HighLevel Updates: New Features for September 7-11, 2026

HighLevel leaned hard into artificial intelligence this week, headlined by the ability to reach the Ask AI copilot from WhatsApp. The week also brought new language models for Conversation AI, a real-time voice model in early access, a publish gate that stops broken workflows, and self-service cancellations for rental bookings. Here is what matters for agencies and small businesses, with fuller context in our September 2026 HighLevel updates roundup.

Ask AI Moves Into WhatsApp

Your Ask AI copilot is now reachable over WhatsApp. Connect a personal number once from the Dispatch tab, scan a QR code, and verify with the code that comes back. After that it behaves like any other contact in your chat list.

Nothing is stripped down in the text version. Look up a contact, move a deal stage, check the day's appointments, request an email template, generate an image or audio clip, schedule recurring work such as a Monday lead summary, or run a custom skill. Product questions route to Support AI in the same thread, and every exchange lands in a searchable Dispatch tab inside the app.

Each number stays private to the person who connected it, and a number claimed elsewhere moves after a confirmation. For an owner running between client meetings, that turns dead time into billable progress. Here is how WhatsApp messaging works elsewhere in the platform.

Conversation AI Runs on Newer Models

Agent Builder now offers eight more OpenAI models, spanning premium accuracy, balanced, and low-cost tiers. GPT-4.1 remains the recommended default.

On real production conversations the strongest new model scored 93.0 percent accuracy against 82.4 percent for GPT-4.1, and every new option matched or beat that baseline. Cost per reply ranges from about 2.5 times the old price down to 89 percent cheaper, so you can put an expensive model on a sales agent and a cheap one on a high-volume support bot. Our guide to the AI tools inside HighLevel covers where each one fits.

Mobile Gets Single Reply Suggestions and Flexible Transfers

With AI Employee in Suggestive Mode, each eligible inbound message now produces one ready-to-send reply above the composer instead of several short options. Send it, edit it, dismiss it, or restore the original with a tap. Faster first responses convert more inbound leads.

Call transfers loosened up too. Staff without a personal phone number can receive transferred calls from the mobile app, warm or blind, on inbound and outbound calls. Re-transferring a call you already received is not supported yet.

Voice AI Adds a Real-Time Model and Outbound Calls

A new speech-to-speech model is in early access with a beta label in the model dropdown. It listens and speaks at once instead of waiting for a turn to end, so callers can interrupt and the agent adapts mid-sentence. A second reasoning model handles bookings, lookups, and tool calls in the background while the speech model keeps talking, which removes the dead air that usually follows a real request. Billing is a flat 0.12 US dollars per minute covering both models and starts September 17.

Speech-to-speech agents can also place outbound calls now. Select one in the Voice AI outbound call workflow action, then set the AI disclosure, welcome message, and verbal opt-out in Agent Builder. One agent covers both directions, which keeps outbound campaigns cheap to scale. Compare approaches on our Voice AI overview or check Voice AI pricing.

Workflow Building Gets Safer

Publishing now stops on unresolved configuration errors. Front-end and back-end validations merged into one Highlight Error sidebar, each issue links straight to the step or offers an AI assisted fix, and saving works even with errors on the canvas so you can finish later. Broken automations stop slipping into client accounts unnoticed.

Dynamic value chips went from two states to three. Blue means the workflow supplies the value, yellow means it exists on the platform but nothing here feeds it, and red is reserved for values that do not exist. Yellow warns without blocking, so deliberate cross-workflow references stop looking like errors, and execution is unchanged.

A new Workflow Settings page in Agency Settings sets Standard or Advanced as the default builder and pushes it to every owned sub-account with one toggle. The AI Image Generation action also takes up to five reference images per run from an upload, the media library, or a URL with custom values, so a logo or location can be shown rather than described.

Calendars, Events, and Courses

Rentals customers can cancel their own bookings from the confirmation link, with an optional reason saved to the record. Operators can set a cutoff window in minutes, hours, or days, bundled listings cancel together, and refunds and deposits stay a manual decision. That clears a routine support queue without giving up control of money.

Custom forms on services and rentals now apply post-booking logic, so redirects, custom messages, and lead disqualification can drive the confirmation page once a global checkbox is on. In-person event tickets can be saved to Apple Wallet or Google Wallet with the same check-in code scanned at the door, and a merge field adds wallet links to confirmation messages. Courses can now be sold without creating a separate offer.

Agency Controls and Builder Upgrades

Attaching a snapshot during SaaS enablement used to overwrite matching sub-account assets silently. A warning modal now appears every time and asks whether to overwrite conflicting assets or keep the existing ones, retiring a recurring class of support tickets.

My Profile and user management were rebuilt on the newer design system. Teams and My Staff are both called Users, My Profile splits into focused sections, the signature editor gained a rich-text toolbar and an HTML view, and custom CSS and JavaScript now apply because these pages left the iframe.

The funnels and websites builder added a breadcrumb bar showing the full hierarchy of the selected element, so you can jump to any ancestor in one click. Managed Agents can now read attached documents, up to five per message including form uploads, and produce images about twice as fast on the high quality setting at the same price.

Smaller Improvements Worth Noting

• Missed call text back: when the number that was called cannot send texts, another eligible number from the location is picked automatically.
• Voice notes in Communities: record notes on posts and comments, upload audio, change playback speed, and autoplay the next clip.
• Dark mode: Communities and Courses support dark mode on the web, with separate light and dark brand palettes for owners.
• Marketplace discovery: Ask AI recommends apps from a plain-language description, and agencies control whether sub-accounts see them.
• Ad Manager: Meta post-publish recommendations appear inside published campaigns with impact levels and a link to the section to edit.
• MCP for OpenAI clients: a dedicated endpoint connects compatible assistants to your account data.
• AI Studio: every new project now ships with server rendering, server functions, and a secrets vault.

Frequently Asked Questions

Do I need a separate phone number to chat with Ask AI on WhatsApp?

No. You connect your own number once through the Dispatch tab and verify it with a code. The connection is private to you, and a number connected elsewhere moves across after a confirmation.

Should I move my Conversation AI agents to a new model right away?

There is no forced migration and GPT-4.1 is still the default recommendation. Test one agent, compare answer quality on your own conversations, and weigh the per-reply cost before rolling a change across sub-accounts.

Will the publish gate break my existing workflows?

No. Live workflows keep running as they did and the chip colors are a display change only. The checks apply at publish time, so republishing an older workflow may surface issues that were previously invisible.

Try These Features Yourself

A copilot that answers on WhatsApp reads well, but texting your own client list from it is the real test. Everything above sits inside the standard platform, so you can test it on live accounts today. Start a free 30-day GoHighLevel trial and connect Dispatch in the first hour. Our beta watch page tracks what is still in testing.