How to build an AI appointment setter in GoHighLevel (2026, step by step)

Every message that comes into my business lands in the Conversations tab: SMS, email, Instagram, TikTok, Facebook. A lot of what you'd see in there are Instagram DMs being answered by AI.

The flow is simple. Somebody comments on one of my videos, a workflow sends them a DM, and the AI agent picks it up from there with a question like: are you looking to start an agency, or are you looking to buy his services? Then it sends them to the right offer depending on the answer. It runs all day and all night, which is the point. The money is made in the follow-ups, and the AI never forgets one.

For clients it's the same idea pointed at bookings. I built an AI appointment setter for a daycare whose problem was following up with leads, qualifying them and getting tours booked. The bot does the qualifying and books the tour, and a nurture sequence keeps working on the ones who aren't ready yet.

Most of the result comes from the prompt. Keep it plain and specific. Something like this, then swap in the business details:

"You're the booking assistant for [business], a [type of business] in [city]. Sound friendly and professional. Your job is to answer questions about our services and book a [consultation, tour, free checkup] for people who are a good fit. Ask one question at a time. Before booking, find out [criteria 1], [criteria 2] and [criteria 3]. If they fit, offer to book. If they don't, thank them and let them know the team will follow up. Only answer from the information you've been given. If you don't know, say you'll check with the team and get back to them. Never make up prices, availability or medical, legal or financial advice. Keep replies short, like a text from a real person."

Two things from my own builds. Give it the context of the first message the lead got, so a reply of "yes" makes sense to it. And tell it what to do when it's stuck, because "I'll check and get back to you" is always better than a confident wrong answer.

Every bot has three modes. Off does nothing. Suggestive drafts a reply in the message box and a person on the team reviews it, edits it and hits send. Auto-Pilot replies on its own.

New bots can start in Off, so check the mode before you wonder why nothing's happening. Run a new client's bot in Suggestive for the first stretch so the owner sees exactly what it would say, then flip it to Auto-Pilot. HighLevel's own docs say to test booking, FAQs and unknown questions before turning on Auto-Pilot. It also makes the owner feel in control, which helps when you're selling it.

Single calendar is the fastest to launch and fine for a business with one appointment type. If they have a few, like a consult and a treatment, pick Multiple Calendars in the Appointment Booking goal, give each calendar a clear description and keywords, and set a fallback calendar for anything that doesn't match. Multiple calendars and services calendars only work on prompt-based bots.

A few limits to know: the bot can't take payment during booking, and private services aren't supported. It does handle "tomorrow" properly even if the lead replies the next morning, and it won't book a contact into a time they already have an appointment.

Turn on Human Handover in Bot Goals (it's a Labs feature, so you may need to enable it first). It has three scenarios: the person asks for a human (add phrases like "can I talk to someone real"), the bot doesn't have the information, or it failed to resolve the issue after two tries.

When it hands off it can assign the conversation to a team member, send a final message like "Thanks! Someone from the team will get back to you", pause itself for as long as you set, create a task due in 24 hours, and tag the contact. Make sure the staff member has notifications on for assigned conversations under Settings > My Staff, or the handoff goes nowhere.

Before this you needed a workflow to send the first text and then hand the reply to the bot, which is how I built my dental setter back in January 2025. Since Oct 27, 2026 the agent can start the conversation itself. In the agent go to Deploy > Outbound > Triggers and pick from nine: form submitted, Facebook or Instagram lead form, Google lead form, TikTok lead form, contact created, tag added, appointment booked or confirmed, opportunity created or moved, and missed call.

Speed to lead is the whole game here. A sales rep takes hours to get back to a Facebook lead, the AI does it in seconds, when the lead is most interested. Missed call text-back is the other easy win for a local business.

The rules: it's in Labs (Settings > Labs > Conversation AI | Outbound AI Agents), outbound only goes out by SMS, you need a sending number and a compliance confirmation before you can activate it, and it's one trigger of each type per agent. Leads that come in at 2 a.m. get held until working hours start. Do Not Disturb contacts are skipped and the first message carries an AI disclaimer. Only text people who opted in.

Since Oct 27, 2026 you're not stuck on OpenAI. Turn on Conversations AI | Claude & Gemini Model Support in Settings > Labs and the Model menu lists Claude Opus 5.5, Sonnet 5.5, Haiku 5.5 and Haiku 4.5, and Gemini 3.8 Flash, 3.7 Flash and 3.5 Flash Lite next to the GPT models. Each one shows a relative cost from $ to $$$$.

For a booking bot on SMS, a fast cheap model is usually plenty. Save the top models for longer sales conversations. You can also set a secondary model from a different provider so the bot keeps working if the main one goes down. Test again after you switch, because each model writes differently. And if any of your bots were on GPT 5 or GPT 5 Mini, move them, those retired on Oct 30.

Don't just send it three friendly messages and hand it over. On my voice agents I run simulations before anything goes live: generate a bunch of scenarios, including weird ones like someone asking for an appointment yesterday or demanding a human, run them all, read what failed, fix the prompt and run it again until it passes almost everything.

Conversation AI has the same thing now. Prompt Optimizer (in Labs) builds scenarios from your prompt, knowledge base and booking setup, runs real chats against a copy of your bot, scores each one pass or fail, and suggests an improved prompt with a side by side diff. Your live prompt doesn't change until you click Use Prompt.

One catch: the actions still run during tests, so point the bot at a test calendar and test workflows first or you'll fill the client's calendar with fake bookings. Then test from your own phone on every channel, ask it something outside its knowledge, and try to cancel and reschedule.

Conversation AI isn't included in the base plans. Without an AI plan it's pay per use, priced on the tokens each reply uses: the prompt, the knowledge base content it pulls, the conversation history and the reply itself. HighLevel's own example is a conversation with 100,000 input tokens and 25,000 output tokens on a $1.25 / $10 per million model costing about 38 cents. A short SMS booking chat on a cheap model like Claude Haiku 5.5 or Gemini 3.5 Flash Lite comes out well under a cent per reply by the same math.

Or put the sub-account on an AI Employee plan. Growth is $50 a month and includes 1,000 AI replies (plus 100 voice minutes). Unlimited is $97 a month, unlimited with fair use. The $20 Starter plan doesn't include Conversation AI and is only offered to agencies on the $497 plan. Texts and calls are billed separately on every option (rates in Usage fees).

When an allowance runs out the bot pauses by default. Turn on Extra Usage in AI Suite if you want it to keep going at pay-per-use rates. Also, raising the max messages per conversation means longer chats, which means more usage.

Don't walk in selling software. Business owners don't care about the CRM or which model it runs on. They want to know one thing: do I have something talking to my leads and booking me appointments? Sell that. With the daycare, I listened to the problems (slow follow-up, no qualifying, tours not getting booked), worked backwards to what to build, and the CRM was a bonus on top.

Build a demo for the niche first, then test the business yourself. Fill in their contact form or DM their Instagram after hours and see how long it takes them to reply. If nobody gets back to you, that's your opening.

You could build this for a dentist, a med spa or a home service company and charge a setup fee plus a monthly fee you set, and people do. It keeps replying and booking every month, which is why it works as a monthly service. You can resell the AI Employee plan at your own price or rebill usage from Agency View, and since Oct 30, 2026 Pro agencies can set usage limits per SaaS client in Labs. My full playbook for running this as an agency is in Start an AI automation agency.

Text is the easier first build. It covers web forms, lead ads, Instagram and Facebook DMs and missed-call text-backs, and leads can reply when it suits them. Voice is for businesses whose leads mostly call. A business can have both, booking into the same calendar. The voice build, settings and per-minute costs are in my AI voice agent guide.

Check in this order: the mode isn't Off, the channel is connected and assigned to the bot, the bot is set as primary (non-primary bots only reply inside workflows), the bot hasn't hit its max messages and gone to sleep, and the sub-account hasn't used up its AI allowance with Extra Usage off. A sleeping bot wakes up when you mark the conversation as Read, or with the Update Conversation AI Bot and Status workflow action.