GoHighLevel AI Agent Web Search and Test Workflow in AI Builder: What's New and How to Set It Up

Status check: both are listed as new under Automations on HighLevel's changelog, posted on 29 October 2026 (Web search, Test workflow). Test workflow in AI Builder is a Labs feature, so you turn on The New Test Workflow in HighLevel Labs first. This post is based on HighLevel's release notes. Neither entry gives rollout timing, so if you don't see a feature, check the changelog entry or ask HighLevel support.

The two updates at a glance

Web search in the AI Agent Action ANNOUNCED 29 OCT

Prices, programmes and public details change all the time, and HighLevel's point is that the agent can now check them at the moment it needs them. The Web search update adds a new Built-in tools section that sits right below Tools. HighLevel describes built-in tools as abilities the agent has on its own, on top of the tools you add.

• One toggle: turn on Web search and the agent can search the internet during any run.
• The agent decides: it searches only when the task needs information it does not have.
• Your own sources: paste a URL in the Instructions, and the agent refers to that page when the run calls for it.
• Static or live: give the agent the pages you trust, and let it search the web for everything else.
• Hover help: hover the (i) on a built-in tool to see what it does, and hover the crown to see what it costs.

How to turn on Web search (steps from HighLevel)

1. Go to Automation > Workflows.
2. Open a workflow, then add or open an AI Agent Action.
3. Scroll down to Built-in tools and expand it.
4. Turn on Web search.
5. To point the agent to specific pages, paste their URLs in the Instructions. This step is optional.
6. Save the action.

What it costs

HighLevel lists Web search as a premium tool, billed per search on top of the AI Agent Action. In HighLevel's words: "OpenAI and Claude models: 1¢ per search." and "Gemini models: 1.4¢ per search." A run with three searches is billed for three searches, and a run that needs no search costs nothing extra. Check the crown icon in your account for the current price.

HighLevel's example: rebate questions answered from the source

A heating and cooling company follows up on every quote request, and rebate programmes change often. The AI Agent Action has Send email and Send SMS as tools, Web search turned on, and Instructions that say to read a named rebate finder page before replying about rebates, and to search the web for any other facts. When a lead asks about heat pump rebates, the agent checks that page first. When a lead asks something the page doesn't cover, the agent searches the web.

Practical tips

• Name the trusted pages first. Put the client's own pricing, service area or policy pages in the Instructions, so the agent reads those before it searches anywhere else.
• Say when to search. Tell the agent in plain words which questions need a lookup (rebates, opening hours of a partner, public rules) and which should go to a human instead.
• Keep the tools tight. Web search only finds information. The agent still needs Send email or Send SMS to act on it, so add only the tools that run actually needs.
• Watch the spend. Because each search is charged, test on your own contact first and check how many searches a typical run makes before switching it on for a busy workflow.

HighLevel also says AI image generation and AI video generation, which are live as separate actions, are coming to the AI Agent Action as built-in tools very soon. If you want the video side now, my AI Video Generation action guide covers the separate action.

Test your workflow right inside AI Builder Labs, opt-in

Labs feature, opt-in, and tests run real steps. You turn it on yourself in HighLevel Labs (The New Test Workflow). HighLevel says a test runs the real steps: tags are added, notes are written and messages are sent. Test with your own contact, not a customer.

AI Builder builds and edits workflows from a description. The Test workflow update adds testing to the same chat: every time AI Builder builds or edits your workflow, its reply ends with a Test workflow button, and the test opens in the same chat.

• Two ways to test: choose Enroll a contact or Test using events, pick the contact and check the contact data before you start.
• Watch the run live: after Start test run, the canvas shows each step as Running, then Success, Action skipped or End.
• Step list in the chat: the chat lists the run step by step, with a check next to every step that ran.
• A summary you can read: AI Builder writes a short test summary covering which steps ran, which were skipped, and whether there were errors or missing data.
• Input and Output per step: click any step on the canvas to see what went in and what came out.
• Run another test: changed something? One click runs the test again.

How to test a workflow in AI Builder (steps from HighLevel)

1. Open HighLevel Labs and turn on The New Test Workflow.
2. Go to Automation > Workflows and open a workflow.
3. Open AI Builder and describe the workflow you want, or the change you want.
4. When AI Builder finishes, click Test workflow in its reply.
5. Choose Enroll a contact and pick a contact, or choose Test using events.
6. Click Start test run.
7. Read the test summary. Click any step on the canvas to see its Input and Output.
8. Save and publish your workflow.

Good to know

• Some actions are skipped in a test run. The summary and the canvas tell you which ones.
• Testing is optional. You can save and publish without testing.

Practical tips

Keep a dedicated test contact in each sub-account with your own email and phone, and use it for every AI Builder test. After AI Builder makes a change, run the test, read the summary for skipped steps or missing data, open the Input and Output of any step that looks wrong, ask AI Builder to fix it in the same chat, then click Run another test before publishing. For bigger workflows, my AI Builder guide covers how it uses your real forms, pipelines and tags, and my Workflow Maps guide covers finding errors once a workflow is live.

Putting the two together

These fit in one flow: describe an AI Agent follow-up in AI Builder, turn on Web search in its Built-in tools with your trusted pages in the Instructions, then use Test workflow on your own contact and open the AI Agent step's Input and Output. Remember the test sends real messages to that contact. For the other workflow updates from the same day, see my Loop, Router and AI Agent Skills guide. If you would rather build and debug workflows from Claude, my guide to connecting Claude to HighLevel's official MCP server covers the workflows private beta.