Voice AI Agents Beat Chat 32% on Long Calls

Voice AI agents resolve long customer conversations more often than text chat agents do, ElevenLabs said in research published Monday, Nov. 2. In conversations longer than two minutes, success rates on voice ran 10.9 percentage points above chat, a relative gain of 31.7%, according to the company's analysis of hundreds of agents running in production.

Key takeaways

• Chat wins short exchanges: under one minute, chat agents led voice by about 9 points. Between one and two minutes the two were roughly even.
• Voice pulls ahead as conversations get longer, with a 31.7% relative lift in success past the two-minute mark.
• Speed matters early: a 16% cut in average end-to-end latency was associated with a 26% rise in success, and calls ending within 15 seconds fell from 21.1% to 3.8%.

How ElevenLabs compared voice and chat agents

The study, written by ElevenLabs' Jake Tennant and Alanna Brown, drew on hundreds of production agents that served customers over both voice (phone and WhatsApp) and text chat. To qualify, an agent had to handle at least 50 voice and 50 chat conversations between May and July 2026.

The researchers compared each agent's voice results with its own chat results, so differences in volume or use case would not skew the outcome. Each business defined "success" for its own deployment, and conversation length stood in for complexity.

Why long AI voice calls resolve faster

Among successful conversations longer than two minutes, voice finished in an average of 4.6 minutes, against 7 minutes for chat. Of 90 agents with enough such conversations on both channels, 78% were faster on voice, and nearly two-thirds of agents did meaningfully better on voice in long conversations.

Part of the explanation is mechanical. People speak roughly three times as fast as they type, the company noted. That figure is consistent with a 2016 study by Stanford University, the University of Washington and Baidu, which found speech input on a smartphone about three times faster than typing, as NPR reported at the time.

"When a problem is complex or stressful, our research suggests that people want to talk it through, not type it out," the authors wrote.

Latency: the first 15 seconds

ElevenLabs also looked at 315 agent conversations from deployments that switched to a faster large language model. A 16% drop in average end-to-end latency coincided with a 26% lift in success rate, and the share of calls ending within 15 seconds fell from 21.1% to 3.8%.

"If an agent is slow to respond, callers are likely to assume the line is dead or the system is broken, and hang up," the post said. The figures show a correlation, not proof that latency alone drove the gain.

The user study and its limits

In a separate test, 98 participants rebooked a simulated cancelled flight with both a voice agent and a chat agent, in random order. Afterward, 85% said they preferred voice, citing faster resolution and empathy.

The research has clear caveats. ElevenLabs did not disclose the total number of conversations analyzed. Success definitions varied by business, and the authors concede that "duration is an imperfect measure of complexity, especially for chat, where customers may pause between messages." The preference result comes from one simulated scenario and is self-reported. ElevenLabs also sells the product under study, its ElevenAgents platform.

Context: the voice agent market

The findings support a business ElevenLabs is leaning on. When the company was valued at $22 billion in a September employee share sale, it said enterprise customers supplied 55% of revenue and named ElevenAgents its fastest-growing product. Competition is rising too, as larger platforms ship cheaper speech models, including Google's new voice models that talk while they think.

What's next

The post argues effective voice agents must track context, connect to business systems so they can act, such as rebooking a flight or issuing a refund, and respond quickly. For support teams, the data points to a split: chat for quick lookups, voice for messy problems. Independent replication with a disclosed dataset would strengthen that case.

FAQ

Are AI voice agents better than chatbots? Not across the board. In ElevenLabs' data, chat led by about 9 points in conversations under one minute, while voice led by 10.9 points, or 31.7% in relative terms, in conversations over two minutes.

How much faster do voice agents resolve problems? For successful conversations longer than two minutes, voice averaged 4.6 minutes versus 7 minutes for chat. Of 90 comparable agents, 78% were faster on voice.

Is the ElevenLabs study independent? No. It is company research about its own platform, the total sample size was not disclosed, and the latency finding is a correlation.

References

• Why voice wins for complex, stressful conversations — ElevenLabs (Nov. 2, 2026)
• Voice Recognition Software Finally Beats Humans At Typing, Study Finds — NPR (Aug. 24, 2016)