Summary: AI is reshaping employer brand, bringing more speed, scale, and accessibility to workforce insight work. But the quality of that work still depends on human judgment: knowing which questions to ask, how to interpret what people actually mean, and how to turn data into stories that move clients forward.

Much like everything else, AI is changing how employer brand research gets done. To find out what’s new and what to watch out for, our Senior Director of Client Strategy, Will Jeffreys, sat down with our resident research whizz, Ed Emery, to talk through how AI is reshaping EVP and employer brand insight work, and where the guardrails still need to be human.

What does a researcher do?

Will: A researcher wears a lot of different hats. Can you walk me through what those actually are for you at Symphony Talent?

Ed: The first thing is understanding the client’s objectives, because you can’t really start a research program or design a methodology without that. The brief is critical. I need to know what we’re trying to do so I can think about how we’ll gather the information to solve the issue or address the problem. Those two things — understanding the client and understanding their needs — change everything that follows. If you said to me, “I want to uncover the challenges women face in engineering,” that changes the tone of my entire discussion guide and analysis compared to “What are the strengths of working in engineering as a woman?” Two very different briefs. Two very different approaches.

From there, you design the methodology. Are we looking for scale, or depth? Do we need a survey? Do we need rigor? Do the stakeholders need numbers? And then, do we need additional qualitative research to understand the segments we’ve identified?

Then you ask the questions, but it’s never just a list. Interviewing is a genuine skill. I spend a lot of time building rapport, especially in qualitative research. People are more open when they’re at ease. So I try to make people laugh, share my own stories and get a connection. 

Listening is just as important as asking. If someone says something that doesn’t quite align with what you asked, you have to go after it. Ask them to define their terms. In sectors like pharma or science, there’s jargon and acronyms I’ve never encountered before. Being curious, interrogating, following the thread — that’s what separates good interviewing from bad.

Will: Reacting and knowing where to delve deeper is so important. What would the next step be?

Ed: After the data’s been collected, there’s a step that doesn’t get talked about enough: protecting anonymity and securing the data. That, for me, is non-negotiable. The participant is the most important person in the research journey. If a client insists they want the transcripts, the answer is no. Firm no. You’ve made a promise, and you keep it.

Then comes analysis — the most time-consuming part, especially with qualitative data. You’re taking a vast lake of information and teasing out the themes. We call it coding: categorizing what people have said into different topic areas. Is this about why people join a company? Why do they leave? What challenges do they face? Once you’ve grouped the themes, you can look across different sets of data and see what’s really coming through.

And finally, you write the report, which is where I’m having the most fun at the moment. It’s storytelling. You’re turning all of this data into a coherent narrative that the client can actually use. What are we seeing? What does it mean? And how do we make it as engaging as possible?

What’s new? What’s working?

Will: It’s 2026, and there’s a new AI product seemingly every week. What are the changes you’re actually seeing on the ground?

Ed: The big one is speed. Research is typically quite manual. Collecting data — whether through surveys, focus groups, or depth interviews — requires significant human input and careful thought. There’s also a huge amount of stakeholder engagement needed just to get things approved. That’s a laborious process before you’ve even got to the analysis.

The analysis itself is the most time-intensive part of all. An hour’s worth of focus group data takes at least two hours to analyze manually, line by line. In a good organization, you’d get a second pair of eyes on it, which roughly doubles that again.

What I’m seeing now is real momentum around AI tools that can dramatically accelerate all of that. I recently shared with the team a piece of research that was completed end-to-end in 35 minutes. The AI designed the discussion guide based on previous data, sourced participants, led live conversations, and then coded and themed everything in under 10 minutes. It could have scaled to 500 people. That speed is genuinely impressive.

Another tool I’m excited about can analyze responses at scale. At the moment, when we gather open comments at, say, 600 responses or more, we have to analyze them manually. There’s nothing that themes them well without serious, structured coding beforehand. But if you can use a large language model to theme those responses at the click of a button — and do it across multiple languages simultaneously, without a separate translation stage — that’s a step change.

Will: AI is really helpful when it’s not necessarily doing the thinking, but more helping you to speed up the thinking process of our researchers. Are there any other instances where it can do that?

Ed: I’m also excited about what AI enables with video and audio data. When people provide video responses rather than written comments, you’re getting tone of voice, gestures, energy and passion. That’s a whole different layer of sentiment you’d never pick up in a written survey answer. AI is making it much easier to transcribe and analyze that kind of data, opening new possibilities for how we collect feedback.

And at the reporting stage, I’m already using AI as a thinking partner. When I’ve got a complex insight in my head, and I’m not quite sure how to articulate it, I’ll use AI to help me find the right framing. It recently helped me land on the concept of “conditional credibility” (The act of individuals, often women or those with atypical profiles, having to constantly prove their expertise, while others are believed by default) — a term I would never have arrived at on my own — that perfectly captured what women were telling me about their experiences in the workplace. It’s not AI thinking. It’s AI helping me think better.

Why use AI in research now?

Will: If I’m a client and I’m thinking about whether to invest in research, what’s the AI-driven case for doing it now versus a few years ago?

ED: Speed is the headline benefit, and speed has a very real commercial consequence: lower cost. Research has historically been expensive, largely because of how much human time it requires. If AI can compress that, particularly during the analysis phase, the cost to clients drops significantly.

But speed isn’t the only thing. Scale becomes possible in ways it simply wasn’t before. Traditionally, if you wanted to speak in depth with 500 people, the analysis alone would take weeks. AI makes that feasible. So you’re not just getting faster research — you’re potentially getting richer research, because you can include more voices.

Multilingual capability is another one. Being able to analyze open comments across languages in one go, without a separate translation stage, removes a practical barrier that has often meant non-English-speaking audiences are underrepresented or analyzed less rigorously than others.

What to watch out for

Will: All of this sounds great up to a point. But, as someone who’s been a researcher for over 10 years, where do you find yourself going, “okay, but…”?

ED: AI without guardrails is really bad. The burden of responsibility lies with the researcher to decide which tool is appropriate, how to use it, and, crucially, when to stop using AI and rely on their own experience and judgment. That’s a burden I’m happy to take on. But I worry that not everyone will do this, especially as we get more comfortable with AI.

The quality trade-off is real, and it starts at the very beginning. Designing a good discussion guide requires you to understand the client — their objectives, their history, the context that comes from dozens of conversations. AI doesn’t have any of that. You’d need to feed it an extraordinarily detailed prompt to get even 70% of the way to what a good researcher would produce naturally. And if the foundation is weak, everything built on it is compromised. 

The same applies to analysis. Qualitative data can be interpreted in so many different ways depending on who’s looking at it. You cannot just run it through AI and send the client what’s been generated. It needs human critical thinking: why does this contradict itself? Does this theme really hold up? What’s actually going on here?

Will: Do you have an example of the kind of knowledge a researcher would have that AI will not?

Ed: Language nuance is a good example of where AI falls short. We had a quote from a Hungarian survey that essentially said: “A new job would be great for new money.” Some AI tools will run wild with that — picking up on “new job” and inferring the person is motivated purely by salary, augmenting what that single quote actually means. Others might try to identify the general sentiment more broadly. But you can’t just rely on the output. That quote might carry a nuance in Hungarian that doesn’t translate cleanly. You don’t know what that person was actually feeling when they wrote it. That’s where human judgment is irreplaceable.

Then there’s synthetic data — and this is something that really worries me when I see it being misused. I saw a platform recently claiming it could “boost underrepresented audiences” by augmenting real responses with synthetic ones. That is not giving underrepresented voices more weight. If you have eight responses from a particular group, all of whom may have their own localized skew, and you use synthetic data to inflate that to fifteen, you’re just amplifying the skew. You cannot assume one LGBT response becomes five LGBT responses. You cannot synthesize lived experience.

Will: While AI can help us be more inclusive with multilingual capabilities, how would you describe the issues with its actual inclusivity?

Ed: My broader worry is that humans — who are naturally drawn to shortcuts and convenience — will start accepting AI-generated themes and outputs without questioning them. If it’s fast and efficient, people like it. So there’s a real risk to quality. If you sacrifice quality for speed, you end up with homogenized, generic output that tells you nothing interesting. And in the work we do — where the goal is to find the nuances, the unexpected things, the individual stories that unlock a brief — that’s precisely the opposite of what we need.

You can say a company is a collaborative place to work, and that’s probably true. But is it interesting? The interesting question is: why is it collaborative? How do people experience it? How do they talk about feedback, connection and working across time zones? That’s where the real story lives. And that’s why I’m actually excited about AI enabling us to do more qualitative research at greater scale and speed. 

If we can speak to more people, represent more views and analyze them faster, we can uncover more interesting stories. Then our clients get better work.

Want to learn more from Inside the Studio? Check out “Turning Experiential Moments Into Memories.”