Anis Mahmud Anik

AI voice calling in GoHighLevel: how Voice AI answers and books

A matte 3D phone with sound waves and a calendar, representing a Voice AI agent answering calls and booking appointments.

Short answer

A Voice AI agent in GoHighLevel is an automated voice assistant that answers a business phone number, talks to the caller, and completes actions during the call. It can answer common questions, collect caller information, book appointments, send a text, update the contact record, trigger a workflow, or transfer the caller to a person. You build it in AI Agents, then Voice AI, then the Agent List tab, then + Create Agent. You test it with a Web Call or a Phone Call before you deploy it, and you deploy it by assigning it a phone number or a Number Pool. Voice AI is not free: the $20 AI Employee Starter plan includes no Voice AI minutes, so every minute bills at pay-per-use rates unless the sub-account is on Growth or Unlimited.

In this post

  • What a Voice AI agent does on a call
  • Before you start: what to check
  • The three ways to create an agent
  • Build the agent: prompt, Knowledge Base, actions
  • Test before you deploy
  • Deploy: number, routing, working hours
  • What Voice AI costs
  • What it cannot do
  • Launch checklist
  • FAQ
  • Sources

What a Voice AI agent does on a call

The Help Center describes a Voice AI agent as an automated voice assistant that communicates with callers using the instructions, business information, and actions you configure. On a live call it can:

  • Answer questions from the Knowledge Base, using approved business information
  • Collect information from the caller, such as a name, a service needed, or a preferred time
  • Book appointments on a connected HighLevel calendar
  • Transfer the caller to a human phone number, or hand the conversation to another Voice AI agent
  • Send an SMS during the conversation
  • Update a contact field with information collected on the call
  • Trigger a workflow
  • Send a real-time webhook to an external system through a Custom Action
  • Run after-call actions that update records or trigger follow-up after the conversation, without interrupting the caller

The point is coverage, not replacement. The agent handles the routine calls the same way every time, following the same approved instructions and escalation rules, while the team handles the conversations that need a person. The Help Center lists the same trade-off: immediate call handling, consistent conversations, automated actions, business knowledge access, flexible routing, and scalable coverage.

Before you start: what to check

The Help Center lists six things to confirm before opening the builder:

  • Access to the correct sub-account
  • The View and Manage Voice AI Agents permission (if Voice AI is missing, ask a sub-account administrator to review user permissions)
  • A defined use case, such as lead qualification, customer support, appointment booking, or call routing
  • Approved business information for the prompt or Knowledge Base
  • Any calendar, workflow, transfer number, or integration the planned actions need
  • An eligible LC Phone number, Twilio number, or Number Pool, when the agent will handle live calls

Two more facts decide the shape of the build. First, outbound calling is a separate setup: outbound Voice AI requires the applicable approval and compliance setup, and the Outbound testing option may remain unavailable until those requirements are completed. Do not promise outbound calling on day one. Second, assigning a number does not guarantee the agent answers every call. Working Hours, Call Routing mode, forwarding, timeout, voicemail, and the number’s call-flow configuration all affect routing.

The three ways to create an agent

HighLevel provides three starting points. Pick by how much control the conversation needs.

Method Use it when What you get
Browse Marketplace A prebuilt agent fits the use case A ready agent for a specific use case; review all included prompts, actions, workflows, calendars, fields, and Knowledge Base content before deployment
Create Custom Agent A standard call-handling use case, built from scratch The full-screen builder: prompt, Knowledge Base, actions, and call settings
Flow Based Builder The conversation has defined paths, routing conditions, or multiple AI steps A visual, node-based flow; a new blank flow begins with a Start Call node connected to an AI Agent node

The Help Center recommends Create Custom Agent for standard call-handling use cases, Browse Marketplace when a suitable prebuilt agent is available, and Flow Based Builder for structured, multi-path conversations. One caution applies to the Flow Based Builder: the starter flow is not retained until you save the agent.

Build the agent: prompt, Knowledge Base, actions

The remaining steps below follow the Create Custom Agent path.

Step 1: open Voice AI

Log in to the right sub-account, go to AI Agents, then Voice AI, then open the Agent List tab. Click + Create Agent, select a creation method, and click Continue.

Step 2: configure the core agent

The full-screen builder keeps the prompt, settings, and test console together. Enter a clear Agent Name, select the Voice, select the Model and Language, use the clock icon in the toolbar to set the Timezone, write or review the agent prompt, configure the Welcome Message for inbound and outbound calls, choose whether the agent or caller speaks first when available, set the pause before the agent begins speaking, then click Save.

The prompt is the whole job. The Help Center says a reliable prompt defines:

  • The agent’s role and primary objective
  • The information it should collect
  • The topics it may answer
  • The claims or topics it must avoid
  • When each action should run
  • When the caller should be transferred or escalated
  • How the conversation should end

Prompt Optimizer can help improve an initial prompt, but review and test the final instructions. One instruction belongs in every prompt: tell the agent not to guess when required information is unavailable. The launch checklist repeats this point: the agent must not invent answers when information is unavailable.

Step 3: connect a Knowledge Base

A Knowledge Base helps the agent answer business-specific questions using approved information instead of relying only on the prompt. Open Knowledge Base in the configuration panel, select the Knowledge Base the agent should use, add or update the relevant sources, describe when the agent should search the Knowledge Base, then test common questions and edge cases. Useful sources include business hours, services, pricing, policies, locations, and frequently asked questions.

The verdict here is simple: if the answer is not in the prompt and not in the Knowledge Base, the agent should say so and offer a transfer, not invent it. That is the difference between a useful agent and a liability.

Step 4: add actions

Actions allow the agent to complete tasks during the conversation or after the call ends. Add only the actions the agent’s objective requires. Open Actions, click + New Action, select the action type, enter a clear action name, define the condition that should trigger the action, complete the required destination, workflow, field, calendar, or integration settings, then save the action.

The available actions are:

Action What it does
Call Transfer Sends the caller to a human phone number
Trigger a Workflow Starts a selected HighLevel workflow
Send SMS Sends a message during the conversation
Update Contact Field Saves information collected from the caller
Appointment Booking Offers availability and schedules an appointment
Custom Action Sends a real-time webhook request to an external system
Agent Transfer Hands the conversation to another Voice AI Agent

Call Transfer and Agent Transfer are different: Call Transfer sends the caller to a person or external number, Agent Transfer hands the conversation to another Voice AI agent. For Appointment Booking, connect an appropriate HighLevel calendar and test availability selection and booking confirmation before deployment.

Step 5: outbound settings, if the agent will call out

Speech-to-speech (S2S) Voice AI agents can also be configured for outbound calling from Agent Builder. Configure the Outbound welcome message, open Outbound Settings, configure the AI Disclosure for the outbound call, review the Verbal Opt-out configuration, then save the agent after reviewing the outbound settings. The same S2S agent can be used for both inbound and outbound calling. Outbound calling needs the applicable approval and compliance setup first.

Step 6: review call and post-call settings

Advanced configuration is hidden by default. At the bottom of the configuration section, enable Show advanced settings to reach Call Settings, Agent Behavior, Transcription and Speech, Voice Settings, or System Prompts. Review Call Settings (call duration, idle reminders, silence handling), Agent Behavior (response speed, interruption sensitivity, response style), Transcription and Speech, Translation, Voice Settings, Post-Call (notifications, workflows, summaries, follow-up behavior, whether the AI-generated call summary is saved to the contact’s notes), and System Prompts. Keep the default settings when they already match the intended experience, change one setting at a time, and retest so you can identify which adjustment affected the call.

Test before you deploy

Save the agent, then open the Test or Test Audio panel. Two test methods exist:

Method What it tests Limit
Web Call The conversation in the browser, without assigning a phone number Call Transfer does not run during a Web Call test
Phone Call The complete telephony experience through an actual phone connection Needs a phone connection

Test the agent with a common caller request, a question answered by the Knowledge Base, a question the agent should not answer, an incomplete or unclear request, each configured action, a transfer or escalation request, a period of silence, an interruption while the agent is speaking, and the intended call-closing sequence. Then review the call history, transcript, recording, summary, and action results. Repeat until the expected behavior is consistent. Use Phone Call testing or call the deployed number to validate a transfer, because Call Transfer does not run during a Web Call test.

Deploy: number, routing, working hours

Open the agent’s Deploy tab. In the Phone section, select an eligible phone number or configured Number Pool. Available phone numbers appear directly on the Deploy screen, and a new number can be purchased without leaving the Deploy screen. One Voice AI agent can handle calls from multiple numbers when a Number Pool is assigned.

Under Call Routing, select how the agent handles incoming calls:

  • Answer Calls Directly: the Voice AI agent answers incoming calls directly
  • Use as Backup: the Voice AI agent acts as a backup when the primary call path does not answer

Under Working Hours, choose All Hours for unrestricted availability, or Custom Schedule to choose active days and time ranges. Custom schedules can extend through midnight.

In the Chat section, configure a Chat Widget when the agent will also handle website conversations: connect an existing Chat Widget, or create and automatically connect a new one. Review the inbound call flow, forwarding, timeout, voicemail, and fallback behavior associated with the assigned number, then save the deployment settings.

Editorial view, not a HighLevel recommendation: Use as Backup is a lower-risk first deployment for most businesses. The agent answers only when the primary path does not, so a weak prompt affects the unanswered calls instead of every call.

What Voice AI costs

Voice AI is not included for agency or sub-account users without an eligible AI Employee subscription. Usage is billed at pay-per-use rates unless covered by an AI Employee plan. The three AI Employee plans, per sub-account:

Plan Price Voice AI included
AI Employee Starter $20/month Not included; pay-per-use
AI Employee Growth $50/month 100 AI Agent minutes/month
AI Employee Unlimited $97/month Unlimited, subject to fair use

Two conditions matter. First, the Starter plan is available exclusively to agencies subscribed to HighLevel’s $497 Founder plan, and it includes no Voice AI allowance at all, so Voice AI on Starter bills at pay-per-use rates. Second, when an included allowance is reached, additional usage is paused by default. To continue at pay-per-use rates, enable Extra Usage in AI Suite. Phone System charges still apply to all phone calls regardless of plan.

The cost formula

For traditional Voice AI, the Help Center gives this formula:

Total estimated cost = call duration in minutes x (Voice Engine rate + selected TTS rate) + LLM token cost + applicable Phone System charges

The Voice Engine rate is $0.045 per minute, effective 20 May 2026 onward. TTS rates, per minute:

TTS provider Model Rate
OpenAI TTS-1, GPT-4o Mini TTS $0.015/min
Cartesia Sonic 3, Sonic 3 Latest $0.015/min
ElevenLabs Flash V2.5, Flash V2 $0.035/min
ElevenLabs Multilingual V2 $0.05/min
ElevenLabs V4 Turbo $0.07/min
ElevenLabs V3, V4 $0.17/min

The Help Center’s own worked example: a 10-minute call using OpenAI TTS costs $0.45 in Voice Engine charges (10 x $0.045) plus $0.15 in TTS charges (10 x $0.015), for an estimated AI voice cost of $0.60, before LLM token usage and Phone System charges. Do not quote a single per-minute figure as the total cost: the formula has four parts, and two calls with the same duration can cost different amounts because the TTS provider, AI model, token usage, and Phone System charges all affect the final cost.

As an example, a month with 100 calls of 3 minutes on OpenAI TTS is 300 minutes at $0.06 per minute (Voice Engine $0.045 plus OpenAI TTS $0.015), which is $18.00 before LLM token cost and Phone System charges.

Speech-to-speech rates

Speech-to-speech (S2S) Voice AI uses a single per-minute rate instead of the formula above. The rate bundles audio, text, cache tokens, transcripts, and infrastructure. S2S agents hear audio and respond with audio through a single realtime model, which removes the speech-to-text and text-to-speech steps and lowers latency. Rates checked 15 November 2026:

Provider Model S2S rate
Google Gemini 3.8 Live $0.10/min
Google Gemini 3.1 Flash Live Preview $0.10/min
OpenAI GPT Realtime 2.1 $0.20/min
OpenAI GPT Realtime 2 $0.20/min
OpenAI GPT Live 1 $0.12/min

Agent prompts are capped at 15,000 tokens on both S2S models. This applies to the agent’s instructions, not to call length or conversation history. The pricing article warns that pricing and model availability may change, and to confirm current in-app pricing and billing settings before quoting externally.

What it cannot do

  • It cannot answer every call on an assigned number automatically. Working Hours, Call Routing mode, forwarding, timeout, voicemail, and the number’s call-flow configuration all affect routing.
  • It cannot run Call Transfer during a Web Call test. Use Phone Call testing or call the deployed number to validate a transfer.
  • It cannot place outbound calls without the applicable approval and compliance setup. Outbound testing may remain unavailable until those requirements are completed.
  • It cannot invent answers it does not have. The prompt must tell it not to guess, and the Knowledge Base must hold the approved facts.
  • It cannot run on the $20 Starter plan for free. Starter includes no Voice AI minutes, so every minute bills at pay-per-use rates.
  • It cannot keep a Flow Based Builder starter flow without saving the agent first.
  • It cannot guarantee a fixed per-minute total. The cost depends on the Voice Engine rate, the TTS rate, LLM tokens, and Phone System charges.

Launch checklist

From the Help Center’s final launch checklist, before the agent takes live calls:

  • The agent name, voice, model, and language are correct
  • The inbound and outbound greetings sound natural
  • The prompt clearly defines the agent’s role, limits, and objective
  • Knowledge Base answers are accurate and relevant
  • The agent does not invent answers when information is unavailable
  • Each action runs only under its intended conditions
  • Phone Call testing confirms that transfers reach the correct destination
  • The correct phone number or Number Pool is assigned
  • Working hours and after-hours behavior are correct
  • The inbound call flow routes callers to the agent as intended
  • A live external call reaches the deployed agent
  • The call log contains the expected transcript, summary, and action results

FAQ

What is a Voice AI agent in GoHighLevel?
An automated voice assistant that answers calls, responds to common questions, collects caller information, and completes actions such as booking appointments or transferring calls, using the instructions, business information, and actions you configure.

Which creation method should I choose?
Use Create Custom Agent for standard call-handling use cases. Use Browse Marketplace when a suitable prebuilt agent is available. Use Flow Based Builder for structured, multi-path conversations.

Can I test an agent without assigning a phone number?
Yes. Use Web Call to test prompts, Knowledge Base responses, and most actions directly in the browser. Call Transfer does not run during a Web Call test, so validate transfers with a Phone Call test or by calling the deployed number.

Can one Voice AI agent handle calls from multiple numbers?
Yes. Assign a configured Number Pool when multiple numbers should route to the same agent.

Can Voice AI book appointments during a call?
Yes. Add the Appointment Booking action and connect an appropriate HighLevel calendar. Test availability selection and booking confirmation before deployment.

How much does Voice AI cost?
It depends on the plan and the usage. AI Employee Growth at $50 a month includes 100 Voice AI minutes a month; Unlimited at $97 a month includes Voice AI subject to fair use; Starter at $20 a month includes no Voice AI minutes, so every minute bills at pay-per-use rates. The pay-per-use cost is call duration times the Voice Engine rate plus the TTS rate, plus LLM token cost, plus Phone System charges. The Voice Engine rate is $0.045 a minute.

Can Voice AI make outbound calls?
Yes, for speech-to-speech agents configured in Agent Builder, with an outbound welcome message, AI Disclosure, and Verbal Opt-out settings. Outbound calling requires the applicable approval and compliance setup first.

Why is the agent not answering the assigned number?
Confirm the number or Number Pool is assigned to the correct agent, then review Working Hours, the selected Call Routing mode, inbound routing, forwarding, timeout, voicemail, and the phone number’s call-flow configuration.

Need Voice AI answering inside your CRM?

If missed calls are going to voicemail while the team is busy, and the business already runs on GoHighLevel, book a free 30-minute call. The first step is one agent with one job, such as answering and booking, tested on a Phone Call before it takes a live call.

Sources

Checked 2026-11-15. All GHL behaviour in this post comes from these sources, not from memory.

  • HighLevel Help Center, “Create, Test, and Deploy a Voice AI Agent”. https://help.gohighlevel.com/support/solutions/articles/155000004107-how-to-create-voice-ai-agents
  • HighLevel Help Center, “HighLevel AI Products Pricing: Voice AI, Agent Studio & More”. https://help.gohighlevel.com/support/solutions/articles/155000006652-ai-product-pricing

Related reading