Until recently, every Conversation AI bot in GoHighLevel ran on OpenAI models. HighLevel has now added Anthropic's Claude and Google's Gemini as model options for Conversation AI agents. For agencies running GoHighLevel AI automation for clients, that means you can pick the model that fits each bot — a fast, low-cost model for simple FAQ replies, a stronger reasoning model for sales qualification — and set a second model to take over if the first is unavailable.
This guide covers what changed, how to enable it, how to choose a model, and a safe way to switch a bot that is already talking to leads.
Quick version: enable Conversations AI Provider Support under Settings → Labs, open your agent's Build tab, choose a Claude, Gemini or OpenAI model from the Model dropdown, optionally pick a secondary (fallback) model, then test before you publish.
What changed in Conversation AI
According to HighLevel's changelog and help documentation, Conversation AI agents can now run on models from three providers instead of one:
- OpenAI — the models Conversation AI has always used.
- Anthropic Claude — several Claude tiers, from fast/low-cost to most capable.
- Google Gemini — Gemini Flash-family models aimed at speed and cost.
Two details matter most for agencies:
- The model is chosen per agent. Your appointment-booking bot and your support bot can run on different models.
- You can set a secondary model. If the primary model's provider has an outage or is unavailable, the secondary model takes over, so conversations keep flowing.
The exact model names in the dropdown change as providers release and retire models, and HighLevel's own documents haven't always listed identical lineups. Treat the dropdown in your account as the source of truth. HighLevel has also been retiring older OpenAI models on a schedule, which is one more reason to review which model each of your bots is on.
How to enable Claude and Gemini in Conversation AI
Step 1: Turn on the Labs feature
Go to Settings → Labs in the sub-account and enable Conversations AI Provider Support. Until this is on, the extra providers won't appear in the model list. If you manage many sub-accounts, remember this is a per-location setting, so add it to your onboarding checklist or snapshot process.
Step 2: Open the agent's Build tab
Go to AI Agents → Conversation AI, open the agent you want to change, and go to the Build tab.
Step 3: Choose a primary model
Open the Model dropdown and select a Claude, Gemini or OpenAI model. This is the model that will handle that agent's conversations.
Step 4: Add a secondary (fallback) model
Pick a secondary model — ideally from a different provider than the primary. If your primary is Claude, a Gemini or OpenAI fallback protects you from a single provider's outage. Using two models from the same provider gives you much less protection.
Step 5: Test before you publish
Each model interprets your prompt slightly differently. Run your agent through the test panel with real-world messages before it goes live (see the checklist below).
Which model should you choose?
There's no single best model — it depends on what the bot does and how many conversations it handles. A practical way to decide:
| Bot type | What matters most | Model tier to start with |
|---|---|---|
| FAQ / support replies | Speed and cost at volume | A fast, low-cost tier (e.g. Gemini Flash or Claude Haiku class) |
| Appointment booking | Following instructions reliably, handling dates and times | A mid tier (e.g. Claude Sonnet class) or your proven OpenAI model |
| Sales qualification / objection handling | Nuance, tone, multi-step reasoning | A stronger tier; test against your best human replies |
| High-volume reactivation campaigns | Cost per conversation | The cheapest model that passes your test script |
Costs differ by model. HighLevel publishes per-model rates on its AI Product Pricing page, priced per million input and output tokens. Before switching a high-volume bot to a bigger model, estimate monthly volume and compare — a model that is a few times more expensive per token adds up quickly across thousands of conversations and many sub-accounts.
A safe switching checklist for live bots
If an agent is already replying to real leads, don't flip the model and walk away. Use this process instead:
- Write a test script of 10–15 real messages pulled from past conversations: easy questions, pricing questions, an angry customer, a booking request with an awkward time, an off-topic message, and a message that should trigger a human handoff.
- Record how the current model answers so you have a baseline.
- Duplicate the agent (or test in a staging sub-account) and switch the copy to the new model.
- Run the same script and compare: accuracy, tone, length, whether it follows your rules (for example, never quoting prices), and whether booking and handoff actions still fire.
- Adjust the prompt. Claude, Gemini and OpenAI models respond to instructions differently. You may need to tighten rules about reply length, formatting, or what the bot must never say.
- Check your workflow triggers and actions. Make sure anything downstream — tags, pipeline moves, appointment confirmations — still fires correctly.
- Roll out gradually: switch one sub-account or one channel first, review transcripts for a few days, then roll out to the rest.
- Answers common questions accurately
- Keeps your tone and reply length
- Never quotes prices or promises it shouldn't
- Handles an angry customer calmly
- Books an appointment end to end
- Hands off to a human when it should
- Tags and pipeline moves still fire
- Ignores off-topic or trick messages
Common problems after switching models
- Replies got longer or more formal. Add explicit length and tone rules to the prompt (for example, “reply in 1–3 short sentences, casual tone”).
- The bot started answering questions it shouldn't. Restate your guardrails clearly near the top of the instructions and re-test the off-topic and pricing messages.
- Booking stopped completing. Re-test the booking flow end to end; check the calendar connection and the agent's goal settings rather than assuming the model is the cause.
- The model you used disappeared from the list. Providers retire models. Check the dropdown regularly and keep a fallback model set so a retirement doesn't take a bot offline.
What this means for agencies
Multi-model support gives agencies two practical advantages. First, cost control: you can run simple bots on cheaper models and reserve stronger models for the conversations that drive revenue. Second, reliability: a fallback model from a different provider means one provider's outage doesn't silence every client's chatbot at once.
The trade-off is more to manage. Every model change should go through a test script, prompts may need tuning per model, and the Labs setting has to be enabled in every sub-account that needs it. If you're running Conversation AI across many client accounts, a short audit — which bots run on which model, with which fallback, and at what monthly cost — is worth doing now, or you can hire a GHL expert to run that audit across every sub-account.
Frequently asked questions
Can GoHighLevel Conversation AI use Claude or Gemini?
Yes. HighLevel added Anthropic Claude and Google Gemini models to Conversation AI alongside OpenAI. Enable Conversations AI Provider Support under Settings → Labs, then choose the model in the agent's Build tab.
Where do I change the AI model for a Conversation AI bot?
Open AI Agents → Conversation AI, select the agent, go to the Build tab and use the Model dropdown. You can also set a secondary model as a fallback.
What is the secondary model in Conversation AI?
It is a backup model that takes over if the primary model is unavailable. Choosing a fallback from a different provider gives the best protection against outages.
Does switching models change the cost?
Yes. Each model has its own per-token rate, listed on HighLevel's AI Product Pricing page. Estimate your monthly conversation volume before moving a busy bot to a more expensive model.
Do I need to rewrite my prompts when I switch to Claude or Gemini?
Often a little. Different models interpret instructions differently, so re-test with real messages and tighten rules on reply length, tone and restricted topics where needed.
Want this set up in your GoHighLevel account?
Get hands-on help from a GHL expert — scoped per project, no retainer. Start with a free 30-minute call, or see how we handle GoHighLevel AI automation and chatbots.
Book a Free Call →Sources
- HighLevel Support: Use Claude and Gemini Models in Conversation AI
- HighLevel Changelog: Multi-LLM Providers in Conversation AI
- HighLevel Support: AI Products Pricing