GoHighLevel AI Agent Skills and Web Search: How to Set Them Up

Short answer: GoHighLevel's AI Agent Action in Workflows now has two new options (Official, HighLevel changelog, October 29, 2026). Skills are reusable instructions the agent loads only when a run needs them, with four built-in skills (Humanizer, Write-well, Cold-email, Sms-writer) and your own custom ones. Web search is a built-in tool that lets the agent look things up online during a run. Skills have no stated extra cost. Web search is billed per search: 1 cent with OpenAI and Claude models, 1.4 cents with Gemini models. Both are set up inside the AI Agent Action under Automation, Workflows.

What Skills are in the AI Agent Action

Until now, everything you wanted the AI Agent Action to know had to live in one long Instructions box. If you wanted the same writing rules in ten workflows, you pasted them ten times, and when you changed your mind you edited ten workflows. Skills fix that. According to HighLevel's changelog (Official), a skill is a reusable procedure that you write once, keep in a library, and attach to any AI Agent Action.

The important detail is how skills load. On each run, the agent only reads each attached skill's name and its "When should the agent use this?" line. The full instructions load only when the run matches that description. If several skills match, they load together. That keeps the agent's working context small and means you can attach a few specialist skills without every message turning into a mix of all of them.

Every account starts with four built-in skills, tagged Built-in, which you can attach, edit or delete like any other:

• Humanizer: removes the usual signs of AI writing from a message.
• Write-well: keeps messages clear, simple and direct.
• Cold-email: writes first-touch and follow-up emails to leads who have not replied.
• Sms-writer: writes short, human-sounding text messages.

You can also make your own skill in three ways: Create with AI (the AI Assistant asks what the skill should do, when it should fire and what it must never do, then drafts it; nothing saves until you click Create), Write skill instructions (Name, "When should the agent use this?" and Instructions, with an Enhance Prompt button), or Add existing skill from your library.

How to add a skill, step by step

1. Go to Automation, then Workflows, and open the workflow.
2. Add a new AI Agent Action or open an existing one.
3. Scroll down to Skills and expand it.
4. Click + Add skill and pick Create with AI, Write skill instructions or Add existing skill.
5. Review the skill, then click Create & attach or Attach.
6. Save the action, then test the workflow on your own contact before real leads hit it.

Three rules from the changelog are worth knowing before you build a library:

• Skills are shared. Editing a skill changes it in every action that uses it. That is the point, but it also means a careless edit reaches every client workflow at once.
• Detach removes a skill from one action. Delete removes it from your library and from every action that uses it.
• A skill describes how to do something; it does not give the agent the ability to do it. The action still needs the right tools, such as Send SMS or Send email. A skill can point at a knowledge base if you name it in the instructions and add the Knowledge base search tool to the action.

HighLevel's own example is a heating and cooling company that attaches Cold-email, Sms-writer and a custom "Emergency service requests" skill. Routine quote requests get a friendly email and text. Leads reporting no heat or a gas smell get a same-day slot or safety steps first, and no price.

What web search adds, and what it costs

Web search is a new built-in tool in the same AI Agent Action (Official, changelog, October 29, 2026). When it is on, the agent can search the internet during a run, and it only searches when a task needs facts it does not already have. You can also paste specific page URLs into the Instructions, and the agent reads those pages when a run calls for them. That lets you mix trusted pages, like your own rebate or service area page, with live search for everything else.

To turn it on: open the AI Agent Action, scroll to Built-in tools, expand it, switch on Web search, optionally paste trusted URLs into the Instructions, and save.

Web search is a premium tool billed per search, on top of the AI Agent Action itself:

• OpenAI and Claude models: 1 cent per search.
• Gemini models: 1.4 cents per search.

Each search is billed separately, so a run with three searches costs three searches. A run that needs no search adds nothing. HighLevel has not stated a cap on searches per run, so if a workflow fires hundreds of times a day, keep an eye on usage in the first week. The changelog also says AI image generation and AI video generation are coming as built-in tools in the AI Agent Action; today they are separate workflow actions, covered in our guide to GoHighLevel AI video generation.

Where this is useful for a service business

For most local service businesses, Skills matter more than web search. The common problem we see is not that the AI lacks facts; it is that it writes in a voice that does not sound like the business, or it quotes prices it should not. A small set of skills solves that once:

• A brand voice skill with your tone, words you never use and sign-off.
• A never quote skill that tells the agent to offer a visit or a call instead of a number.
• An urgent request skill for emergencies, with the exact steps and who gets notified.

Web search earns its cost in narrower cases: answering questions about public rebates, permit rules or product specs, or checking a business before a B2B follow-up. For anything about your own services, prices and policies, a knowledge base is the safer source, because the web may say something different from what you actually offer.

Agencies get the most from Skills. One tested set of skills can sit behind every client's AI Agent Actions, and a fix to the tone or a compliance rule happens in one place instead of forty workflows. Pair that with the new Router workflow action to send each lead down the right branch before the agent writes anything.

Mistakes to avoid

• Vague "when to use" lines. The agent decides from that one line. "Use for emails" is too broad; "Use when the lead has not replied to the first email after 2 days" is clear.
• Editing a shared skill live. Duplicate it, test the copy on one workflow, then switch.
• Forgetting the tools. A Sms-writer skill with no Send SMS tool writes a text that goes nowhere.
• Letting web search answer pricing questions. Tell the agent in a skill to use your knowledge base for anything about your own business.
• Skipping tests. Run the workflow on your own contact and read every message before turning it on for real leads. If messages already go out wrong, our guide on why a workflow can send the wrong message walks through the checks.

What to do now

Start with the two built-in skills that fit your channel (Sms-writer for text, Write-well or Cold-email for email), add one custom skill with your never-do rules, and test on one workflow. Only switch on web search where a question really needs outside facts. For the full list of what HighLevel shipped this week, see our LevelUp 2026 announcements hub.

If you want this set up and tested for you, our Speed-to-Lead Sprint builds the follow-up workflows, and the GoHighLevel account audit finds where AI messages are going wrong today. Or book a free call at meet.yotomations.com.

Yaakov (Koby) Oranski, Founder and CEO of Yotomations

Certified GoHighLevel Admin and Partner. Koby designs CRM migrations, two-way integrations, billing systems and AI agents for service businesses, and writes from the engagements described on this site.

Published 2026-10-30