GoHighLevel AI Agent Action: Claude, Gemini and GPT Guide

On August 20, 2026, HighLevel quietly changed one of the most-used steps in the Workflow Builder. The action that used to be called GPT Powered by OpenAI – the one agencies drop into a workflow to draft a reply, classify a lead, or summarize a conversation – now runs models from Anthropic and Google as well, and it ships with a reasoning-effort dial that didn’t exist before. HighLevel’s changelog refers to the updated step as the AI Agent action. If you built automations on the old OpenAI-only version, here is exactly what’s different and how to decide which model actually belongs in which workflow.

This isn’t a UI refresh you can ignore. The model you pick now changes both the quality of what the action outputs and what it costs you every time a contact runs through that step – and with three providers and a three-level effort dial, the number of reasonable combinations went from “pick a GPT version” to something worth actually thinking through.

What the AI Agent Action Does

Inside the Workflow Builder, this action takes a prompt you write – plain instructions, optionally loaded with dynamic variables like {{contact.first_name}} or a custom field pulled from earlier in the workflow – and sends it to the model you’ve selected. What comes back can be written into a Custom Field, dropped straight into the next Email or SMS action, or logged out through an internal notification, Slack message or webhook. Nothing about that mechanic changed. What changed is which model does the thinking, and how hard you tell it to think.

The Model Lineup by Provider

As of the August 20 changelog entry, the model dropdown is organized by provider, with each option carrying a short description and a “thinking chip” on the ones that support reasoning effort:

One thing worth flagging before you build around this list: HighLevel’s own dedicated support article for this action (still titled “GPT Powered by OpenAI” at the time of writing) shows a different OpenAI lineup – GPT-5, GPT-5.1, GPT-5 Mini, GPT-5 Nano, GPT-4o, GPT-4 Turbo and GPT-4o Mini. The two HighLevel sources don’t agree with each other, which is presumably a documentation lag rather than two different feature sets, but it means the model names you actually see in your own account’s dropdown are the ones to trust, not either published list. Confirm what’s live before you name a specific model in an internal SOP.

Reasoning Effort: Low, Medium, High

The other new control sits next to the model dropdown: a Low / Medium / High effort setting, available on models that support it. HighLevel’s own description is straightforward – higher effort produces deeper reasoning on complex tasks, lower effort returns faster responses on simpler ones. In practice this is the same idea as temperature, but aimed at a different axis. Temperature controls how creative or conservative the wording is; reasoning effort controls how much internal work the model does before it answers at all.

That distinction matters for how you configure a step. A workflow that classifies an inbound SMS as “hot lead” or “spam” doesn’t need the model to reason deeply – it needs a fast, cheap, consistent answer, so Low effort on a smaller model is the right call. A workflow that drafts a first-touch proposal email from a messy intake form benefits from High effort, because getting the structure and the specifics right is worth the extra latency.

Configuring the Action Step by Step

• Open the AI Agent action inside your workflow and click the model dropdown.
• Pick a provider, then a model – options are grouped so you’re choosing Anthropic, Google or OpenAI first, model second.
• If the model shows a thinking chip, set the reasoning effort to Low, Medium or High.
• Write the prompt, including any dynamic variables the workflow has available at that step.
• Set where the output goes – a Custom Field, directly into a following Email or SMS action, or out to Slack, an internal notification, or a webhook.
• Save and publish, then run a test contact through the workflow before trusting it on live traffic.

The advanced Temperature control from the pre-update version of this action isn’t mentioned anywhere in the August 20 changelog entry, so whether it still sits underneath the reasoning-effort setting or was replaced by it is not something either HighLevel source confirms – check your own account’s advanced section rather than assuming either way.

Picking a Model for the Job, Not by Habit

The honest failure mode with a longer model list isn’t picking a bad model – it’s leaving every workflow on whatever the default was when you first built it. A few practical starting points, based on what each tier is actually built for rather than a benchmark:

This isn’t a claim about which provider writes better copy – that’s genuinely subjective and depends on your prompt and your niche. It’s a claim about matching effort to the actual difficulty of the task, which is the lever HighLevel just handed you that didn’t exist before.

Where This Sits Next to HighLevel’s Other AI Actions

Agencies building anything past a single message already have AI Decision Maker and AI Intent Detection in the same action list, and it’s worth being clear on the boundary between them so you don’t reach for the wrong one. AI Intent Detection reads a message and classifies sentiment or intent – it’s a narrow, single-purpose read. AI Decision Maker routes a contact down different workflow branches based on plain-English instructions you give it, weighing signals like engagement or company size – it decides where the contact goes next, but it doesn’t generate content. The AI Agent action is the generation step: it’s the one that writes the email draft, the summary, or the classification label those other two actions then act on. In a workflow that needs to both decide a branch and draft a message, it’s common to use AI Decision Maker for the branch and the AI Agent action, on its own model and effort setting, for the copy – rather than trying to get one action to do both jobs.

The Cost Angle

HighLevel’s changelog entry states that, combined with earlier token-optimization work, the same task now runs “up to 50% cheaper” under the new setup. That’s HighLevel’s own figure, not an independent benchmark, and it isn’t broken out by provider or model tier – treat it as a directional claim worth testing against your own AI-usage billing rather than a number to plan a client price around.

What to Check in Your Own Account Before You Rely on This

Two things neither HighLevel source confirms, and that are worth five minutes in your own sub-account before you rebuild anything: whether a workflow already running the old OpenAI-only version of this action needs to be re-saved to pick up the new model list, and whether the model names shown to you match either list in this article or have shifted again since. Both are the kind of detail that changes fast on a feature four days old – verify in-account rather than trusting a screenshot from launch week, including this one.

Frequently Asked Questions

Is this a new action or an update to the existing OpenAI action?

It’s an update to the existing action. The action agencies already had in their workflows – previously OpenAI-only – now supports Anthropic and Google models from the same dropdown, and HighLevel’s changelog refers to it as the AI Agent action.

Do I need to rebuild workflows that already use the old action?

Neither HighLevel source confirms this either way. Open an existing workflow using this action and check whether the new model dropdown and effort setting appear, rather than assuming a rebuild is or isn’t required.

Does a higher reasoning effort always produce a better result?

Not for every task. HighLevel describes High effort as producing deeper reasoning on complex tasks and Low effort as returning faster responses on simpler ones – for a straightforward classification step, a lower effort setting is usually the more efficient choice, not a worse one.

Which provider should an agency default to?

There’s no single correct default across every use case. Match the model tier to how much reasoning the specific step actually needs, and treat provider choice as a per-workflow decision rather than an account-wide setting.

Does this change what the action costs per run?

HighLevel’s changelog claims up to 50% lower cost for the same task when combined with earlier optimizations, but does not break that figure out by model or provider. Check your own AI-usage billing after switching a workflow rather than assuming the saving applies uniformly.

We build your GoHighLevel account for you

CRM, pipelines, funnels, SMS and automation — configured for your business, not left for you to figure out.

• Setup (one-time)$850
• Ongoing (per month)$38.80

Get my free setup quote

No obligation quote · reply within 1 business day

This site is independently operated and is not officially affiliated with, endorsed by, or sponsored by HighLevel LLC. We do not receive referral commissions — this is our own flat-fee implementation service.