ElevenLabs blends voice and text for Conversational AI

Trying to spell out an alphanumeric order code or an email address to a voice bot is arguably one of the most painful customer service experiences on the internet. Miss one letter, and the whole lookup fails.

ElevenLabs just solved that headache by adding concurrent text and voice multimodal capabilities to its Conversational AI platform. Now, instead of picking between talking or typing, users can do both inside the same conversation.

Stop reading credit card numbers out loud

While voice agents feel futuristic, voice-only setups regularly break down in the real world. ElevenLabs pointed out two major corporate deployment failures: speech-to-text inaccuracies with tricky data like tracking IDs or user handles, and the awkward friction of forcing people to dictate sensitive credentials or long numbers.

With this update, AI agents can process spoken words and typed inputs simultaneously. You can chat freely by voice, switch to typing when you need surgical precision or privacy, and jump right back to talking without restarting the session.

If you prefer a silent flow, the agents can also be toggled into traditional text-only chatbot mode.

One-line HTML toggle

For developers, rolling this out requires almost zero lift. In the ElevenLabs widget dashboard, it is literally a single switch called Allow Text Input, deployable via one line of HTML embed code.

For custom setups, the multimodal feature works across ElevenLabs SDKs and bidirectional real-time WebSockets. It inherits all the existing platform perks, including access to expressive voices in over 32 languages and compatibility with Twilio and SIP telecom infrastructure.

Why it matters

We think this is a refreshingly pragmatic move from ElevenLabs. Many voice tech companies spent the last year pretending that keyboards were obsolete, pushing users into clunky, all-voice interactions that sound great in demos but fall apart in messy enterprise deployments.

Giving a voice agent a typing box is not a step backward; it is acknowledging how humans actually communicate. If you run customer support, product onboarding, or lead capture bots, enabling this should immediately slash abandonment rates and transcription errors on hard alphanumeric queries. It is worth turning on right away if you are already using their web widget.

For the broader industry, it sets a much-needed design pattern: conversational AI does not need to be exclusively voice or text. The winning play is hybrid utility.