How to Set Up GoHighLevel Reviews AI Agents: Brand Voice, Guardrails and Sensitive Review Holds

Replying to every review is one of those jobs every local business knows it should do and almost none keep up with. Reviews AI has been able to draft and post replies for a while, but the worry was always the same: what if it says something daft, or worse, something that sounds like an admission of fault, on a public page that sits next to the business name for years?

This update is aimed squarely at that worry. This guide covers exactly what changed, why it matters if you look after client accounts, and how to set it up step by step, with example guardrails you can adapt. Every fact about the feature comes from HighLevel's own changelog and help docs, linked at the bottom. These went into the changelog on the opening day of LevelUp 2026, but HighLevel has not labelled them as LevelUp announcements, so I am treating them as normal releases.

More short GoHighLevel fixes like this are on my YouTube channel, Code With Urooj.

What HighLevel changed in Reviews AI

There are two changelog entries, both dated 26 October 2026.

1. Reviews AI agents got four new controls

According to the Reviews AI Agents changelog entry:

• Brand voice controls. For each agent you can set a "we" or "I" response style, the business name, tone traits, response length and character limits, and sign-off or footer preferences.
• Quality guardrails. You can switch on controls to remember recent responses, avoid repeating the same opening lines, and avoid risky language such as refund promises, guarantees, fault admissions, liability statements, or medical and financial advice.
• Hold sensitive reviews for human review. When this is on, higher-risk reviews can be held for a person to check before anything is posted, instead of being answered automatically.
• Live response preview. Before you save an agent, you can see how it would reply to a positive, a neutral and a negative review, and regenerate those replies as you change the tone, prompt and other settings.

2. Agencies can manage Reviews AI in bulk

The Bulk Manage Reviews AI changelog entry adds a Reviews AI control at Agency, AI Suite, AI Setup. From there you can set the response mode to Off, Suggestive or Auto for one sub-account or many at once. If a sub-account changes its own mode under Settings, Reviews AI, the agency view updates to match.

HighLevel also notes that Reviews AI can be switched off completely for a sub-account at Agency, Sub-Accounts, Features & Limits, Reviews AI. When it is off, the Reviews AI settings and the AI Reply and Regenerate buttons are hidden, and every AI review reply is blocked, including manual AI replies, auto responses and Drip Campaign replies.

A quick reminder: Suggestive versus Auto

HighLevel's help article on the two modes explains the difference. In Suggestive mode, someone clicks AI Reply on a review and Reviews AI drafts a reply for them to post. In Auto mode (called Auto-Pilot in the older help article), replies go out automatically, and you can set things like a wait time and footers. The new guardrails and sensitive review hold matter most in Auto mode, because that is where nobody reads the reply before it goes live.

Why this matters for agencies and local businesses

• Replies sound like the business, not like a robot. A sole trader who signs off with their first name and a dental group that always says "we" can now both get replies that fit, set once per agent.
• Fewer copy-paste replies. Anyone who has scrolled a Google profile full of "Thank you so much for your kind words!" knows how bad that looks. The guardrail that avoids repeating opening lines goes straight at this.
• Less risk on angry reviews. The risky reply is almost always on a one-star review: an apology that reads like an admission, or a promise of a refund nobody approved. Guardrails plus a human hold give you a safety net there.
• You can test before it goes live. The live preview means you can show a client exactly how their replies will read before you switch Auto on, which makes the conversation about turning on AI a lot easier.
• Rollout across clients is quicker. If you manage dozens of sub-accounts, setting the mode in bulk from the agency view saves a lot of clicking, and you can see at a glance which accounts are on Off, Suggestive or Auto.
• You can prove it is working. HighLevel also released 16 Reputation widgets for location dashboards on the same day, including response rate, average response time and AI versus manual responses, so you can show a client the before and after.

Before you start

• Permissions. HighLevel's getting started guide for Reviews AI agents says that if you cannot create or edit agents, an admin needs to enable the Manage Reviews AI Agents permission for your role.
• Review sources. Make sure the client's review sources, such as their Google Business Profile, are connected in Reputation, or there will be nothing for the agents to reply to.
• Agree the rules with the client. Ask who should see sensitive reviews, what the business is never allowed to say in public, and how they like to sign off. Ten minutes on this saves a lot of awkward edits later.
• Labels may vary. HighLevel's changelog names the new controls but does not show every screen, so the exact wording of buttons in your account may be slightly different from what I use below.

Step by step: set up Reviews AI agents with brand voice and guardrails

Step 1: Choose the response mode for each sub-account

If you are an agency, go to Agency, AI Suite, AI Setup. Select one or more sub-accounts and set the response mode to Off, Suggestive or Auto. For a client who is new to AI replies, I start on Suggestive for a week or two so they can see the drafts before trusting Auto. Inside a single sub-account, the same choice lives under Settings, Reviews AI, and the two stay in sync.

Step 2: Create or open an agent

In the sub-account's Reviews AI settings, open the agents section and click Create Agent, or open one you already have. HighLevel's help guide says you can build an agent from scratch or start from a ready-made template, and there is a Create Starter Agents option that switches on the pre-built agents in one go. Give each agent a clear name, a tone and a short prompt.

Step 3: Set the brand voice

In the agent's settings, fill in the brand voice options from the changelog:

• "We" or "I": "we" suits most businesses with a team. "I" suits sole traders where the owner really is the one replying.
• Business name: use the name customers know, not the registered company name.
• Tone traits: pick two or three at most, for example warm, plain and professional. A long list tends to produce mushy replies.
• Length and character limits: keep positive review replies short. Two or three sentences reads as genuine; a paragraph reads as automated.
• Sign-off or footer: a first name or "The team at [business name]" works well. Avoid adding phone numbers to every reply.

Step 4: Switch on the quality guardrails

Turn on the guardrails HighLevel lists: remember recent responses, avoid repeating the same opening lines, and avoid risky language such as refund promises, guarantees, fault admissions, liability statements, and medical or financial advice. If you also write your own rules into the agent's prompt, keep them short and specific. There are some examples you can adapt in the next section.

Step 5: Hold sensitive reviews for a person to approve

Switch on the option to hold sensitive reviews for human review. When it is on, higher-risk reviews wait for a person before a reply is posted, so they are not answered automatically. Decide with the client who checks these and how quickly. A held review that nobody looks at for a week is worse than a slightly clumsy reply on the day. HighLevel's changelog does not spell out exactly what it counts as a sensitive review, so test it with the preview and keep an eye on what gets held in the first few weeks.

Step 6: Test with the live preview

Before you save, use the live preview. It shows how the agent would reply to a positive, a neutral and a negative review. Change the tone, prompt or other settings and regenerate until all three read like something the business owner would actually write. Pay most attention to the negative one. If it apologises in a way that sounds like taking the blame, tighten the prompt and regenerate.

Step 7: Assign agents and go live gradually

HighLevel's help guide explains that you can assign agents by review sentiment (positive, neutral or negative), turn on round robin so different agents take turns, set a fallback language for reviews whose language cannot be detected, and assign agents to specific Google Business Pages if one sub-account covers several locations. A setup that works well for most local businesses is a short, warm agent for positive reviews and a calmer, more careful agent for negative ones. Save, leave it on Suggestive or watch Auto closely for the first couple of weeks, and check the replies on the live profile.

Example guardrails you can adapt

Example rules for any local business:

• Example: "Never offer a refund, discount, voucher or free service in a reply. If the customer asks for one, invite them to contact the team directly."
• Example: "Do not say we were at fault or that something was our mistake. Thank them for the feedback and say we would like to look into it."
• Example: "Never mention the customer's job details, address, booking date or anything they did not write in the review themselves."
• Example: "Do not argue with the reviewer or say their account is wrong, even if it seems to be."
• Example: "Keep replies to positive reviews under 3 sentences. Do not start two replies in a row the same way."
• Example: "For any negative review, end with an invitation to contact us by phone or email so we can put things right."

Extra examples for clinics, dentists and similar:

• Example: "Never confirm or suggest that the reviewer is or was a patient. Reply in general terms only."
• Example: "Never give advice about treatment, symptoms or medication."

Extra examples for trades and home services:

• Example: "Never promise a return visit, a fix or a time frame in the reply."
• Example: "Do not mention insurance, warranties or guarantees."

Things to watch out for

• Switching off at agency level blocks everything. If you disable Reviews AI under Features & Limits, HighLevel says all AI review replies are blocked, including manual AI replies and Drip Campaign replies. Do not use that switch just to pause Auto mode; set the mode to Suggestive or Off instead.
• Sub-account changes show up at agency level. Because the two stay in sync, a client user who flips their own account to Auto will change what you see in AI Setup. Check AI Setup now and then rather than assuming nothing has moved.
• A held review still needs a reply. The hold stops a risky auto reply, but the customer is still waiting. Make sure the person who approves held reviews knows it is their job.
• Guardrails reduce risk, they do not remove it. HighLevel describes them as reducing risky output. Keep checking live replies, especially in the first month and after you change a prompt.
• Drip campaigns use your agents too. HighLevel's Drip Mode help article says Drip campaigns for older, unanswered reviews use your Reviews AI agents. Set up brand voice and guardrails before you start one, and pause a running campaign before you make big changes to an agent.
• Permissions can hide the settings. If a team member says they cannot see the agent settings, check the Manage Reviews AI Agents permission on their role before assuming the feature has not reached the account.

If you are looking at AI across the rest of the account too, read my guide to using Claude or Gemini in Conversation AI, the GoHighLevel AI automation guide, and my LevelUp 2026 preview. For the inbox side of the same update, see my guide to keyword search and social comments in Conversations. If you want to run your accounts by typing plain English instructions, see the GoHighLevel MCP server.