HighLevel’s Release Radar for September 28 to October 2, 2026, announced five updates: Muse connectivity through MCP, Voice AI flow generation, Ask AI funnel and website creation, verified-number outbound calling in the mobile app, and Meta campaign budget controls.
This guide turns that release summary into an operating plan. Each capability has a different endpoint: a CRM action, a Voice AI flow, a site asset, a phone call, or an advertising configuration. The original announcement explains what is new; current HighLevel support documentation supplies the setup conditions. Generated work and enabled settings still need an accountable review before production use.
| Capability | Operational endpoint | First owner |
|---|---|---|
| Muse and MCP | Supported CRM read or write | CRM administrator |
| Voice AI Build with AI | Editable call flow | Voice AI owner |
| Ask AI | Draft funnel, website, or page asset | Marketing or site owner |
| Verified mobile caller ID | Supported outbound call | Phone administrator |
| Meta budget controls | Campaign or ad-set configuration | Campaign owner |
Read the October 2 Release Radar announcement, then check the current interface and support documentation before rollout.
What did HighLevel announce for September 28 to October 2, 2026?
The release week combined AI-assisted creation with operational controls. Muse can connect to supported HighLevel CRM operations through an official OAuth connection. Voice AI Build with AI can create or modify a flow graph from natural-language requirements. Ask AI supports assisted creation of funnel, website, and page assets. The mobile app can use an existing verified personal or business number for supported outbound calls. Ad Manager adds campaign-level and ad-set-level Meta budget choices, including an optional sharing setting.
The practical distinction is important. A generated graph is not a tested call path, a generated page is not a working conversion path, and a budget-sharing setting is not evidence that Meta moved money. The operating model below keeps those states separate.
Use a release feature as a controlled operating step
This is a proposed team operating model, not a built-in HighLevel approval feature. It works across the five updates by separating the request, the product action, the review, and the measurable result.
Teams configuring their wider environment can review GoHighLevel setup support. The feature-specific sections below identify the documented product behavior and the team decisions that remain outside it.
Operationalize Muse and HighLevel MCP for CRM work
HighLevel documents an official Muse connection using OAuth and a Muse-specific MCP endpoint: https://services.leadconnectorhq.com/mcp/muse/v2/. During authorization, select the intended CRM sub-account or sub-accounts and review the requested scopes. HighLevel currently describes more than 600 supported operations across 38 CRM domains through four unified MCP tools. This is an aggregate catalog, not a guarantee that every operation is available in every session.
OAuth scopes and HighLevel user permissions are separate controls. Availability depends on both. The general MCP documentation describes supported CRM work such as retrieving records and updating data, but it does not promise unrestricted access to every workflow, calendar, or customer context for every connected user.
- Add the official HighLevel app in Muse or use the documented Muse connection.
- Complete OAuth, select the intended sub-account, and approve only the scopes needed for the task.
- Start with a read-only request that returns record IDs and current values.
- Confirm the target records and proposed change, then issue a separate write instruction if approved.
- Check the result in HighLevel and record the initiating user, sub-account, operation, and outcome in the team’s own audit process.
A useful read-first prompt is: “In sub-account [ID], find contacts matching [criteria]. Return contact IDs and current lifecycle stage. Do not update records, send messages, or trigger workflows.” For a later write, identify approved records and expected current values. If a value has changed since review, skip that record rather than overwriting the newer value.
The following is an illustrative review record, not a Muse response schema. The key combines a source run and operation reference so multiple runs cannot be confused with one another:
{
"source_run_id": "run_2026_10_02_001",
"operation_attempt_id": "op_0007",
"sub_account_id": "sub_account_123",
"record_ids": [
"contact_456"
],
"proposed_changes": [
{
"field": "lifecycle_stage",
"current_value": "new_lead",
"proposed_value": "qualified"
}
],
"approval_status": "pending"
}
For a receiving audit system, enforce uniqueness with a database constraint or transactional upsert around the operation attempt. A lookup followed by insert is not safe when concurrent workers process the same task. The CRM administrator or authorizing user should investigate scope, permission, and changed-value failures before retrying. See CRM systems consulting for broader data ownership and system-of-record planning.
Generate Voice AI flows, then test the actual call path
Build with AI in the Voice AI Flow Builder can create or modify a flow graph from natural-language requirements. HighLevel documents a sequence in which the user describes the call experience and information to collect, answers clarification questions if asked, reviews a build plan, previews the graph, and applies it to the editable canvas.
Give the builder concrete inputs: caller type, information to collect, decision points, transfer destination, and post-call action. Inspect every node and connection after generation. Confirm that collected values map to intended contact fields and complete required action settings. Test each important branch, including transfers and post-call actions, using the actual call path.
HighLevel separately documents predefined Voice AI templates, including lead qualification and post-call automation starting configurations. Templates are not the same workflow as custom Build with AI generation. Required action settings must be completed, and incomplete setup can block testing or deployment. See Build with AI for Voice AI flows and the separate Voice AI templates and guided setup guide.
For a lead qualification flow, test both an urgent answer that should transfer and a non-urgent answer that should continue. Verify the captured service need and callback time on the intended contact, then confirm the post-call action. If a field mapping is missing, correct the configuration before deployment. Teams defining bounded AI responsibilities and review paths can explore AI agent implementation support.
Use Ask AI for a draft funnel or website
HighLevel documents AI-assisted creation under Sites, with Funnels or Websites paths that include Assist and Build options. Depending on the selected path and current product version, the result may be a funnel, website, page, or related asset. Provide the business, audience, goal, offer, page purpose, and design information requested by the selected path, then continue in the visual builder.
After generation, the marketing or site owner should check factual claims, prices, service areas, offer terms, images, forms, calendar links, and consent language. The technical owner should test form submissions and calendar destinations and verify relevant SEO fields, domain, tracking, and publishing settings. These are manual review tasks, not automatic guarantees of the generation workflow.
A practical handoff is to save the asset as a draft, list the intended conversion path and integrations, and have the content owner approve customer-facing claims before the technical owner publishes. If a form fails, fix its destination and repeat the submission test. The Funnels and Websites AI guidance describes the assisted creation paths and builder editing. The separate Funnel Creation Agent documentation describes a particular funnel workflow and is version-sensitive; it should not be treated as proof that every prompt creates a complete multi-page website.
Use a verified number for supported mobile outbound calls
The mobile calling update lets a user select an existing verified personal or business number for supported outbound calls. First add the number in HighLevel and complete its verification call and code. Then open the mobile Call Dialer, choose the caller ID selection, and select the number marked Verified. This is not SIM-based calling and does not convert the number into a HighLevel-managed number.
HighLevel documents HighLevel or LeadConnector mobile app version 4.25.0 or later for this workflow and a default limit of 10 verified caller IDs per sub-account. Confirm the current app version, verify the badge, and place a test call before routine use. Country, carrier, and regulatory restrictions can affect availability. HighLevel specifically documents a restriction for calls from +91 numbers to +91 numbers using this workflow.
The phone administrator owns verification and caller-ID issues. If the number is absent or not marked Verified, resolve verification or eligibility before placing the call. An eligible verified number can be used without first purchasing a HighLevel-managed number, but that does not establish that calling is free or that other phone-system costs disappear. See HighLevel’s verified-number mobile calling instructions.
Choose campaign-level or ad-set-level Meta budgets deliberately
In HighLevel Ad Manager, campaign-level budgeting uses one campaign budget that Meta distributes across ad sets. Ad-set budgeting assigns separate budgets to individual ad sets. With ad-set budgets, the advertiser can optionally enable sharing of up to 20% of a budget with other eligible ad sets. Daily and lifetime budgets are supported, with applicable bid-strategy choices such as Highest volume and Cost per result goal.
- Choose campaign-level budgeting when centralized distribution across ad sets fits the campaign objective.
- Choose ad-set budgeting when separate amounts or tighter per-ad-set control matter.
- Enable sharing only when conditional redistribution is acceptable. Record the budget mode, amount, sharing state, and bid strategy.
- After a budget setup change, allow for HighLevel’s documented possibility that Meta may temporarily restrict further setup changes for approximately two hours.
Enabling sharing permits Meta to redistribute up to 20% under applicable eligibility, performance, and delivery rules. It does not guarantee a transfer or prove one occurred. Store the configuration snapshot separately from observed spend or delivery data.
For distinct audience tests, an advertiser might use ad-set budgets and enable sharing only if some movement between eligible ad sets is acceptable. The campaign owner should review delivery or spend independently from the saved configuration. If the interface rejects a change, check campaign state and allow for the temporary restriction before escalating. See HighLevel’s Meta campaign budget optimizer documentation.
Apply shared controls once, then measure the right outcome
Across these features, use least-necessary permissions, keep the system of record clear, review high-impact writes and publication, and name the exception owner. Measure the actual endpoint: successful and corrected CRM actions, tested Voice AI branches and routing, working form and calendar submissions, completed verified-number call attempts, and campaign delivery or spend observations.
Any logging or synchronization layer should use a row grain that matches what happened. One MCP log row can represent one attempted tool invocation. A Voice AI call and its individual actions should be separate records. A generation run should not be conflated with each generated page. A Meta configuration version is not a spend observation or a redistribution event.
These are proposed implementation patterns, not vendor-published HighLevel schemas. Use stable source IDs where available. If no source ID exists, assign a source run ID and preserve the source response. For concurrent processing, enforce uniqueness with a database constraint or use a transactional upsert. Do not rely on lookup followed by create.
- Name the feature owner and the person who handles exceptions.
- Confirm access, scopes, app version, number status, or campaign eligibility as relevant.
- Save evidence from the actual workflow endpoint, not only a preview or setting.
- Record where approved changes, configuration snapshots, and source responses belong.
- Define the correction path and the measure that shows routine operation is working.
These five updates can reduce setup effort, but dependable use comes from matching each feature to its proper endpoint and checking that endpoint before relying on it. The Release Radar article describes the September 28 to October 2, 2026 announcement window; verify current documentation, interface labels, permissions, and account availability when implementing.
