HighLevel Updates: September 2026 Roundup

September 2026 closed with 68 releases. The month opened with workflow validation and AI image actions, and it has gone somewhere else entirely since. The full Ask AI copilot is reachable from a WhatsApp thread, the Workflow AI builder answers from that same thread, seven new permissions decide who is allowed to touch WhatsApp at all, and AI Studio is now generally available with its own permission gate to unlock. In the last week the workflow AI actions moved to GPT-6 at half the previous cost, Communities gained automated moderation, every general widget on a sub-account dashboard became editable with filter conditions that stick, Ask AI on mobile learned to hold a spoken conversation and invoices gained an AI layout builder. On September 29 Company and Custom Object Smart Lists became snapshot assets and Ad Manager gained full control of Meta campaign budgets. Underneath all of it, the CRM quietly learned to create and match company records on its own, and two separate price lists change in October.

September is closed and this page is the final record of the month, September 1 to September 30. The heavy late-August wave is documented in full in the August roundup.

Ask AI moved out of the app, into WhatsApp and into voice

On September 18 the copilot stopped being something you open. Connect your WhatsApp number once and the full Ask AI copilot, the same one that runs inside the platform, is reachable from a text. HighLevel is explicit that this is not a trimmed-down mobile version or a notification relay. If Ask AI can do it on the web, it can do it from a message.

• It runs your CRM without you opening your CRM. Look up a contact, add a note, update a field, move a deal to the next stage, or ask what today's calendar looks like. A client mentions something on a call and you log it before hanging up
• It produces real work. Ask for a blog draft and a subagent returns one in the thread. Ask for a follow-up email and you get a reusable template. Images, video and audio are generated in the conversation
• It schedules itself. Say every Monday at 9am summarize new leads and it runs from then on. Results wait in the Scheduled Tasks tab. A one-line message sets a follow-up check days out
• Custom skills travel with it. Anything you have taught your copilot behaves identically over WhatsApp, and product questions route to Support AI inside the same thread

The Workflow AI builder arrived in Ask AI on September 17, which is what makes the WhatsApp move more than a convenience. Describe the automation you want and the agent builds it. The generated workflow renders as a live preview in the Ask AI panel, and clicking any trigger or action shows its full configuration. Edits are conversational. A CTA opens the result in the workflow builder. Because Ask AI runs on WhatsApp, a workflow can now be created from a text message. Two limits are worth knowing: manual editing is not supported inside Ask AI, and publishing always happens in the builder after a final test. Drafting moved; the review and ship path did not.

Granular WhatsApp permissions landed the same week. WhatsApp is now its own module under Settings, Users, Edit, Roles and permissions, at both agency and sub-account level, with seven per-action permissions: create and edit templates, delete templates, create and edit flows, delete or deprecate flows, update business profile, add and manage phone numbers, and disconnect the integration. The template one matters more than it reads. An edited template goes back to Meta for review, so one unplanned edit can pull a working template out of service. All seven are granted by default to existing roles, so nothing breaks until you decide to tighten it, and the rules are enforced behind the interface rather than only hiding buttons. Our WhatsApp guide covers how the channel is set up.

On September 25 Ask AI in the mobile app learned to talk. Tap the mic and speak. Your words transcribe live in a bubble, then settle into a final transcript with a duration label. The reply comes back as audio through the speaker or a headset, with the text highlighting in time with the voice so you can read along. Start talking while the assistant is speaking and it stops immediately to listen, so a correction no longer means waiting out the answer. When it shows a question or an approval card the mic pauses on its own, then picks back up once you have tapped. A dropped connection shows a reconnecting pill and resumes the same conversation instead of starting over, and closing the Ask AI sheet keeps voice live behind the floating button. AirPods and Bluetooth headsets are picked up automatically, with echo cancellation so the assistant never hears itself. Voice and typed messages sit in one thread with one history, so you can switch between them mid-conversation. The web version already had voice. Mobile is where it earns its place, because the reason to speak to your CRM is that your hands are busy.

October brings two pricing resets

HighLevel is removing the 5 percent processing fee on every WhatsApp message. Marketing, utility and service, in every market, from October 1. You pay Meta's published rate to the cent, which means a client-facing quote can now be proved against a public rate card instead of explained. Per marketing message the United States and Canada go from $0.0263 to $0.0250, the United Kingdom from $0.0667 to $0.0635, Brazil from $0.0656 to $0.0625 and India from $0.0124 to $0.0118. Nothing needs enabling. If you resell WhatsApp, the rebilling multiplier sits under Agency dashboard, Reselling, WhatsApp, Manage, and your client price moves with the cost unless you reset it.

Two categories that were free become billable on the same day. Service messages, meaning any free-form reply inside an open 24 hour customer service window, have not been charged since November 2024. From October 1 they are billed per message, with the first 1,000 per phone number free each calendar month. That allowance is tracked per WhatsApp number, not per sub-account or contact, and unused volume does not roll over. In-window utility messages, free since July 2025, also become billable. Read those two together with the first section: manual Inbox replies, workflow replies and Conversation AI replies all count as service messages. An AI agent answering WhatsApp around the clock is now a line item past the first thousand messages on that number.

• Free Entry Point windows are untouched. A conversation started from a Click to WhatsApp ad or a Facebook Page call-to-action button still opens a 72 hour window in which every message type is free, including the service and utility messages that become billable elsewhere. The window starts when you reply. The 24 hour customer service window runs independently, so once it closes you can only send templates, though those stay free inside the Free Entry Point period. Routing first touch through Click to WhatsApp ads is the clearest lever against the October 1 change
• SMS carrier fees rise on the same date. Outbound SMS and RCS to Verizon numbers move from $0.0045 to $0.0050 per message, and the Other carrier category from $0.0040 to $0.0042. Short Code Group A stays at $0.0025. At 100,000 monthly Verizon messages that is $50, so this is a repricing exercise rather than a strategy change. Plan-level costs are covered in the pricing guide
• Toll-free verification now requires a Privacy Policy URL and a Terms and Conditions URL on Step 3 of the form. Carriers have rejected submissions missing either link since September 15. Numbers verified before that date need nothing. If the funnel behind the number has no published policy pages, build them before submitting, because a rejection costs days
• A second price list changes on October 13. Announced on September 21, it covers international outbound SMS to twelve destinations and monthly rentals on phone numbers in eleven countries. The messaging moves are small in absolute terms and large in percentage. Mongolia goes from $0.2957 to $0.3541 per message, the Maldives from $0.3581 to $0.4047, Burkina Faso from $0.2233 to $0.2568 and Nicaragua from $0.1567 to $0.1856, with Albania, Moldova, Congo, Guam and the Northern Mariana Islands, Angola, Mali, Montenegro and Nepal also rising. Number rentals are the sharper change. A Hong Kong local number goes from $6.00 to $14.70 a month, a Philippines mobile number from $120.00 to $142.00, a Norway mobile number from $12.00 to $16.40 and a Colombia local number from $14.00 to $17.50, with Austria, Belgium, Chile, Finland, Indonesia, Switzerland and Thailand moving too. HighLevel attributes all of it to underlying carrier costs. Nothing needs doing and an active fixed-price agreement keeps its contracted rate for its term, but if you are holding unused numbers in any of those countries the rental is worth auditing before the 13th

Workflows: GPT-6 models, image actions, and a publish gate

GPT-6 Sol and GPT-6 Luna replaced the GPT-5.6 models on September 23. Both appear in the Model dropdown of the AI Agent action and the GPT action. Sol is the model for complex work: multi-step agent tasks, processes that span several apps, and decisions that need careful reasoning. HighLevel reports it makes roughly half as many factual mistakes as GPT-5.6 Sol. Luna is the fast one for classification, summaries, tagging and quick replies, and at higher effort settings it matches GPT-5.6 Sol on factual accuracy at a fraction of the cost. Pricing on both is half what the GPT-5.6 models cost, and on OpenAI's end-to-end workflow benchmarks Luna returns better results at up to 58 percent lower cost per task. The accuracy gain matters most where one AI step feeds another, because a wrong tag or a misrouted contact propagates. The price cut matters more strategically: AI steps that were too expensive to run on every contact are worth pricing again.

AI Image Generation shipped on September 7 and looked different by September 10. The action sits under AI Actions. Give it a prompt, pick a model, and it creates the image, stores it in Media Storage at a public URL and returns that URL as a variable for later steps. The prompt can be typed or pulled from a custom value, so a contact field, a webhook payload or an earlier action can drive it. One workflow can produce a different image for every contact.

Reference images arrived three days later and change what the action is good for. Attach up to five images to a single action as visual context alongside the prompt, by upload, from the Media Library, or by URL. The URL field accepts custom values, so the reference itself can vary per contact. Images pass in the order you add them and the prompt can address each by position. HighLevel's own example uses three: a setting, a portrait as the subject, and a brand logo, with the prompt assigning each a job. A face, a room or an exact logo never survives a text description. GPT Image 2.5 landed the same day in two variants, Flare as the default with higher quality than GPT Image 2 at up to 50 percent lower latency, and Sunburst for premium work needing tighter control across edits. Note that output URLs are public and open without a login, and both image actions are premium with a per-execution charge that varies by model, quality and size.

AI Analyze Image is the reverse operation, also from September 7. Give it a public image URL and a prompt and a vision model returns text. It reads receipts, invoices, screenshots and documents, or describes what is in a picture. Two models are offered, GPT-5.6 Luna and GPT-5.6 Tera, with a detail level of Auto, High or Low, where High is more accurate on small text. Output drops into a custom field, an email, an SMS, a webhook or another AI action. Our AI feature guide covers where these sit alongside the AI Agent action.

Publish became a validation gate on September 11. Hit Publish with unresolved configuration issues and a modal lists every one instead of letting the workflow through. Review Issues opens a Resolve Errors sidebar, and each error offers Open Step, which jumps to the trigger or action at fault, or an AI-assisted fix. Frontend and backend validations that used to surface in different places with different wording are merged into one sidebar. Saving is a separate concern now: auto save and manual save both work with errors on the canvas, where the old behavior blocked the save and fired repeated error toasts. You can stop halfway, keep the work, and return to the unresolved steps. Only the publish is gated.

• Dynamic value chips gained a third state on September 4 and it pairs with the publish gate. Blue means the workflow can fill the value. Red means nothing on the platform matches it. Yellow covers values that are real on the platform but have no source in the workflow you have open. Yellow warns and never blocks a save. The distinction is between a value that will not render at all and one that renders blank, which is the failure that used to reach a live email silently
• Jotform became a native workflow integration on September 20. Two triggers, New Submission and New Signed Document, the second checked every five minutes. One action, Create Submission, writes a record into any Jotform form with values mapped from the workflow. That covers the two directions people were paying a third-party connector for: acting on a submission or a signed contract the moment it lands, and pushing phone and walk-in sign-ups back into Jotform so form reports show every lead. Connect from the builder when prompted, or under Settings, Integrations, Jotform
• Video testimonials can now start a workflow. From September 18 a Video Testimonial Received trigger fires on submission through a collector, filterable by marketing consent or by specific collector. Reviewer name, email, phone, consent, submission time, collector name and video link are all available as custom values inside notifications, notes, emails, SMS and webhooks. Reputation widgets also gained a publishing choice: auto-publish new video testimonials, which stays the default, or require manual approval first

AI beyond the workflow builder

Working Hours for Conversation AI shipped on September 7. Each agent gets its own schedule with different hours per day and multiple slots for breaks or split shifts, and you choose which channels follow it. SMS, email, WhatsApp, Instagram, Facebook, chat widget, live chat and TikTok can each be in or out, and unselected channels keep answering around the clock. The schedule runs on the business timezone or each contact's local timezone. A conversation that started inside working hours can continue after closing. Anyone who messages outside hours can get an off-hours auto-reply, be tagged for follow-up and trigger an email notification, with an automatic follow-up when hours resume. It is optional and off by default. Given the WhatsApp service-message charge arriving on October 1, capping agent hours on that channel is now a cost decision as well as a service one.

Ask AI can find marketplace apps. Since September 10 you can describe what you need, an IDX integration for example, and Ask AI searches the App Marketplace and returns the top matches in the conversation with name, category and capabilities. Agencies keep control: App Marketplace, Settings, Marketplace Visibility carries an Ask AI Discoverability toggle for sub-account users. Turn it off and sub-account admins get no app recommendations through Ask AI, while agency admins keep discovery either way.

AI Studio became a full-stack builder on September 8. New projects run on TanStack Start instead of a static single-page app. Every page is server rendered on every request, so search engines and social unfurls receive real HTML rather than an empty root element. Server functions run backend logic the browser never sees, REST endpoints under /api give the app a public HTTP surface for webhooks, and a secrets vault holds API keys out of the client bundle. Existing Vite and React projects keep running unchanged, with a migration guide promised later. If you have been prerendering AI Studio pages to get them indexed, that workaround is no longer the plan.

AI Studio left Labs on September 22 and is now live for every location. Access does not come with it. Sub-account user and admin permissions for AI Studio are off by default, and only people who already had it enabled in Labs keep it. A banner on the agency Home points at Configure, where you search for users and grant access to everyone or to named people, and setup finishes in a few minutes. Agency owners, admins and agency users with write access see the banner. Sub-account users and read-only agency users do not, so nobody in a sub-account will tell you the box is unticked. If AI Studio is part of what you sell, this is a five minute job worth doing before a client opens the tab and finds nothing there.

Existing React projects can now move to the new stack. The server rendering, server functions and secrets that shipped on September 8 were for new projects only. Project Settings now carries a Migrate to SSR action, marked Advanced, on React and Vite projects you have edit rights to. Confirming creates a second project in the same sub-account, named after the original with TanStack appended, and runs the upgrade there. Your pages come across as a draft. The published site keeps its domain and its version history on the original project, and publish, code save, version restore, domain changes, slug changes and remix are all blocked on the copy until the migration finishes. Projects already on TanStack do not get the action, and a React project stays on React until someone runs it.

Invoices got an AI layout builder on September 28. Describe the invoice you want, a two column layout with the logo top left, an itemized table in the middle and payment terms at the bottom, and Layout AI builds it. You can also upload or describe an invoice you already send and have it rebuilt, instead of starting from a blank template. Keep prompting for changes, or open the layout in the visual builder and move, align, add or delete elements by hand. Every AI change saves a restore point, so you can go back to an earlier version from the chat. Layouts live in their own list where you preview, search, sort, clone, download and rename them, with a history of who created and edited each one. Publishing applies the layout to the invoices your customers see. If you bill several clients with different branding out of one sub-account, that is now a layout choice rather than a code job.

The CRM starts organizing itself

Companies can now create and match themselves from an email domain. Enabled through Labs on September 17, the feature checks for a suitable existing Company whenever a Contact is created or their email changes. A contact at acme.com is associated with the Company that already holds acme.com as its domain. If an Acme record exists with no domain stored, HighLevel populates the domain and reuses that record rather than creating a duplicate. If no suitable Company exists, one is created and enriched with company name and logo. A new Company Domain Name field enforces uniqueness by default, free email providers are excluded out of the box, and you can add your own domains to the exclusion list. Configuration lives under Settings, Objects, Companies, Automations.

That matters beyond tidiness. Company-level Smart Lists, workflows and reporting are only as good as the association behind them, and until now that association depended on someone remembering to make it. The point of the feature is that existing Companies carrying contacts, opportunities and notes are preserved rather than shadowed by a duplicate. Our CRM overview covers how contacts, companies and custom objects fit together.

• Bulk email now starts from Companies and Custom Objects. Since September 17 you can select Company, Property, Policy or other Custom Object records, pick which associated contacts should receive the message, and send without rebuilding the audience in the Contacts list. Choose the most recently associated contact, the earliest, or all of them, and contacts reached through multiple association labels are de-duplicated before sending. Company and Custom Object fields work as merge fields. Send now, schedule, or batch, and track results under Bulk Actions. Filter policies by renewal date and email the holders. Filter companies by industry or account tier and run the B2B campaign from there
• Four new custom field types shipped on September 16. URL fields with domain controls, where you allow all domains, allow only specific ones, or block specific ones. Date and time fields in three shapes, date and time, date only or time only, restrictable to future dates, past dates or a fixed range. Users fields that attach one or more team members to a record and grant them follower-level access. And radio button options that carry an image or icon alongside each choice
• Tag colors can be changed in bulk. From September 20 you can select up to 100 tags on the Tags management page and apply one color in a single action, instead of editing them one at a time. Categories can also be created from inside the category selector, so assigning a tag to a category that does not exist yet no longer means leaving the flow
• Field options can finally be deleted on Companies and Custom Objects. From September 29 you can remove options from dropdown, multi-select, radio and checkbox fields on Custom Objects and Companies, which Contact and Opportunity fields already allowed. Unused options delete outright. An option already in use is removed from future selection while existing records keep the value they have, and the confirmation step shows how many records are affected before you commit. Existing filters, Smart Lists and workflows keep working, and a deleted option simply stops being selectable for new records, filters and workflow conditions. Cleaning up a field someone set up badly eighteen months ago no longer means rewriting history

Agencies: two brands, safer snapshots, tighter permissions

SaaS Multi-Brands went live on September 1 for agencies on the $497 Pro plan. One SaaS account can run up to two independent brand identities side by side, each with its own white label domain, API domain, logo, brand name, brand email and phone, theme, custom CSS and JS, and privacy and terms URLs. Brands map to individual sub-accounts directly, or get assigned automatically by SaaS plan category, with central switching and fallbacks. For anyone selling into two niches under different names, this removes the duplicate-agency-account workaround. Our white-label overview covers what the Pro plan includes.

• Opportunity Smart Lists and Smart Tags became snapshot assets on September 23. Both are selectable when you create or push a snapshot, so a pipeline view built once in a template sub-account arrives intact everywhere else. Smart List configuration carries its filters, sort order, manage-fields setup, list or board type and pipeline visibility. Smart Tags carry their colour coding and linked pipeline. Conflict handling on load stops a push into a sub-account that already has them from creating duplicates. One catch worth respecting: both depend on pipelines, so the related pipelines have to travel in the same push or nothing gets created at the destination
• Company and Custom Object Smart Lists joined them on September 29. Both are now selectable when you create or refresh an Account Snapshot under Agency View, Account Snapshots, so the segmentation you built once travels to every destination sub-account. Companies filtered by industry or account tier, properties by city or price band, policies approaching renewal, memberships nearing expiry: the filter set arrives already built. Refreshing an existing snapshot picks up Smart Lists created since it was last saved, so this does not mean rebuilding your snapshot library. The same dependency applies as with opportunities. Include the underlying object and field configuration the Smart List filters on, or the list lands at the destination with nothing to work with. Test the Smart List in the source sub-account first, because a snapshot carries a broken filter just as faithfully as a good one
• Snapshots stop overwriting your work silently. From September 10, enabling SaaS on a sub-account with a snapshot attached checks for conflicts first. If the snapshot would replace funnels, websites or other assets you already customized, you get a warning and choose which version wins. This applies to SaaS V1 and V2
• General dashboard widgets became editable on September 24. Open Edit dashboard, then the three-dot menu on any general widget, and the editor opens on three tabs. Configure renames the widget and sets a date range override that ignores the dashboard-level date filter. Conditions applies permanent filters by assigned user, tags, source, lead value, custom fields and attribution, and those filters hold every time the dashboard is opened. Themes applies a visual style to match client branding. Widgets can also be duplicated and put in comparison mode against last period or last year. Setting Assigned user to Logged in user once gives every team member a dashboard scoped to their own numbers, which is the version worth building for a client login. Each widget face now states its date property and date range, so two similar widgets that disagree stop being a mystery. Pipeline stays a runtime filter on the widget face, Funnel and Manual Actions widgets are not configurable yet, and general widgets still ignore dashboard Quick Filters
• The default dashboard can be set during snapshot load. Since September 17 each selected dashboard carries a Set as default option at load time, so the destination sub-account opens on the dashboard you intended. Only one can be marked default, and leaving it unset keeps the snapshot's existing default. This used to be a per-sub-account setting, which meant opening every destination after the load. At a few hundred sub-accounts that was most of a day
• Gift Cards got eight permissions on September 17. View orders, view products, create and edit products, access the sell link and QR code, send gift cards, view the full gift card code, open the order details page, and delete or deactivate. The full code is protected separately, so a support user can look at orders without being able to reveal a code or issue a card. Assigned-data granularity is supported, so a user can be limited to their own records
• Agency Settings has a Workflow Settings page as of September 9, with a default workflow builder choice, Standard or Advanced, and an Apply to all sub-accounts toggle. Sub-accounts created later inherit the agency default and existing workflows keep the builder they were built on. A refreshed user management experience also arrived through Labs on September 9, splitting My Profile into tabs and replacing the long roles-and-permissions scroll with a searchable module panel. Capabilities are unchanged
• Marketplace app detail pages were rebuilt on September 18. Rating, install count, target user, install eligibility, white-label support, pricing and developer details are scannable upfront, with separate sections for pricing tiers, grouped permissions showing read and write scope, and reviews. Similar apps and other apps from the same developer sit at the bottom. For agencies vetting what sub-accounts are allowed to install, the permissions view is the part that earns its keep

Services, calendars and rentals got their own release day

September 22 published nine releases in a single day, and most of them landed on the booking side of the platform. Taken together they close the gap between what a service business actually charges and what the platform was willing to record.

• Tax Zones put manual tax rates behind a map. Under Manual Taxes you can now group countries and states into a named zone, attach up to ten user-defined rates to it, and have those rates apply only to customers in that zone. A country or state belongs to one zone at a time, so configurations cannot overlap. A rate has to be attached to both the product and the zone before it charges, a zone with no rates charges nothing, and any region outside every zone falls back to whatever the product carries. Automatic Taxes still take priority, and an existing setup with no zones defined behaves exactly as before. This is the piece that was missing for anyone selling across provinces or states from one store
• Service prices can be edited on the appointment itself. Creating, editing or duplicating a Services appointment now lets you change that service's price in the appointment modal, and the total updates immediately. The change applies to that one booking and never touches the global price. Zero is allowed for complimentary work, negative values are blocked with a validation error, and an edited amount carries a Custom tag next to it. It also overrides a staff-specific price if one is configured. The mobile app has it too
• Staff can carry a role badge. Create reusable badges such as Lead Therapist or Senior Consultant, up to 30 characters, give each one a color, and assign one per staff member. The badge appears on the staff card during booking, tinted from your widget's primary color. Staff without a badge show nothing at all, so the card stays clean
• Calendar AI now explains why a slot is blocked. Ask why 5 PM on a given date is unavailable on a named calendar and it answers in plain language instead of sending you into the Troubleshooting View grid. Ask about a person rather than a calendar and it resolves across every calendar they serve. Ask about a whole day and it returns one de-duplicated summary per segment, with slot-level detail on a follow-up. When the reason engine cannot produce a clean answer it says so and links into the full Troubleshooting View rather than inventing a reason
• Third-party reschedules finally sync back. A single-service appointment rescheduled or cancelled directly in Google Calendar or Outlook now updates the service appointment record as well as the calendar event, and the calendar view, the list view and the detail view all agree. Multi-service appointments are still not covered, so keep managing those inside the app
• Notification templates speak the business language. Default notification templates across Calendars, Services and Rentals now localize from the Business Profile platform language, on email, SMS and WhatsApp wherever each product supports them. Templates you have already customized are never overwritten or retranslated. New calendars and new Services and Rentals locations get localized defaults automatically, and existing ones keep their current templates until someone resets them to default
• The mobile app can diagnose its own notifications. Settings, Diagnose Notifications runs four checks in one tap covering network, permission, sound and an actual test delivery, tells you which one failed and what to change, and copies the result for a support ticket

Communities grew up in a single day

Content Moderation shipped on September 23 and gives admins a first line of defense that runs before anything reaches the feed. Set the words, phrases or domains you do not want under Settings, Content Moderation, and any member post or comment that matches is held back automatically and routed to a Needs review tab. Matching is case insensitive, and HighLevel recommends phrases over single words for precision. A starter pack of common spam and scam phrases loads with one click, and a Try your rules tool lets you paste sample text and see whether it would be held before anything goes live. In the reviewer screen an admin sees which keyword triggered the flag, previews the full post, and either approves and publishes or removes it. The author still sees their own post with an In review status, so nothing disappears without explanation, and everyone else sees nothing at all. Member-reported content lands in its own Reported by members tab with the reporter named. Admin and owner posts are never held. Email, in-app and push alerts fire the moment something is flagged, and HighLevel says AI-powered spam detection is coming on the same foundation.

Post and comment search was rebuilt the same day. It now handles typos, alternative phrasing, synonyms and accented or non-English text, and results rank by relevance instead of newest first.

• Courses gained filtering and sorting. Members narrow the Learning page by topic, difficulty and language, combining values across filters, and sort by title A to Z or Z to A, newest or oldest. Creators assign those values in the course builder, and until someone does the filters have nothing to work with. Filter chips appear at the top of the Learning tab on mobile as well
• Dark mode arrived for Communities and Courses on GoKollab web. Members switch from the top navigation and the choice persists across refreshes. Owners configure light and dark palettes independently under Settings, Appearance, with live preview, so the dark palette is a design decision rather than an automatic inversion of the light one. Coverage runs across posts, comments, channels, chats, members, events, calendars, About pages, courses, lessons, assessments, quizzes and live sessions, plus the wider platform shell. Communities on a legacy custom theme keep their existing palette until a dark one is configured
• Voice notes in posts and comments landed earlier in the month, on September 11
• Community email unsubscribes stopped being all or nothing on September 30. Clicking Unsubscribe in a Communities email used to risk opting the member out of every platform email. The preference is now scoped to the one community that sent the email and the one notification type behind it, such as a comment mention or an @everyone post. A confirmation page names both before anything is saved, a reused or expired link says plainly that nothing was changed, and the member can switch that notification back on from their own settings at any time. For a paid community this is the difference between losing one notification and losing a member from every email you send

Taken together, this is the week Communities stopped behaving like a bolt-on. Moderation, discovery and appearance are the three things that decide whether a paid community survives its first hundred members, and all three moved on the same day. If a membership or course offer is part of what you sell, the industry playbooks show where it fits alongside the rest of the stack.

Bookings, payments, phone, and what to watch

Rental customers can cancel their own bookings. Since September 10 the confirmation link carries a self-service cancellation flow. The booking releases immediately with no approval step, any time before it starts. An optional Cancellation Reason field saves the customer's own words to the booking. Operators can set a cutoff under Rentals, Global settings, Booking settings, in minutes, hours or days before the start. Bookings that bundle several listings cancel together, and expired links now show a plain message instead of an error page. Add the link anywhere with the rentalBooking.cancellation_link custom value. Refunds and security deposits stay entirely in the operator's hands. Separately, custom forms can now override calendar-level confirmation behavior, so redirect to URL, custom message and disqualify lead can all fire from the form's own conditions once you tick the override under Global settings, Form and confirmation.

Failed subscription payments can be retried by hand as of September 11, on NMI, Authorize.Net, Square and Mercado Pago. Open any subscription marked Unpaid or Past due under Payments, Subscriptions, and once automatic retries are exhausted you can retry the oldest unpaid payment on a saved or new card, optionally making it the subscription default. Three attempts per card, then a 24 hour cool-down. Involuntary churn is the cheapest churn to fix. On September 17 SaaS V2 went one better: updating the default card now updates the card on active SaaS subscriptions too, so a renewal no longer fails against a card the client already replaced. Sub-accounts on SaaS v2 have also been able to delete saved payment methods since September 1.

• Local presence dialing was rebuilt on September 18. Location matching now covers the United States, Canada, United Kingdom and Australia using real postal code distance rather than static lookup tables, and it works across the web dialer and desk phones, with mobile app support still to come. Contacts see the same number from you on every repeat attempt, which builds familiarity and helps carrier standing. Up to 50 numbers per sub-account can be reserved so they are never auto-selected, protecting a main office line while keeping it dialable by hand. The matching ladder runs zipcode, area code, overlay code, nearest available distance, same region, account default, and a call is never delayed or blocked if no match exists. Settings, Phone System, Voice, Other Settings, Outbound calls
• Calls transfer to staff with no personal phone number. From September 7 the mobile app transfers a live call to a colleague's web app, mobile app, phone or SIP desk phone, ringing every configured device at once. Warm and blind transfers both work. Re-transferring an already transferred call is not supported yet. Missed Call Text Back also got smarter about sender selection on September 4: it tries the called number, then a number that has messaged the contact before, then the default outbound number, then any eligible messaging number, so the follow-up no longer fails silently when the dialed number cannot send SMS
• Chat widgets got spam protection on September 18. Bot and high-volume submissions are detected in the background with no CAPTCHA and no extra fields for genuine visitors. Flagged contacts are kept rather than deleted, tagged chat_widget_high_volume_flag, and you choose per widget whether to tag only, pause SMS, pause email or pause both. Existing widgets default to off, new widgets default to on with Pause Both. The point is protecting sender reputation and messaging spend without losing a real lead to a false positive
• App QR Codes shipped on September 18. One QR code detects the device and sends iOS to the App Store, Android to Play, and anything else to a fallback URL, replacing the two-codes-or-a-landing-page workaround for app download campaigns
• Ad Manager gained real Meta budget control on September 29. A Meta campaign can now be budgeted at campaign level, where Meta distributes one pot across ad sets on its own, or at ad set level, where each ad set carries its own budget and schedule. Daily and lifetime budgets both work at either level. On ad set budgets there is an opt-in that lets Meta move up to 20 percent of one ad set's budget to better-performing ones, which keeps your split intact while leaving Meta room to react. Bid strategy is selectable between highest volume and cost per result goal, with the target cost set per ad set. The structure can be changed after publish, moving between campaign and ad set budgets, though Meta may block further budget edits for a period after each change. Ad Manager has also surfaced Meta's post-publish optimization recommendations inside a published campaign's Review screen since September 10. Between the two, a Meta campaign no longer has to be finished off in Meta's own interface
• Mexico local numbers lost SMS. An upstream carrier migration means SMS is no longer supported on Mexico local phone numbers, and from September 29 the Phone Numbers table carries an SMS not supported pill on every affected number, existing and newly purchased, with a tooltip explaining why. Voice keeps working, and WhatsApp where it applies. Nothing needs enabling and the buying flow is unchanged, so the listed capabilities on a Mexico local number are now the thing to read before purchasing one. Settings, Phone Numbers, Manage Numbers, Phone Numbers tab. If a Mexican sub-account runs SMS follow-up off a local number, that automation is already failing and this pill is the only place the platform says so

Elsewhere this month: the User login workflow trigger started firing for the new Client Portal experience, and a display bug showing Hungarian Forint plan prices 100 times too high is fixed.

One new deadline is aimed at developers rather than users. A Labs change announced on September 23 moves conversation attachments behind permission-based, authenticated access instead of directly reachable links, starting with email attachments and extending to other channels and file-based custom fields in later phases. Existing attachment links keep working until November 30, 2026, after which the new controls take effect permanently. For most HighLevel accounts nothing changes. Marketplace apps that fetch or display email attachments outside the platform need the new file access permission and have to ask existing users to reauthorize. If you run a custom integration that pulls attachments, that reauthorization prompt is the part clients will ask you about, so schedule it rather than meet it on November 30.

The late-August wave is documented in full in the August 2026 roundup. Three dates matter now. On October 1 WhatsApp rates drop, WhatsApp service and in-window utility messages start costing money, and SMS carrier fees rise. On October 13 international SMS rates and phone number rentals rise across roughly two dozen countries. On November 30 the old conversation attachment links stop working. Model your own volumes, audit your number inventory and check your custom integrations before those dates rather than after. The wider direction is now hard to miss. HighLevel spent the first half of September making the builder safer, the second half making the copilot reachable from outside the product entirely, and the last week halving the price of the models underneath it. Then it gave that copilot a voice on your phone. This page is the closed record of September, and the weekly roundup tracks new releases as they land.

Frequently Asked Questions