, right?”.​

Sensitize the Leadership

“So now more than ever, the aspect is more relevant. If you add more meaning to the research process itself, it doesn’t become like a checkbox, it becomes meaningful activity. And I would say, I think a good way to is sensitize the leadership with the same aspect again. I circle back to that idea because I have seen that personally happen as a transition, right?”.​

“Because before that, what they thought researchers were doing was just generating insight documents. They do some research in their own silos, and they come back with some well-polished slides of what the insights are and stuff like that. But if you convert that into workshops which are co-creating in nature, co-creation is happening, people start realizing the value of it. It doesn’t become like a checkbox. It becomes like a thoughtful checkbox, right?”.​

Meaningful Workshops

“So that’s how I would put it: add more meaning to it. If even if you’re putting one thought together in one slide, can there be a way to collaboratively come to that thought, for example? That’s another way a lot of people can come together, and then that meaning is added to that thing”.​

“So if all in the organization product development cycle, right, they are sensitive to the idea of users and the users’ real point of views and stuff like that, then it is a meaningful activity. And even if a small action like a workshop happens, it’s a meaningful action. So it’s not even—it’s a meaningful workshop and a meaningful sort of a way of collaborating and coming together basically”.​

The Mindset Shift: Keep Asking Why

For his closing advice, Divya returns to fundamentals:​

“I mean, coming from my—I think it will be like a very—I don’t want to put any big words out there, right? Because I’m still in my journey in terms of evolving my thinking from what is—capturing the user perspective, insights, value of insight into strategies into product design. So I think I’m still in my journey, so I will not say any big words in terms of what the mindset should be for the industry”.​

The Three-Word Answer

“But what do I see as my own takeaway in this thing so far to kind of guide my own work? I would say I always be curious and ask the why questions, right? The three-level why. If you can ask the fundamental—there’s so many ways people—first principle, somebody calls it, somebody calls it five whys and how-might-we statements and stuff like that, right? So there are many keywords around it”.​

The Common Theme

“But my only, like a theme that I’m always getting from that, is keep questioning, keep asking. Let’s say, for example, you were thinking about, you know, asking critiquing to the ChatGPT itself, right? Critique its own choices. That is another thing, result of a why question only, right?”.​​

“Why can I not answer and ask a question to GPT back? Why can I not go to the core of how it is putting together an answer, for example, right?”.​

NID’s Academic Training

“So I would say I think a lot of the thought process shift has to happen towards whys. And that I think academic training from NID actually has helped me to do that, by the way, right? I mean, the fundamental way that first day of the course, there’s nothing like a course. Everything is like open. So you do what you want to do, and you ask all the whys that you want to ask, and then you put together an idea”.​

“I think that has been part of my rigor. Maybe that has—why I’m still talking about that—because that changes the way, I mean, that changes the way in which you think about any product, any research statement, any brief, any strategy document that you come across, any research output that you come across. We always question why. Why is somebody talking about this in this way? Why could we not be in this way, right? Like that kind of question, right?”.​

Designers Thrive On It, Then Forget

“I mean, designers thrive on that, but then we kind of forget that sometimes, you know, we consider it to be like basics. Moreover, with LLMs giving you the answer instantly—we ask why, but it’s not always a meaningful why”.​

Fallen Into the Trap

“So I—I like—I fall in this trap actually. Yeah, yeah. I myself have fallen in this trap. Like I mean, if I be very honest, right? I mean, I’ve been trying to churn out a lot of my ideas and a lot of my products also using assistance from AI quite a lot”.​​

Prompt Laddering

“But what I’m trying to do recently in my current work is keep thinking of a thoughtful question around that, like keep taking a pause in that. So like one of the things that I was coming across recently in prompt engineering is the concept, right? What is that actually is—just to break it down into pieces and then start building a ladder around it, right? And then the whole output will be much more logical than what you’re actually doing right now”.​

Build the Building Blocks

“So that’s how I think about it. So I have fallen in this trap myself, the love of seeing something come alive so fast. But you can use that as a reference as a—okay, this is what it is, somebody’s recommending. It’s hypothesis. Can I try something from 10 different questions starting from the 10 different why questions that I think about, right?”.​

Daily Emotional Journaling

“Keep—I think practicing that muscle, I think that is the best way to do it. Like for example, every day when we wake up, it’s very easy for us to feel we are feeling happy, we are feeling sad sometimes, right?”.​

“In my—I remember one of the exercises that I used to do in NID was to every day when I used to wake up, I used to write how my emotions are and visualizing my emotions. Now that became like a half-an-hour ritual of thoughtfully questioning why. And you know what is happening to me?”.​

“And then over one month or two months and six months of span, I could see a trend emerging. And that is where the aha moment happens. So you also can actually keep the AI output as a reference but try to build it into still the building blocks of it, and then you can see the value of it, I would say”.​

Key Takeaways: Guardrails, Picnics, and Reflective Questions

Divya’s decade transitioning from pan-India researcher to Principal Designer at Naukri offers crucial lessons for navigating AI’s impact:​

1. Garbage in, garbage out is the foundational principle​

Guardrails at the input source—who conducts interviews, who ingests them into NotebookLM, ensuring empathy and sanity—prevent hallucinations masquerading as insights.​

2. Synthetic users build hypotheses, not answers​

You want an evolved understanding, not complete virtualization; synthetic users are hypothesis generators like internet research used to be.​​

3. Researchers have become managers of insights​

AI handles sense-making across feedback sources; researchers ensure outputs do justice to lived realities and are consumable for decision-makers.​​

4. Always start with existing knowledge​

Query your Dovetail, NotebookLM, and Gong repositories first; understand what you already know before deciding on quick validation versus deep ethnographic studies.​​

5. Selling research is selling a picnic​

You need stakeholder buy-in for the experience, not just the deliverable; immerse leadership, do workshops, show competitive threats, make it commercial.​

6. Four sentences nobody needs​

That product person doesn’t need your heart-and-soul four sentences—they need one good idea to sell to the next person; co-creation workshops beat polished decks.​

7. Remote research struggles with the human connection​

On video calls you’re a moving graphic, not a human; real truth emerges from comfort, which requires context or at minimum in-person moderators who sensitize participants.​

8. AI moderation works for observation, not interrogation​

Heat map analysis and behavioral pattern detection add value; AI conducting interviews creates shallow, surface-level responses because the person itself isn’t real.​

9. Focused research, not fast research​

Adding meaning to actions prevents checkbox theater; if there’s purpose behind talking to synthetic users for specific validation, it’s thoughtful, not performative.​

10. User-first is more relevant than ever​

With 10 LLMs and endless engagement methods, thoughtful user understanding differentiates products; sensitizing organizations to this creates meaningful research culture.​

11. Use AI for reflection, not just answers​

GPT as sounding board providing counter-thoughts and reflective questions beats using it as answer machine; critique the model, ask why it recommends things.​​

12. Keep asking why with prompt laddering​

Break concepts into pieces, build ladders of logic, practice the questioning muscle; NID’s training of doing what you want and asking all the whys changes how you think.​

Final Thoughts: The Half-Hour Ritual

Divya’s confession about falling into the trap of loving how fast AI makes ideas come alive reveals the central tension of this moment.​

We can churn out concepts, validate hypotheses with synthetic users, and accelerate from idea to shipped product faster than ever. NotebookLM synthesizes interviews, Gong aggregates sales conversations, AI agents analyze behavioral recordings to surface pausing patterns and flow bottlenecks. The tools democratize research access across designers, product managers, and marketers who now conduct their own studies.​

But speed without guardrails produces garbage. Virtualized experiences.

Thank you for reading!

If Divya’s insights on garbage-in-garbage-out guardrails, selling the picnic to stakeholders, and using AI for reflective questions instead of instant answers resonated with you, share this article with researchers, designers, and product teams navigating the messy collision of AI tools and user understanding.

Have questions about NotebookLM synthesis, building research functions from scratch, or preventing performative checkbox research? Connect with us at hi@uxarmy.com.

Special thanks to Divya Kulshreshtha for sharing hard-won lessons from a decade transitioning from pan-India research for Metronic and Samsung through full-stack design roles at Paytm, Udaan, Capillary, and Freshworks to his current work as Principal Designer at Naukri.

And thank you to all of you for being part of the User Insights community.

⚡ This podcast is brought to you by UXArmy, an all-in-one UX research tool.