What goes wrong when nobody reads a reply first?
The Google API sends whatever it is given. It does not know your business. It will not stop a reply that thanks the wrong person, promises a refund, or repeats a customer's name in a public thread.
Picture a plumbing company that sets a tool to answer every review the minute it arrives. A customer leaves two stars about a missed appointment. The automatic reply thanks them for their kind words and invites them to book again. Nothing in it breaks a rule. Everything in it is wrong. (That one is made up, but we have seen replies that were close.)
A reply is public, it carries your business name, and it stays under the review for as long as the review does. That is a lot to hand to a script.
What do rules catch, and what do they miss?
Rules are good at mechanical mistakes, and OpenGMB runs them on every reply. A link in a reply is stopped by U7.url, an email address by U7.email, a hashtag by U7.hashtag and an offer by U7.offer. A reply that opens the way the last one did gets a warning from R6.opening. For healthcare practices, H1 stops wording that confirms the reviewer was a patient.
A block cannot be overridden. A warning has to be acknowledged by whoever approves. Both are cheap and fast, and a person would miss some of them on a busy day.
What rules cannot do is read the room. They do not know the customer was upset for a good reason. They do not know the owner phoned them yesterday.
What does a person catch?
- Tone. A reply can pass every rule and still sound cold or defensive.
- Context. The owner already called the customer, and the reply should say so.
- Facts only your staff know. Whether the job happened, who did it, and whether a refund is really on the table.
- The review you should not answer in public at all.
Honestly, ten seconds of a human reading a draft that already passed the rules is the cheapest quality check there is. And it is the part that makes a reply sound like your business wrote it.
How does approval work in OpenGMB?
A reply moves from draft to pending approval, then approved, then published. Approving does not send it by itself. Publishing is a separate step, unless an owner has set a workspace to publish automatically once a reply is approved. That setting is off by default.
- Rules check the reply when it is submitted. Blocks have to be fixed.
- Warnings need an acknowledgement from the approver.
- Each approved reply records who approved it.
- Posts always need someone with approval rights. There is no auto-approve for posts.
- Profile edits wait for approval too.
Two things can relax this, and both are choices. An owner can switch on auto-approve for replies to 4 or 5 star reviews. It is off until then, and replies to 1, 2 and 3 star reviews always need a person. And a read and write MCP token acts with the rights of whoever made it, so an agent holding one can approve too. We think the person-first default is right for most businesses, and those options are there for owners who weigh it differently.
The Google Business Profile guardrails and approvals page lists every rule pack. For the reply steps themselves, see how to reply to Google reviews.
Does approval slow you down?
Less than you would think. Drafts wait in one queue, so you review them in one sitting. Set aside a slot once a day, read what is waiting, approve what is good and fix the rest.
And if you use an AI agent, you choose how much it can do. If you keep approval for yourself, the queue is where the agent hands its work to you: it drafts, and you read.
Do it with OpenGMB
- Open Reviews and look for replies marked as waiting for approval.
- Read each draft. Fix anything a rule blocks, and acknowledge a warning only if you agree with it.
- Approve the ones that are right, then publish them.
With Claude or another MCP client, connect the server at https://opengmb.synectus.com/mcp (the MCP setup docs have the steps). The tool reviewReplies_listQueue lists replies in one status, such as pending approval. This prompt only looks at the queue.
Prompt for Claude or any MCP client
Use the opengmb MCP server. Call reviewReplies_listQueue with status pending_approval. For each reply, show the review rating, the reply text and any rule that fired. Do not approve, reject or publish anything.
You choose how much your agent can do. A read-only token can look but not change anything. A read and write token acts with the rights of the person who made it, so give it only to an agent you are happy to have approve.
Questions about approving replies
Can a rule replace a person reading the reply?
No. Rules catch patterns such as links, emails and offers. They cannot judge tone or know what happened with the customer.
Do all replies need approval in OpenGMB?
Replies do, unless the owner turns on auto-approve for 4 or 5 star reviews. It is off by default. Replies to 1, 2 and 3 star reviews always need a person. A read and write MCP token acts with its owner's rights, so an agent holding one can approve as well.
Do posts and profile edits need approval too?
Yes. Posts always need someone with approval rights, and profile edits wait for approval as well. Auto-approve exists for replies only.
Can I override a rule that blocks a reply?
No. A block has to be fixed. A warning can be acknowledged by the person approving.
Does approving a reply publish it?
Not by itself. Publishing is a separate step, unless an owner has set the workspace to publish automatically once a reply is approved. That is off by default.
Related posts
- Google Business Profile API: access, APIs and limits
How the Google Business Profile API works: which APIs it includes, how to get access, what it cannot do and its current limits, read from Google's docs.
- $10 a month: how usage credit works
What the monthly credit pays for, how AI drafts and rank scans are priced, what happens at zero, and what rolls over, with worked examples.