Review requests: the moment, the follow-up and the unhappy branch

Most review automation is one workflow with one message in it. Job marked complete, text goes out, link attached, done.

That version works for about a week, produces a handful of reviews from the customers who were going to leave one anyway, and then flattens out. The business concludes that review automation does not do much.

A review engine is actually three separate decisions, and the single-message version only makes the first one, badly. The three are the moment, the follow-up and the branch for the customer who is not happy.

Decision one: the moment

The ask has to fire at the completion moment for that specific business, and the completion moment is not the same event in any two trades.

For a trade it is the point the work is signed off and the van is pulling off the drive. For a clinic it is the end of the visit, before the patient has reached the car park. For a broker it is the close, when relief is the dominant emotion. For a restaurant or a rental it is the end of the stay. Every crate in the depot carries a completion moment written into it for exactly this reason, which is why the roofing contractor crate fires post-install and the dental practice crate fires post-visit rather than both firing on a generic “opportunity won” event.

Timing does most of the work. An ask that arrives while the customer is still standing in the finished room converts. The same ask on Tuesday arrives while they are at work thinking about something else, and it converts at a fraction of the rate. If you change one thing about an existing review setup, change how quickly it fires.

Two practical points about the message itself. Put the direct link in it, pointing at the review form rather than the profile page, so the customer does not have to find the button. And ask once, in one sentence, for one thing. A message that also asks them to follow you on Instagram and join the mailing list gets none of the three.

Decision two: the follow-up

One ask is not a campaign. Most people who intend to leave a review do not leave it on the first message, and they are not refusing, they are busy.

The follow-up has three properties worth building deliberately. It waits long enough that the second message is not nagging, which in practice is a few days rather than a few hours. It changes channel, because somebody who ignored a text may well answer an email, and the reverse is just as true. And it stops the moment a review appears, which means the sequence has to be able to see that a review appeared.

Three touches across a fortnight is a reasonable shape. More than that and you are annoying the people who already decided no, which costs you nothing today and something later.

This is the half that produces most of the volume, and it is the half that almost never gets built by hand, because it is fiddly and the first message felt like the job. It is also the difference between a review profile that grows and one that sits still.

Decision three: the unhappy branch

Before the public ask, a short satisfaction check: one question, one tap, a rating out of five or a straight yes or no.

The purpose of that check is to route the conversation. A customer who answers low has told you something is unresolved, and the useful response is a call from somebody who can fix it, today, while it is still fixable. Without the check, the first you hear about it is a one star review written by somebody who has given up on being helped.

Be precise about what this is, because there is a version of it that is against Google’s own policy and it is worth knowing which side of the line you are on. Selectively soliciting reviews only from customers you already know are happy is review gating, and Google prohibits it. Withholding the link from somebody because they answered low is that. Sending the link to everybody, and additionally raising an internal alert and a recovery call when somebody answers low, is not. The difference is whether the unhappy customer still has the same access to your profile as the happy one. Keep it so they do.

The other two rules in the same family are simple and absolute. Never offer anything in exchange for a review, including a discount, a prize draw or a free upgrade. Never write the review for the customer or send them a suggested wording.

Why the depot ships this as two components

Two of the nine fitted components in every crate are about reviews, and people occasionally read that as padding. They are genuinely different machines.

Review Harvesting is the capture. It watches for the completion moment, asks once, immediately, on the channel that customer already replies on, with the direct link in the message. It is one decision and one message, and its whole job is timing.

Review Automation is the standing engine behind it: the multi-touch cadence for everyone who did not respond, the channel switch on silence, the internal routing of an unhappy answer, and the replies to reviews that get published. It runs continuously rather than firing once.

Split that way, you can change one without breaking the other. Move the completion moment for a niche and the cadence is untouched. Rewrite the cadence and the timing of the ask is untouched. Built as one workflow, every change is a change to everything, which is why the single-workflow version tends to be abandoned rather than tuned. The review engine setup service builds them as two for the same reason.

The replies are part of the system

A review profile with fifty reviews and no owner replies looks abandoned. Replying is half of what the engine is for, and the two cases need genuinely different templates.

A five star reply is short, specific and thankful, and it names the service so the reply itself carries a keyword the next reader can see. A one star reply is not a defence. It acknowledges, it does not argue the facts in public, it names a way to continue the conversation privately, and it is written by somebody who is not angry. Automating the first case is straightforward. The second should be drafted automatically and sent by a human, and every crate is wired that way rather than letting the AI answer a complaint unattended.

What the profile is actually for

Reviews are not a trophy. They are an input to the next local search and the next hesitant reader.

Recency matters as much as volume, which is the real argument for a standing engine rather than a push once a year. Forty reviews all dated eighteen months ago reads worse than twelve spread across the last quarter. The system’s design target across the crates is five to ten reviews a month, which is a rate rather than a total, and a rate is the thing that keeps a profile looking alive.

The other use is quieter. A steady review flow tells you, in near real time, what customers think of a change you made. That signal arrives sooner and more honestly than any survey, and it is worth reading rather than only counting.

The next decision

Check one thing before you build anything: how long after completion does your current ask fire. If the answer is “the next day” or “when somebody remembers”, fixing that is worth more than every other change on this page combined.

Then decide whether the follow-up exists at all, because in most accounts it does not. The same standing-engine argument applies to the appointment side of the calendar, which is the hour after a missed appointment, and it applies across most of the health and wellness aisle, where recall intervals and reviews run on the same rhythm. There is a fuller picture of how a clinic’s ask sits alongside its booking system in GoHighLevel for clinics and wellness businesses.

Every crate in the catalog ships with both review components already timed to that trade, and the five-station walkthrough shows where the ask sits on the line.