How to create a voice assistant with AI: from idea to a functional app that listens, responds and acts for you

Before building it, it helps to understand the 3 pieces that make it up:

1. Listens (speech-to-text): converts your voice into text. A transcription model does this (like OpenAI’s Whisper).
2. Thinks (LLM): the text goes to a language model (GPT-4, Claude, Gemini) that generates the response.
3. Speaks (text-to-speech): the written response is converted to voice with a model like ElevenLabs or the browser’s native voice.

The no-code platforms we’ll use in this guide already have these 3 pieces ready. You just configure them.

ElevenLabs is the fastest tool to get a voice assistant running today. It has a conversational agent builder that requires no code.

1. Go to https://elevenlabs.io and create a free account.
2. Verify your email.
3. Once inside, explore the panel. The main menu shows: Speech (text to voice), Dubbing, Projects, and Conversational AI (what we’ll use).

The free plan includes 10,000 credits per month. Enough for this guide.

1. In the side menu, click “Conversational AI”.
2. Click “Create agent” (or “New agent”).
3. You can start from a template or from scratch. To learn, choose “Blank agent”.
4. Give your agent a name (for example, “Support Assistant”).
5. You’ll see 4 main tabs:
   • Agent: the agent’s personality and instructions.
   • Voice: the voice it will use to speak.
   • Analysis: usage metrics.
   • Integrations: how to embed it on your website or app.

The system prompt is where you tell the agent who it is, what it knows, and how it should behave. It’s the most important part.

In the “Agent” tab, find the “System prompt” field and write something like this (adapt it to your case):

You are Sofia, the virtual assistant for [Your Company]. You speak in English with a friendly and professional tone.

Your role is to:
- Answer questions about our products and services.
- Help customers find information on our website.
- If you don't know something, honestly say you'll verify it and ask for the customer's email.

What you should NOT do:
- Make up information that wasn't provided to you.
- Talk about topics irrelevant to the company.

Replace the examples with real information about your business or use case. The more specific you are, the better the agent behaves.

You can also:

• Upload reference documents (FAQs, product catalog, policies) with the “Knowledge base” button.
• Adjust the language model (GPT-4o, Gemini, etc.) from the agent settings.

1. Go to the “Voice” tab.
2. ElevenLabs shows a voice library. You can filter by language, accent, and gender.
3. Listen to samples and pick the one that best fits your assistant.
4. For a natural-sounding Spanish voice, filter by “Spanish”.
5. Save your selection.

Additional options:

• Clone your own voice: if you want the assistant to sound exactly like you, see the tutorial How to clone your voice with AI for the full step-by-step process.
• Adjust parameters: in the Voice tab you can change the speed, stability, and similarity of the voice.

1. At the top of the editor, you’ll find a play or test chat button. Click it and speak (or type) to test your agent.
2. Adjust the system prompt if the agent responds incorrectly or out of context.
3. When you’re satisfied, go to the “Integrations” tab:
   • Chat widget: generates an HTML snippet you can paste into any website to make the assistant appear as a button in the corner.
   • Public link: generates a direct link you can share via WhatsApp, email, or social media.
   • API: if you later want to connect it to your own app, the API gives you full access.
4. Copy the public link and share. Your assistant is now live.

For a voice assistant integrated into your own website, see the tutorial How to create a chatbot for your website without coding, which covers integration options in detail.

If you need an assistant with more complex dialogue paths (menus, conditional questions, multiple possible routes), Voiceflow is the best no-code option:

1. Create a free account at https://voiceflow.com.
2. Create a new project of type “Chat assistant” or “Voice assistant”.
3. Use the visual flow editor to design the dialogue: text blocks, AI blocks, conditionals, external API calls.
4. To add the voice, connect ElevenLabs directly from Voiceflow’s integration settings.
5. Publish and share with the link or chat widget.

Voiceflow has a free plan for personal and learning projects. Confirm current prices at https://voiceflow.com/pricing before scaling.

Problem	Fix
The agent doesn’t understand what I say	Speak more slowly and close to the microphone. Check that your browser has microphone permission.
It responds off-topic	The system prompt isn’t specific enough. Add more instructions about what it should NOT do.
The voice sounds robotic	Try a different voice from ElevenLabs’ library or adjust the stability parameters.
Free plan credits run out	Credits reset monthly. Consider the Starter plan ($6/mo) for regular use.
I want it to remember previous conversations	Basic agents don’t have persistent memory. For that you need an external database connected via API (intermediate level).