Skip to content
ConsultEvo

HighLevel Release Radar: August 31 to September 4 Setup Guide

HighLevel’s Release Radar for August 31 to September 4, 2026 introduced six practical feature areas: full-stack AI Studio projects, grouped announcement audiences, Chat Widget Spam Protection, two AI workflow actions, and Ask AI Dispatch through WhatsApp. This guide turns that recap into current setup paths, operating decisions, and validation gates for teams preparing features for production.

The original release post described AI Studio full-stack capabilities as private beta. Current HighLevel documentation says the private beta is over for newly created projects, which now use the full-stack architecture. Existing Vite/React projects are not automatically migrated. The release post remains useful for historical context, while the current Help Center documentation should control implementation decisions.

Across the features, the practical pattern is consistent: identify the source event, define the output, validate it before taking action, and assign an owner for exceptions. An AI response, a spam flag, a generated image, or a WhatsApp instruction should not become an unchecked customer or CRM change.

Treat every release feature as an operating process, not just a switch: document its input, destination, validation gate, and exception owner before enabling it at volume.

What changed in HighLevel’s August 31 to September 4 Release Radar?

The release covered full-stack AI Studio projects, targeted audiences for agency announcements, server-side Chat Widget Spam Protection, AI Image Generation, AI Analyze Image, and WhatsApp access to Ask AI Dispatch. The table is a route map. The detailed sections below provide the current setup path and the control that matters most.

Feature Input or trigger Output or decision Main control
AI Studio Page request, app call, or external HTTP request Rendered HTML, private response, or endpoint response Choose the correct server boundary
Announcements User and sub-account attributes Announcement visibility Test grouped AND and OR logic
Spam Protection New chat-widget submission Tag and selected channel pause Review flagged contacts
AI workflow actions Prompt, workflow data, or image URL Asset or plain-text analysis Validate before distribution or CRM write
Ask AI Dispatch WhatsApp text, voice, file, or media request Supported response or action Confirm consequential changes

The original HighLevel release announcement is a dated feature recap, not a complete API or implementation specification. Feature access, permissions, billing, model availability, and navigation can depend on the account and current product documentation.

AI Studio: choose the right server-side boundary

Current HighLevel documentation says newly created AI Studio projects use the TanStack Start full-stack architecture. Existing Vite/React projects are not automatically converted. Before planning a migration, confirm which architecture the project currently uses and which settings or integrations must be recreated.

  • Use SSR for public pages whose HTML should be rendered on the server before delivery. This can make page content more available to crawlers and services that generate link previews, but it does not guarantee indexing, rankings, or correct previews.
  • Use a Server Function for private backend logic called by the project frontend, such as a server-side request to a payment or data provider.
  • Use a REST API route when an external application, webhook, or integration needs a callable HTTP endpoint.

Store API keys, tokens, and webhook secrets in AI Studio Secrets and reference them only from server-side functionality. A Secret is not a guarantee of security if a value is returned to the browser, written to exposed logs, embedded in frontend code, or made available through an insecure endpoint. Define the endpoint method, request contract, and authentication requirements for the application. Validate webhook signatures where the integration requires them; HighLevel does not establish one universal authentication scheme for every route.

For a published page, inspect the actual HTML, metadata, canonical URL, robots directives, redirects, and social preview. SSR is an architectural capability, not a ranking promise.

Availability update

The September 4 release post described full-stack AI Studio as private beta. Current documentation says the beta is over for newly created projects. That update does not migrate existing Vite/React projects automatically, so verify the project architecture before relying on SSR or planning a conversion.

See HighLevel’s current AI Studio documentation for SSR, Server Functions, REST API routes, and Secrets for the current product boundaries and migration caveats.

Announcements Builder: make audience logic testable

In Agency View, go to Settings > Announcements, create or edit an announcement, and open the Audience settings. Choose Agency Level, Sub-account Level, or both. Documented user fields include role and platform language. Documented sub-account fields include status, country, timezone, niche, HIPAA compliance, SaaS mode, and other attributes listed in the current interface.

Conditions in one filter group use AND logic. Separate filter groups use OR logic. This documented group model should not be described as unlimited nested Boolean logic or assumed to support arbitrary custom CRM fields.

For example, an agency could express this rule: show the announcement to US sub-accounts using LC Email, or to HIPAA-compliant sub-accounts in the healthcare niche.

  • Group A: Country = United States AND Has LC Email = true.
  • Group B: HIPAA compliance = true AND Niche = Healthcare.

Test at least one expected match and one expected non-match from each group. Also review schedule, priority, frequency, and CTA behavior before publishing. A CTA can have a destination URL and open in a new tab. Announcement management may depend on agency role and permissions. The Announcements Builder documentation lists the current scopes, fields, logic, and controls.

Chat Widget Spam Protection: flag first, review before restoring

Go to Sites > Chat Widget, open a widget, and choose Spam Protection. Select Tag only, Pause SMS, Pause Email, or Pause Both, then save. The server-side feature does not add CAPTCHA, extra form fields, or intentional visible delay.

Flagged submissions remain in Contacts with the tag chat_widget_high_volume_flag. The tag is a review marker, not proof that a contact is malicious. The selected outbound channels are paused for flagged contacts. Calls, inbound messages, and unrelated DND settings are not affected.

New widgets default to protection enabled with Pause Both. Existing widgets remain off until someone enables protection. The feature applies to new submissions after activation, not retroactively. When a contact is flagged, inspect the contact and activity log to identify the paused channel. For a legitimate lead, remove the relevant DND setting and tag as appropriate before resuming follow-up.

Use Tag only when staff can review promptly or when pausing outreach would be too disruptive. During a known event, launch, or campaign surge, temporarily disable protection or increase review rather than deleting contacts automatically. HighLevel documents no detection score or threshold to tune. The Spam Protection guide covers configuration, defaults, channel behavior, and review.

AI workflow actions: generate assets or inspect incoming images

Generate an image asset

In Automation > Workflows, open a workflow, add an action under AI Actions, and select AI Image Generation. Build the prompt from fixed text, contact fields, webhook data, or previous workflow outputs. Configure the model and quality, then use the available additional settings, such as reference images, size, background, output format, Design Kit, or Brand Voice. The documentation allows up to five reference images.

The premium action is billed per execution. Generated images are saved to Media Storage. Later workflow steps can use Image URL, Image File Path, and Image File Name. The Image URL is publicly accessible, so do not treat it as private storage. If transparency is needed, use a compatible format such as PNG or WebP. JPEG does not preserve transparency.

Before a generated asset reaches a customer, have a marketing, brand, legal, or product owner check claims, product details, accessibility concerns, and rights requirements that apply to the use case.

Analyze one image

In a workflow, add AI Analyze Image under AI Actions. Supply one publicly accessible URL that points directly to the image file, not a webpage containing the image. Select a model and detail level: Auto, High, or Low. High detail is recommended for small or dense text. The action processes one image URL per action and returns plain text.

A later workflow action can use that text in a contact field, notification, outbound webhook, or conditional branch. For known values such as campaign ID, source widget, file type, or required-field presence, use deterministic workflow conditions instead of image AI. Use image analysis when the required signal is visible in the image and unavailable in structured data.

01SourceSupply one anonymously accessible image-file URL and confirm that it resolves without login or private-share permission.
02Ask narrowlyUse a bounded prompt such as “Read the total amount. Return only the amount, or null if unreadable.”
03Check the textTreat the response as plain text. Validate numeric format, currency, date, expected prefix, plausible range, and source-image quality before accepting a value.
04Route or escalateWrite an accepted value to the intended field or branch on it. Route unreadable, inconsistent, financial, identity, medical, or compliance-sensitive cases to a named reviewer.
05Measure correctionsTrack reviewer corrections on real images before increasing volume, and retain the raw response with the analysis context when an audit trail is required.

The following is a hypothetical normalized record that a separate parser or validation step might create. It is not the native action’s guaranteed output:

{
  "amount": "48.20",
  "currency": "USD",
  "review_status": "pending"
}

For a financial extraction, retain the original image reference, analysis timestamp, prompt version, model, detail level, and raw response alongside any accepted normalized fields. The AI Image Generation guide and AI Analyze Image guide document the native actions and their constraints. Teams defining ownership of validated CRM data can also review CRM systems consulting.

Ask AI Dispatch through WhatsApp: treat it as an operator interface

Open Ask AI > Customize > Dispatch and choose Connect WhatsApp. Scan the QR code, request the verification code through WhatsApp, and enter it in HighLevel. The code expires after five minutes. Each location supports one connected WhatsApp number, and the user who connects it owns that connection.

Documented interaction types include text, voice messages, file attachments, and multimedia responses. Supported examples include retrieving contact information, reviewing opportunities and appointments, managing supported CRM records, generating blog drafts, emails, and images, creating supported recurring tasks, and running configured Ask AI Skills. Some requests that modify CRM data may require confirmation.

Use the connected number as an operator interface for supported tasks, not as an unrestricted command API. Assign an owner for the number and for ambiguous or consequential requests. Where the team’s audit process permits, record the location, operator, request, confirmation state, target record, and result. Review Dispatch history when investigating an unexpected action. Broader teams defining bounded AI responsibilities may also consider AI agent design and implementation.

Production checks: validate outputs, access, and ownership

Keep shared controls here rather than repeating the same warning in every feature section. AI-generated text and images are candidate outputs, not authoritative business records. Public image URLs should be treated as shareable. Keep confidential information out of prompts and assets unless its use has been approved.

Define the actual record grain before sending data to an external audit store. A contact is not a chat submission. An image is not an analysis execution. A generated asset is not a workflow run. A WhatsApp message is not a Dispatch session. An announcement definition, audience rule, and display impression are also different records.

Where a stable source event or execution identifier is available, retain it. Confirm that it is actually exposed before designing around it. For image analysis, a proposed collision-resistant key might include location ID, workflow execution ID, image hash or URL, prompt version, model, and attempt. For generated assets, retain the workflow execution ID, prompt version, source-variable snapshot, asset URL, asset index, and attempt. For WhatsApp tasks, retain location ID, connected number, session or message identifier if available, request timestamp, confirmation state, and resulting CRM object ID. These are illustrative external data-model recommendations, not HighLevel schema.

A contact ID alone is not a unique spam submission identifier, and the same image may be analyzed repeatedly with different prompts or models. If concurrent workers write to an external store, do not use lookup-then-create as the sole duplicate safeguard. Enforce uniqueness in the database and use a transactional upsert. If HighLevel does not expose a stable identifier for the integration, state that limitation and confirm an alternative key before implementation.

Review before rollout
  • Confirm current feature access, account permissions, plan conditions, and setup path.
  • Test representative inputs, anonymous image access, expected matches, non-matches, and failure cases.
  • Write the expected output and deterministic validation rule before a CRM write or customer message.
  • Name the person who reviews exceptions, restores legitimate leads, approves assets, or confirms CRM changes.
  • Check privacy implications for public image URLs, prompt content, generated assets, and WhatsApp requests.
  • Define the record grain, source identifier, uniqueness rule, and concurrency-safe upsert for external storage.
  • Measure audience mis-targets, legitimate-lead restoration time, extraction correction rate, asset rejection rate, and supported Dispatch completion time.

These measures are recommended operating metrics, not reported HighLevel benchmarks. Recheck current documentation immediately before launch because navigation, model availability, billing, permissions, and product behavior can change. The right setup is the one whose input, output destination, validation gate, and human owner are explicit before it affects production.