Phone surveys used to be the gold standard. A trained interviewer, a structured questionnaire, a real conversation. The data quality was excellent. The cost was not.
CATI interviews typically run $30–50 per complete, sometimes higher for specialized audiences. At those rates, phone has been priced out of most research budgets for years, replaced by cheaper online panels that trade conversational depth for scale.
What if you could get the conversation back without the cost?
AI-Powered Voice Surveys on MX8 Labs
Starting today, MX8 Labs supports voice surveys powered by Twilio. The platform calls your respondents, asks your survey questions aloud using an AI voice, listens to their spoken answers, and maps those responses back to your survey structure (all automatically).
The experience is conversational. Rather than reading questions like a script, the AI adapts its delivery to feel like a natural exchange. Respondents aren't pressing buttons on a keypad or using a phone tree. They're talking, and the system understands.
All question types are supported. The only limitation is what you'd expect: media-based content like images and video obviously can't be presented over audio. Everything else (single select, multi-select, open-ends, scales, logic and branching) works the way it does on any other MX8 source.
The Business Case
The economics here are striking. Traditional CATI research requires recruiting, training, and managing live interviewers. You're paying for labor on every single call, productive or not. Supervisors monitor quality. Scheduling is constrained by working hours and interviewer availability.
AI voice surveys eliminate the labor component entirely. There are no interviewers to staff, no shifts to schedule, no per-hour costs accumulating while your team waits for someone to pick up. You pay for telephony and compute, not headcount. For most studies, that translates to a cost per interview that's a fraction of traditional CATI; often 80–90% lower.
The economics are also fundamentally different from text-to-web outreach. With SMS campaigns, you pay for every message you send, whether the respondent engages or not. Send a hundred thousand texts and get a 3% response rate, and you've paid for ninety-seven thousand messages that produced nothing. With voice surveys, costs are concentrated on calls that are actually answered. If nobody picks up, you're not burning through your budget on dead-end outreach. That changes the math significantly, especially for large contact lists with uncertain response rates.
That cost reduction doesn't just save budget. It changes what's possible. Phone-based research becomes viable again for studies that couldn't justify it before: tracking studies, customer satisfaction programs, audience segments that don't respond well to online panels. When the price drops by an order of magnitude, the use cases multiply.
And unlike CATI, AI voice surveys scale instantly. Need to field a thousand calls tonight? There's no staffing constraint. Need to run the same study in three languages? You configure it and go (no multilingual interviewer recruitment required).
What This Means for Hard-to-Reach Audiences
Some of the most valuable research audiences are the ones that don't respond to email invitations or online surveys. Older demographics, rural populations, people without reliable internet access, frontline workers who aren't sitting at a desk; these groups have been systematically underrepresented in online research for years.
They do, however, answer the phone. Voice surveys reopen a channel to populations that the industry has struggled to reach cost-effectively since CATI became prohibitively expensive.
An Honest Assessment: This Is Early
We're excited about voice surveys, and we believe the technology is ready for production use today. But we want to be straightforward: this is the first generation of a capability that will improve significantly over the coming months and years.
AI voice models have reached a level where conversations sound natural and respondents engage comfortably. But they are not yet indistinguishable from a human interviewer in every scenario. Accents, background noise, and unexpected phrasing can occasionally trip up transcription. Complex open-ended probing doesn't yet match what a skilled human interviewer can do.
There's also the call screening question. Apple's Live Voicemail and similar screening features on Android mean that a growing share of unknown calls never ring through to the respondent at all. The phone silently sends the call to voicemail, and the respondent decides after the fact whether to engage. This is a real consideration for response rates, and it affects all outbound phone research (not just AI voice). We're actively exploring ways to work with call screening rather than against it, but it's an evolving challenge for the industry and we want to be upfront that it's not a solved problem yet.
We think the quality is strong enough to field real studies today (and the data supports that). But this is a technology that's going to get materially better as real-time voice LLM models continue to advance. If you're evaluating voice surveys, you're adopting a capability on an upward trajectory, not a finished product.
Understanding Regulatory Requirements
Any conversation about automated voice calls has to address the elephant in the room: is this robocalling?
The short answer is that AI voice surveys are subject to the same rules as any other automated phone outreach. The FCC has confirmed that AI-generated voices fall under the Telephone Consumer Protection Act (TCPA), which means prior express consent is required before placing these calls. That's the same consent framework that governs CATI research conducted with autodialers; it's not a new burden, but it is one you need to take seriously.
In practice, this means voice surveys are best suited for first-party contact lists where you have an existing relationship and documented consent. They're a natural fit for customer research, member surveys, patient follow-ups, and similar use cases where consent is already part of your workflow.
The regulatory environment around AI voice is still evolving. The FCC has proposed additional disclosure requirements specifically for AI-generated calls, and state-level rules vary. We strongly recommend working with your legal team to ensure compliance for your specific use case and jurisdiction. MX8 Labs builds in opt-out handling and stop-message compliance, but the consent obligation starts with you.
Getting Started
Setting up a voice survey is straightforward. In your survey's respondent sources, search for "voice" and select Twilio Voice. Configure your Twilio credentials, choose your preferred AI voice, upload your contact list, and you're ready to test.
Test links offer both phone and online modes (so you can hear the full experience yourself before going live). Just enter your number in E.164 international format (e.g. +1 555 123 4567) and the platform will call you.
For the full setup guide, visit our documentation on setting up a Twilio Voice respondent source.
Reporting works exactly the same as any other source type. Crosstabs, data exports, weighting; it's all there, no additional configuration needed.
The Conversation Is Just Starting
Voice surveys bring something back to research that's been missing: the richness of a spoken conversation, at a price point that makes it accessible again. We think this is a meaningful step forward for the industry, and we're eager to see what teams do with it.
If you'd like to see voice surveys in action, request a demo. We'll walk you through the setup, play back a sample call, and help you evaluate whether voice is the right channel for your next study.