GoHighLevel Conversation AI: Setup & Training Guide (2026)

By Dr Priya Jaganathan, GoHighLevel Certified Admin · HL Growth Partner, Australia · Updated 18 July 2026 · 10 min read

On this page: What Conversation AI is · Pricing · Step-by-step setup · Training best practice · Connecting to Workflows · Measuring performance · Common mistakes · FAQ

GoHighLevel Conversation AI is the text-based bot built into every GHL sub-account that can answer enquiries and book appointments across SMS, your website chat widget, Facebook Messenger, Instagram DMs and Google Business Profile chat. I have set it up for dozens of Australian clients, and the difference between a bot that books ten appointments a week and one that gets switched off in embarrassment almost always comes down to configuration and training, not the underlying model.

This guide walks through the full setup in the order I do it for clients: mode selection, channel activation, bot training, appointment booking configuration, testing, and the workflow connections that make it useful. I will also flag the SMS compliance considerations that apply in Australia, which most US-centric tutorials skip.

What Conversation AI is, and how its modes work

Conversation AI is HighLevel's native messaging bot. It reads inbound messages in the Conversations tab, generates replies from the training material you give it, and can execute an appointment booking action against one of your calendars. It is not the same feature as Voice AI, which answers phone calls — if you want the phone side covered as well, see my guide to using GoHighLevel Voice AI as an after-hours receptionist. Conversation AI is strictly text: SMS, web chat, social DMs and GMB chat.

Auto-Pilot vs Suggestive vs Off

The bot runs in one of three modes, set per sub-account and adjustable per channel. Choosing the right one for your rollout stage matters more than most people realise.

My standard rollout is Suggestive for the first fortnight: every edit your team makes to a draft shows you where the training data has gaps. Once drafts go out unedited nine times out of ten, switch to Auto-Pilot.

What Conversation AI costs

Conversation AI is included in the AI Employee bundle, which HighLevel prices at $97 USD per month per sub-account for unlimited usage. That bundle also covers Voice AI, Reviews AI, Content AI, Funnel AI and Workflow AI Assistant. If your usage is light, you can instead stay on usage-based pricing, where you pay per message generated — fine for testing, but the per-message charges add up quickly once a bot is handling real enquiry volume on Auto-Pilot.

For most agencies and local businesses I work with, the unlimited bundle pays for itself if the bot books two or three appointments a month that would otherwise have gone cold. I have broken down the full bundle economics, including rebilling margins for agencies, in my GoHighLevel AI Employee features, pricing and ROI guide. Current platform plan pricing is on the official GoHighLevel pricing page; note the AI Employee bundle sits on top of your base plan.

Step-by-step setup

Step 1: Turn on the bot and pick your channels

In the sub-account, go to Settings → Conversation AI. Create or open your bot, and start in Suggestive mode. Then tick the channels the bot should operate on:

• SMS — requires a working number through LeadConnector or your own Twilio account. In Australia, A2P messaging must comply with ACMA rules under the Spam Act: consent before texting, sender identification, and a working opt-out (STOP handling is built into LeadConnector). Use a reply-capable number, not an alphanumeric sender ID, or the conversation cannot happen.
• Web chat widget — the LeadConnector chat widget installed on your site. Make sure the widget is set to the live chat or SMS-chat type, and that it is actually deployed on the pages where enquiries land.
• Facebook and Instagram DMs — connect the pages under Settings → Integrations first. The bot then answers Messenger and IG DMs inside the same Conversations view.
• GMB chat — if your Google Business Profile is integrated, the bot can respond there too; verify chat is currently available for your listing and region.

Set the response delay to something human — I use 60 to 90 seconds for SMS and 15 to 30 seconds for web chat. Instant replies at 2 am read as robotic and invite people to test the bot rather than talk to it.

Step 2: Train the bot on URLs and FAQs

Under the Bot Training tab you have two inputs: web pages and manual FAQs. For URLs, you can crawl a full site or add individual pages. Do not crawl everything — feed it the pages that answer real customer questions: services, pricing, service areas, about, and any policy pages. Blog archives and old event pages add noise.

Then add FAQs manually for anything not cleanly stated on the site: parking instructions, whether you bulk bill, cancellation policy, what happens on a first appointment. In my experience the manual FAQs do more of the heavy lifting than the crawled pages, because you control the exact wording of the answer.

Step 3: Configure the appointment booking action

This is the part that makes Conversation AI commercially useful. In the bot's goals or intents configuration, set the primary intent to appointment booking and tie it to a specific calendar in the sub-account. The bot will then offer available times pulled from that calendar's real availability, confirm a slot, and create the appointment — which fires any confirmation and reminder automation you already have.

Check three things on the calendar before you connect it: the availability window and timezone are correct (Australian sub-accounts inherit the location timezone, but verify it), the slot duration matches the service, and the calendar's own confirmations will not double up with the bot's messages. Decide which fields the bot must collect before booking — at minimum name and phone; anything extra should exist as a custom field first.

Step 4: Test in Bot Trial before going live

The Bot Trial tab gives you a sandbox chat against your trained bot. Spend at least 30 minutes here pretending to be your most awkward customer: ask about pricing three different ways, try to book outside business hours, and ask something the bot cannot know. A good bot offers a human follow-up; a badly configured one invents an answer. Every wrong or waffly reply becomes a new FAQ entry. The official GoHighLevel help centre keeps the current walkthrough of the Bot Trial and training screens, which change layout a few times a year.

Training best practice

A few rules I apply on every build:

• Write FAQs as answers, not essays. Two or three sentences per answer. The bot composes from your material; long rambling source answers produce long rambling bot replies, which is fatal on SMS.
• State prices explicitly or forbid them explicitly. If you do not want the bot quoting prices, say so in the bot's personality prompt and give it a deflection line. Silence on pricing is how bots hallucinate figures.
• Give the bot a name and boundary in its prompt. "You are Mia, the assistant for XYZ Clinic. You help with questions and book consultations. You never give clinical advice." Boundaries in the prompt outperform hoping the model behaves.
• Retrain after every business change. New prices, new hours, a closed location — the bot only knows what you told it.
• Keep one source of truth. If the website says one price and an FAQ says another, the bot will alternate between them. Update both or remove one.

Connecting Conversation AI to Workflows

Conversation AI gets much stronger when it is wired into Workflows rather than left running standalone. Three connections I set up on almost every account:

Missed call hand-off. A missed-call-text-back workflow sends the first SMS, and Conversation AI takes over the reply thread from there — the caller gets an instant text, then a bot that can actually book them in. If you have not built that first piece yet, start with my missed call text back setup guide.

Tag-based escalation. Use an inbound message trigger filtered for phrases like "complaint", "refund" or "speak to a person" to apply a tag, notify a human, and pause the bot for that contact. The dedicated Conversation AI workflow action can switch the bot off per contact, so it stops talking over your sales team once a human picks up the thread.

Appointment follow-through. When the bot books an appointment, the appointment trigger fires your normal confirmation, reminder and no-show sequences. Add a "booked-by-ai" tag in that workflow so you can report on bot-sourced bookings separately — you will want that number when someone asks whether the $97 is worth it.

Measuring performance

Do not judge the bot on vibes. Four numbers tell you whether it is working:

• Bot-booked appointments per week — filter appointments by your booked-by-ai tag. This is the headline metric.
• Containment rate — the share of conversations resolved without human takeover; 60 to 80 per cent is realistic for a well-trained bot on routine enquiries.
• Edit rate in Suggestive mode — while ramping up, track how often staff edit drafts before sending. Under ten per cent means you are ready for Auto-Pilot.
• Speed to first response — cutting this from four hours to two minutes usually moves booking rates more than any copy tweak.

Review transcripts fortnightly for the first quarter. Ten minutes of reading real conversations surfaces more training gaps than any dashboard.

Common mistakes to avoid

• Going straight to Auto-Pilot on day one. Untested bots answer confidently and wrongly in front of real customers. Run Suggestive mode first and earn the switch.
• Crawling the entire website as training data. Old blog posts and expired offers pollute answers. Curate the URLs and lean on manual FAQs.
• Connecting the wrong calendar, or one with broken availability. The bot will cheerfully offer slots at 3 am if the calendar timezone or availability is misconfigured. Test a booking end to end.
• Ignoring SMS consent obligations. Australian A2P traffic sits under the Spam Act and ACMA enforcement; texting scraped or cold lists via the bot risks penalties, not just carrier filtering.
• No human escape hatch. Every bot needs an escalation path — a phrase trigger that pauses the AI and pings a person. Trapping a frustrated customer in a bot loop costs more than the bot saves.
• Set-and-forget training. Businesses change; bots do not, unless retrained. Diarise a monthly review of FAQs, pricing and hours.

If you want Conversation AI set up and trained properly on your own GoHighLevel account, book a strategy call with the HL Growth Partner team.

Book Your Strategy Call →

Frequently asked questions

Is Conversation AI included in the standard GoHighLevel plans?

No. Conversation AI is an add-on: it is part of the AI Employee bundle at $97 USD per month per sub-account for unlimited usage, or you can pay usage-based per-message pricing on top of your base $97 or $297 platform plan.

Can Conversation AI book appointments by itself?

Yes. In Auto-Pilot mode, with the appointment booking intent tied to a calendar, the bot offers real available times, confirms a slot with the contact, and creates the appointment — which then fires your normal confirmation and reminder workflows.

Which channels does Conversation AI work on?

SMS (via LeadConnector or Twilio numbers), the LeadConnector web chat widget, Facebook Messenger, Instagram DMs and Google Business Profile chat. It does not answer phone calls; that is the separate Voice AI feature.

How long does it take to set up and train the bot?

The technical setup takes about an hour: mode, channels, calendar connection and initial training on your key pages and FAQs. Budget another one to two weeks in Suggestive mode to refine answers from real conversations before switching to Auto-Pilot.

Is it legal to use an AI bot over SMS in Australia?

Yes, provided you follow the Spam Act and ACMA rules that apply to all commercial SMS: express or inferred consent before messaging, clear sender identification, and a functional opt-out. LeadConnector handles STOP opt-outs automatically, but consent is your responsibility.