How to Use Claude or Gemini in GoHighLevel Conversation AI (and Set a Fallback Model)

This is one of the more useful Conversation AI changes this year. Until now, every Conversation AI agent was powered by OpenAI models, so if you thought a different model would follow your booking rules better, or you wanted a backup in case a provider went down, there was nothing you could do inside HighLevel. That has changed.

In this guide I will walk you through what HighLevel actually released, how to switch an agent over step by step, how to choose a sensible fallback, and how to test the change without risking a live client account. Everything here is based on HighLevel's own release notes and help docs, linked at the bottom.

What HighLevel changed

In its Release Radar for the week of 19 October 2026, HighLevel confirmed that Conversation AI agents can now be powered by Anthropic Claude or Google Gemini models, as well as OpenAI. The key points from that release:

• Per-agent choice. You pick the model agent by agent, so a booking agent and a support agent in the same sub-account can run on different models.
• Seven new models. HighLevel lists seven new options. The full list is in the table below.
• Fallback models. You can set a secondary model that automatically takes over if the primary provider has downtime or slow responses.
• A better conversation engine. HighLevel says the engine underneath has been upgraded to handle several requests in one reply, stick to workflow restrictions more strictly, and work out relative dates such as "next Tuesday" more accurately.
• It sits in Labs for now. You have to switch it on in Labs before the new models appear.

HighLevel's changelog entry of 27 October 2026 adds more detail:

• Cost labels. Every model in the menu shows a relative cost from $ to $$$$, so you can weigh quality against cost before you switch.
• A secondary model from any provider. It cannot be the same model as the primary, and the agent switches to it automatically when the primary model is down.
• Switching by chat. In Build with AI you can ask for a different model and the agent's primary model is updated for you.
• Nothing else changes when you switch. Your prompt, actions, knowledge base, channels and settings stay as they are.
• GPT 5 and GPT 5 Mini retire on 30 October. Agents on these models should be moved to another model before then. See how to move them.

If you would rather watch than read, this 44 second Short covers the same changelog entry: the new Claude and Gemini models, the cost labels, the secondary model and the GPT 5 retirement date. The feature is still in Labs, so switch it on before you look for the new models.

More short GoHighLevel walkthroughs like this are on my YouTube channel, Code With Urooj.

HighLevel's Bot Goals overview explains where the model selector lives: in the agent's Bot Goals tab, inside the Prompt area, alongside the personality, goal and additional information fields. At the time of writing, HighLevel's help articles still describe that selector in terms of OpenAI models. The Claude and Gemini options are covered by the Release Radar and appear once the Labs feature is switched on.

The current model list (27 October 2026)

These are the Claude and Gemini models HighLevel's changelog lists in the Conversation AI Model menu, with the Labs feature switched on:

About the cost labels: HighLevel says every model shows a relative cost from $ to $$$$, but its changelog does not say which label each model carries. Read the label next to each model in your own Model menu rather than relying on a list. HighLevel has also not published per-reply prices for these models in the changelog.

Conversation AI is not the same as the AI Agent action in workflows

HighLevel has also added Anthropic and Google models to the AI Agent action inside workflows. That is a separate setting with its own model picker and a reasoning effort control (Low, Medium or High). Changing your Conversation AI agent does not change any AI Agent actions in your workflows, and the other way round. If you use both, check both.

The model names listed for each feature are not identical, and HighLevel updates them often, so always trust what you see in your own dropdown over any list in a blog post, including this one.

Before you start

Ten minutes of preparation here saves a lot of awkward client messages later.

• Check your access. You need admin access to the sub-account, or the right Settings permissions, to see Labs.
• Copy your current prompt. Paste the agent's personality, goal and additional information into a document so you can roll back quickly.
• Note the current model. Write down which model the agent uses today. That is your baseline for testing.
• Pick one agent to start with. Choose a low-risk agent, ideally one that handles general questions rather than paid bookings.
• Write five test conversations. Use real questions your leads ask, including a multi-part question, a relative date ("can I come in next Thursday afternoon?"), a booking request and a request for a human.

Step by step: switch a Conversation AI agent to Claude or Gemini

Step 1: Open Labs in the sub-account

Inside the sub-account, go to Settings and then Labs. If you cannot see the feature, your agency may control Labs centrally. In that case an agency owner or admin goes to Agency View, Settings, Labs, opens the Sub-account tab, chooses Manage, selects the sub-account and saves. HighLevel's Labs overview covers this in more detail.

Step 2: Turn on Claude & Gemini Model Support

Find Conversations AI | Claude & Gemini Model Support in the Labs list and switch it on. HighLevel's changelog uses two names for it: that one under Availability, and Conversations AI Provider Support in its setup steps, so look for either. This only makes the new models available. It does not change any of your existing agents by itself.

Step 3: Open the agent you want to change

Go to AI Agents, then Conversation AI. Open the agents list and click to edit the agent you picked for testing.

Step 4: Choose the new model

Open the Bot Goals tab. At the top of the Prompt area you will see the Model menu. With the Labs feature on, it now includes Claude and Gemini options as well as OpenAI, and each model shows a relative cost label from $ to $$$$. Select the model you want to try.

Change only the model at this stage. Leave your prompt exactly as it is. If you change the prompt and the model together, you will not know which change made the difference.

Step 5: Set a secondary (fallback) model

Click the settings icon next to the model. You will see your Primary Model and can pick a Secondary Model. HighLevel's changelog says the secondary model can come from any provider but cannot be the same model as the primary, and that the agent switches to it automatically when the primary model is down. Its Release Radar also mentions latency. My advice is to pick a fallback from a different provider to your main model. If your primary is a Claude model, a fallback from OpenAI or Google gives you real protection, because a problem at one provider is less likely to affect another.

Test the fallback with the same prompt too. A fallback you have never tried is a guess, not a backup.

Step 6: Save

Click Save. Do not switch the agent on for more channels or more sub-accounts yet.

Step 7: Test before it talks to real leads

HighLevel's own instructions end with "test the agent before using the updated configuration in live conversations", and I would underline that twice. Run your five test conversations and check each one against this list:

• Did it answer every part of a multi-part question?
• Did it get the relative date right and offer real calendar slots?
• Did it stick to the rules in your prompt, such as not quoting prices it should not quote?
• Did it hand over to a human when asked?
• Did the tone sound like your client's business, in the right spelling (UK or US)?
• Was the reply speed acceptable for chat and SMS?

If the new model does better on your tests, roll it out to that one agent for a few days and watch the conversations. Only then move on to the next agent.

Or switch the model by chatting in Build with AI

If you build agents with Build with AI, you can change the model by asking for it in the chat. HighLevel's example is "Switch the model to Claude", and the agent's primary model is updated for you. Two things to know:

• Build with AI is also a Labs feature. HighLevel says to turn it on under Settings, Labs.
• Check the result. HighLevel's changelog only mentions the primary model here, so open the Model setting afterwards to confirm which model was chosen and to set or check the secondary model yourself.

Move off GPT 5 and GPT 5 Mini before 30 October

HighLevel's 27 October changelog entry says GPT 5 and GPT 5 Mini retire on 30 October 2026, and that agents on these models should be moved to another model before then. HighLevel does not say what happens to agents still on them after that date, so do not wait to find out. For each sub-account:

1. List your agents. Go to AI Agents, then Conversation AI, and open each agent in turn.
2. Check the Model setting. Note any agent whose primary model is GPT 5 or GPT 5 Mini. Open the settings icon next to the model and check the secondary model too, so your fallback is not on a retiring model either.
3. Copy the prompt first. Save the personality, goal and additional information somewhere safe, as in the steps above.
4. Pick a replacement. Choose another model from the Model menu. If you want Claude or Gemini, switch on the Labs feature first (Step 2). If you would rather stay on OpenAI, pick another GPT model from your menu. Use the $ to $$$$ labels to compare cost.
5. Set a secondary model from a different provider, as in Step 5.
6. Test before you save. HighLevel notes that each model writes in its own way, so run your test conversations in Test your agent, then click Save.
7. Watch the first live conversations on the new model, and tell clients if the AI provider behind their agent has changed.

The AI Agent action in workflows has its own model picker, so check it separately.

How to choose between OpenAI, Claude and Gemini

I am not going to tell you one model is "the best". HighLevel itself does not, and the honest answer depends on your prompt, your client's industry and what the agent has to do. Here is a simple way to decide that does not rely on anyone's hype.

On cost: the Model menu now shows a relative cost label from $ to $$$$ for every model, and HighLevel suggests lower-cost models such as Claude Haiku 5.5 and Gemini 3.5 Flash Lite for simple, high-volume conversations. Its release notes still do not set out per-reply prices for Conversation AI, so check the AI usage and billing details in your own account before you move a busy agent to a different model. Do not assume every model costs the same per reply.

Common mistakes to avoid

• Switching every agent at once. If something goes wrong, you will not know where. Move one agent at a time.
• Editing the prompt and the model together. Change one thing, test, then change the next.
• Skipping the fallback. It takes seconds to set and protects you when a provider has a bad day.
• Forgetting workflow AI actions. The AI Agent action in workflows has its own model setting. Update it separately if you want them to match.
• Not telling the client. If you run agents for clients, let them know which AI provider now handles their customers' messages. Some businesses, particularly regulated ones, will want to know and may need to update their own privacy information.

Where this fits with Claude MCP

People sometimes mix these two up, so to be clear: choosing Claude inside Conversation AI changes the model that talks to your leads in chat and SMS. Connecting Claude to GoHighLevel through MCP changes how you run the account, such as updating contacts, building pipelines or checking workflows by typing plain English instructions.

They work well together, but they solve different problems. If you want the second one, read how to connect Claude AI to GoHighLevel using MCP, see the GoHighLevel MCP server, or compare the options in GoHighLevel AI Studio vs Claude MCP. If you are building an AI sales assistant across CloseBot and GoHighLevel, the CloseBot MCP server page explains how that side works.