How to Build an AI Receptionist with n8n

In short

An AI receptionist built in n8n connects four pieces: a phone or voice layer that captures the call, speech-to-text and text-to-speech to convert between audio and words, a large language model that decides what to say, and n8n as the workflow that ties it all to your calendar and CRM. You wire, host, and pay each piece separately.

• Four pieces: a voice/telephony layer, speech-to-text and text-to-speech, an LLM, and n8n as the workflow connecting them to your calendar and CRM.
• The real cost is not n8n itself. It is the per-minute usage across every connected service, billed separately by each vendor.
• GoHighLevel Voice AI bundles the same pieces into one platform with one published rate and no workflow to build or maintain.

What pieces do you actually need to build an AI receptionist with n8n?

An AI receptionist is not one product. It is several separate services chained together, and n8n is the tool that does the chaining. At a minimum you need four pieces.

Each of those is usually a separate vendor with its own account, its own API key, and its own bill. Building the receptionist means wiring them together correctly, then keeping that wiring working every time one of them changes.

How does a call actually flow through the workflow?

The shape of the flow is the same no matter which specific tools you pick. A call lands on your phone number and the telephony layer forwards it, usually as a webhook, into an n8n workflow. n8n triggers on that event, sends the caller's audio to a speech-to-text step, and passes the resulting text to the LLM step along with your prompt and whatever context you want it to have. The LLM's reply goes through text-to-speech and back down the same connection so the caller hears it. If the conversation reaches a booking, the workflow calls your calendar's API to check availability and create the appointment, then writes the contact and a call summary into your CRM.

We are not naming a specific pre-built template here on purpose. The right combination of telephony, speech, and LLM nodes changes as those tools update their integrations, and a screenshot of someone else's workflow from a few months ago is often already out of date. Build the flow above from the current nodes n8n offers for your chosen vendors.

This is also real-time work, not a background automation nobody is watching. Latency, error handling, and uptime all matter in a way they never do for a workflow that runs quietly overnight.

What does it actually cost to run in n8n?

As with any AI receptionist, the number that matters is not the platform fee. It is the per-minute usage, and building it yourself in n8n does not remove that cost. It splits it across more bills instead of one.

n8n itself is usually the smallest line. It can be free if you self-host, or a paid monthly plan on n8n's cloud. The real cost sits with the services n8n calls: your telephony provider bills per minute, your speech-to-text and text-to-speech provider usually bills per minute or per character, and your LLM provider bills per token. None of those rates are fixed, so the honest way to estimate your bill is to add up each vendor's own published rate for your expected call volume rather than trust one quoted number.

Is n8n reliable enough for real phone calls?

n8n is a general workflow tool, not phone infrastructure. It is reliable at what workflow tools are good at: rules, data, and retries. A live phone call adds requirements a background automation never faces, low latency because the caller is waiting in real time, uptime for the exact moment someone calls, and clean error handling so a caller never sits on dead air if one step fails.

If you self-host n8n, that reliability is your job. The instance has to stay up, and every connected service has to respond fast enough for the conversation to feel natural. n8n's cloud plan removes the hosting question, but the latency and error-handling design is still yours to build.

That is the honest tradeoff of doing it yourself: full control over every piece, and full responsibility for every piece staying connected.

The no-glue-code alternative: GoHighLevel Voice AI

If maintaining four connected services across n8n is more than you want to own, GoHighLevel Voice AI is the same job pre-wired. The phone number, the LLM, the speech conversion, and the calendar and CRM sync all live inside one platform, billed on one published per-minute rate instead of four separate vendor bills to reconcile.

You still write the prompt and decide what the receptionist should say and do. You are just not the one keeping a webhook, a speech-to-text step, and a calendar API call all working together every time one of them updates.