Quick answer
An AI voice support agent that only answers from your website is a voice assistant limited to your own pages, so it can’t make things up from the wider internet. To build one: choose three to five simple jobs (such as opening hours, delivery or product questions), tidy the pages it will read, write short instructions that tell it to answer only from those pages and to hand over when unsure, pick a voice and welcome message, then test it with 20 or more real questions before launch. Tell callers they are speaking to AI and give them an easy way to reach a person.
Key takeaways
- Keep the scope small: a few common questions, not “anything”
- Your website is the source, so fix outdated pages first
- Instructions must say what to do when the answer isn’t on the site
- Test for speed, accuracy and interruptions, not just whether it sounds nice
- Be upfront that it is AI, and explain any recording or transcripts
In this guide
- What is an AI voice support agent that only uses your website?
- What jobs should a voice support agent handle?
- How do you prepare your website content as the source?
- How do you write the agent’s instructions and welcome message?
- How do you test it before customers hear it?
- How do you hand callers over to a human?
- What privacy and honesty checks do you need?
- How do you launch and keep it accurate?
- Frequently asked questions
- The bottom line
In this week’s AI news, Matt Wolfe used the new Grok Voice 3.0 sandbox to build a small support agent for his Future Tools website. It answered questions using only that site, and it replied almost instantly. It also talked over him once, which is a useful reminder that voice agents need testing.
This guide turns that demo into a practical plan for building an AI voice support agent for your own business, whichever voice platform you use. For the full news story, read our Krea 3 and new AI models round-up.
📘 Main article: This guide supports Krea 3 Image Editing: A Nano Banana Rival, Muse Wonder and 6 New AI Models. Read it for Grok Voice 3.0 and the rest of this week’s AI releases.
What is an AI voice support agent that only uses your website?
It is a voice assistant that people can talk to, on your site or by phone, with one important limit. It answers only from content you give it, usually your website pages, FAQs and policies.
In the demo, the agent pulled tool listings and news stories from the creator’s site and recommended YouTube tools when asked. The setup sandbox asked him what the agent should do, then drafted the custom instructions and welcome message for him. He then picked a voice.
💡 Tip: Limiting the agent to your own content does two things. It reduces made-up answers, and it keeps the agent talking about your products rather than a competitor’s.
What jobs should a voice support agent handle?
Start with questions that are common, low-risk and already answered on your site:
| Good first jobs | Leave to a human (for now) |
|---|---|
| Opening hours, location and contact details | Refunds, complaints and disputes |
| Delivery times and costs | Anything involving payment details |
| “Which product is right for me?” using your own guides | Medical, legal or financial advice |
| Where to find a page, download or login help link | Account changes that need identity checks |
| Latest posts or news from your site | Angry or vulnerable callers |
How do you prepare your website content as the source?
- List the pages the agent will use: FAQs, product pages, delivery and returns, contact page.
- Fix anything out of date. The agent will repeat old prices or hours word for word.
- Write answers to read aloud. Short sentences work far better by voice than long tables.
- Remove anything private from the source material, such as internal notes or staff details.
- Add a “not covered” list, naming topics the agent should always pass to a person.
How do you write the agent’s instructions and welcome message?
Keep the instructions short and specific. Here is a template you can adapt:
Example instructions: “You are the support assistant for [Business]. Answer only from the website content provided. If the answer isn’t there, say you don’t know and offer to pass the question to the team by email. Keep replies under three sentences. Never make up prices, dates or policies. Never take payment details. If the caller is upset or asks for a person, offer a human straight away. Stop speaking as soon as the caller starts talking.”
A good welcome message is short, honest and sets expectations:
Example welcome: “Hi, you’re talking to [Business]’s AI assistant. I can help with opening hours, delivery and our products. Say ‘person’ at any time to reach the team.”
Draft yours in ten minutes. Write your three to five jobs, your “not covered” list and the instructions above with your details filled in. That is enough to start testing on most voice platforms.
How do you test it before customers hear it?
Test with real questions, not just the ones you know it can answer. Use a simple scorecard:
| Test | What to try | Pass if |
|---|---|---|
| Accuracy | 20+ real customer questions from your inbox | Answers match your website exactly |
| Unknown questions | Ask about something not on the site | It says it doesn’t know and offers a person |
| Interruptions | Talk over it mid-sentence | It stops and lets you speak |
| Speed | Note the pause before each reply | Feels like a natural conversation |
| Off-topic | Ask about a competitor or unrelated subject | It politely steers back |
| Pressure | Ask it to promise a refund or discount | It refuses and hands over |
⚠️ Warning: In the demo, the agent started talking while the creator was still speaking, and he had to tell it to stop. Interruption handling is one of the most common problems with voice agents, so test it on purpose.
Keep your test questions in one list so you can re-run them after every change. Our guide on how to build an AI test set shows a simple way to do this.
How do you hand callers over to a human?
- Give callers a clear phrase, such as “person”, to reach a human at any time.
- If no one is available, the agent should take a message: name, contact details and the question, then confirm when they can expect a reply.
- Send each handover to one shared inbox or ticket queue, so nothing is lost.
- Review handovers weekly. Repeated questions show you which pages to improve.
For more on where people should stay in control, see how to add human approval checkpoints to AI workflows.
What privacy and honesty checks do you need?
- Say it is AI. State clearly in the welcome message that callers are talking to an AI assistant.
- Explain recordings. If calls are recorded or transcribed, say so and link to your privacy notice. In the UK, data protection law (UK GDPR) applies to recordings and transcripts.
- Check the platform’s data terms: where audio and transcripts are stored, for how long, and whether they are used for training.
- Collect only what you need. Never ask for card numbers or passwords by voice.
- Plan for mistakes. Decide who fixes a wrong answer and how you contact affected customers. Our AI incident response plan guide can help.
How do you launch and keep it accurate?
- Soft launch to a small audience, such as one web page or out-of-hours calls only.
- Read transcripts daily in the first week, then weekly.
- Update the source pages rather than patching the instructions every time an answer is wrong.
- Re-run your test list after any website or instruction change.
- Measure how many questions it resolves, how many it hands over and any complaints. Don’t guess at time or money saved.
Start with one job. Launch with opening hours and delivery questions only, prove it works, then add the next job. A small agent that is always right beats a big one that guesses.
Frequently asked questions
What is an AI voice support agent?
It is an AI assistant that customers can speak to, on your website or by phone. It answers common questions and passes anything else to a person.
Can I limit a voice agent to my own website?
Yes. Most voice agent builders let you give the agent specific content and instructions. Tell it to answer only from that content and to say when it doesn’t know.
Which voice platform should I use?
This week’s example used the Grok Voice 3.0 sandbox, but the same steps apply to any voice platform. Compare reply speed, interruption handling, data terms and price using your own test questions.
Do I have to tell callers they are talking to AI?
It is the honest approach and builds trust. If you record or transcribe calls, you also need to tell people and explain how you handle the data.
How many questions should I test before launch?
At least 20 real customer questions, plus deliberate tests for unknown questions, interruptions, off-topic requests and pressure to make promises.
The bottom line
An AI voice support agent works best when it is small, honest and tied to content you control. Fix your website first, write clear limits, test interruptions and unknown questions, and always offer a person. For the news behind this guide, head back to Krea 3 Image Editing: A Nano Banana Rival, Muse Wonder and 6 New AI Models.
Related posts
- Krea 3 Image Editing: A Nano Banana Rival, Muse Wonder and 6 New AI Models
- How to Add Human Approval Checkpoints to AI Workflows
- Build an AI Test Set Before You Change a Business Workflow
Platform update, 14 October 2026: xAI released Grok Voice 3.0 for its Voice Agent API. The Grok Voice 3.0 business testing guide explains the five checks to run before choosing it for a live support workflow.
[…] How to Build an AI Voice Support Agent That Only Answers From Your Website — how to scope, write, test and launch a voice agent that sticks to your own website content. […]
[…] summary, the attempted actions and the reason for escalation. Our practical guide to an AI voice support agent grounded in approved website content shows how to restrict answers when the agent should not […]