Back to the Future: Will LLMs Make Voice the Future of Surveys Again?

Or why your next survey won’t be a phone call

For years, surveys lived by the phone line. From meticulously recorded in-person interviews to cold calls during dinner, voice was the king of market research. However, the COVID-19 pandemic did what market disruptors couldn’t: it forced surveys almost entirely online. Large Language Models (LLMs) and advances in conversational AI suggest we might bring voice back—but should we? Does voice have a place in the future of digital surveys, or is it an answer to a question we’re not asking?

Let’s examine voice surveys, what they could be, and where they’re likely to go.

When Voice Was King of Surveys

Back in the day, surveys were practically synonymous with human interaction. Phone and in-person surveys captured respondents’ tone, pauses, and hesitations—nuances that added depth and context to raw answers. This voice-first approach worked because it was natural and adaptable. Researchers could follow up on interesting answers, adjust the conversation, and dig deep, using voice to get beyond “yes” or “no.”

Then came the pandemic. Overnight, survey methodologies pivoted to digital formats, bringing the efficiency and reach of online platforms to the forefront. However, something valuable was lost in this transition, particularly in qualitative research, where the voice’s warmth and spontaneity could make a difference.

LLMs and the Promise of Conversational Surveys

Enter LLMs, the newest player in conversational AI. They can hold remarkably fluid conversations, seemingly made for the natural flow we once found in voice interviews. The potential here seems obvious: voice-based surveys powered by AI could allow us to tap back into that old depth without the staffing and cost of phone calls.

But here’s the rub: LLMs are fantastic responders, not questioners. Designed to answer, they lack the “instinct” or training to ask targeted follow-ups, handle unexpected answers, or probe deeper based on real-time feedback. Without these skills, an AI-driven voice survey loses the adaptability that makes traditional voice so valuable. It’s more conversation-ish than conversational, which could leave respondents feeling slightly… talked at.

Is Voice Really the Best Input Form for Surveys?

When it comes to quantitative surveys, voice has its drawbacks. Sure, there’s value in hearing someone’s answers directly, but online forms have become the standard for a reason. They’re simple, efficient, and—importantly—consistent. Operational challenges aside, quantitative surveys thrive on standardized data, something text-based methods excel at.

Voice could theoretically capture subtleties that a text response might miss. Yet, today’s reality is that voice isn’t the primary communication tool for many respondents, especially younger ones. They spend more time in chat apps than on phone calls, and their relationship with voice-based interactions is fundamentally different. In other words, while voice might be powerful, it doesn’t necessarily align with the ways respondents prefer to communicate.

In addition, brains work differently. Some people connect through voice, thriving on intonation and flow, while others process information better visually. A voice-first approach risks alienating respondents who are more comfortable or expressive through text, making surveys less effective for a broad audience.

Voice’s Role in the Future of Quant and Qual Research

Quantitative Surveys: Text-based or online formats will likely remain the go-to for quick, large-scale data collection. The long lists of grids to choose from in quant don’t map well to voice interactions. Even if we could streamline voice responses, it would probably be a longer interview, and we all know that the length of the interview drives cost.

Qualitative Surveys: Voice could shine more brightly in qualitative studies, where capturing tone, emotion, and elaboration adds value. Imagine an LLM capable of asking probing follow-ups, working almost like a skilled moderator. That’s the dream, but we’re not there yet. Even as LLMs improve, the primary cost in qualitative surveys remains participant time, not moderator time. No matter how sophisticated AI becomes, it can’t reduce the time and effort a respondent spends in a voice interaction.

The Verdict: A Tool, Not the Solution

So, is voice the future of surveys? For quant, probably not. Digital forms and text-based interactions seem better suited to the data’s needs and most users' habits. There’s likely a hybrid future where voice becomes commonplace for some respondents, especially those less well-educated, with chat-based forms likely the most common format for most consumers.

However, voice already leads in qualitative research, and assuming LLMs can become active, intelligent questioners instead of passive responders, this will become automated. However, I don’t see this leading to a massive resurgence in qualitative studies as the cost hurdle remains the time and patience of the participant, something AI can’t automate away.

The bottom line: voice has its place, but it’s part of a larger, multimodal toolkit. In the end, effective surveys will adapt to the diverse ways people communicate—not try to fit them all into one voice-first box.