To automate review requests with AI, let your job software trigger the ask when the work is actually finished, wait a short settling period, and send every customer one personal request with a direct review link. Follow up once, route complaints to a person, and match each review back to the job and tech that earned it.
Why don't review requests go out when you tell the team to ask?
Because asking depends on memory and nerve at the worst moment. The owner tells the crew to ask for reviews and means it. Then a busy quarter happens, and the business finds it collected a handful of reviews in three months while a competitor down the road added dozens.
That isn't a discipline problem. A tech packing up tools at the end of a long day, already thinking about the next call, is the wrong person to rely on for an awkward request. The fix is to make the ask a consequence of the job record instead of an act of will. When the work order closes and the work has had time to hold up, the request goes out, no matter how busy the week was.
It's worth doing. BrightLocal's 2026 survey of 1,002 US consumers found that 74% only care about reviews written in the last three months, and 28% say they will always write a review when asked. A steady stream of recent reviews beats a burst from two years ago.
What should trigger an automated review request?
A change in the job record that means the work is done, not a timer. The trigger is the product: most of what separates a good review program from an annoying one is when the ask fires.
Firing the instant a job closes, which is what most simple automations do, means the customer gets the request before they've seen the invoice or know whether the repair held. A few days later, the same customer has lived with the work and is a different respondent. The only thing that changed is the trigger.
Build the trigger from four checks, all about whether the work is finished:
| Check | What it looks at | Why it exists | Limit |
|---|---|---|---|
| Job closed and verified | Status complete, tech's completion notes recorded | A request on an open job reads like the business lost track | None: this is the starting line |
| No open callback | Return visits or warranty tickets still open on the same job | If it isn't fixed, the job isn't done | The request goes out when the callback closes |
| Invoice settled or sent | Payment status | A request next to a billing question gets lost in it | Hold no longer than your normal payment terms |
| Settling time | Days since completion, set by job type | Lets the customer see the work hold up | Two to seven days is a sensible range; pick yours |
Every hold has an end date. A customer whose job had a callback still gets asked once it closes, exactly like everyone else. Nobody is dropped because of what they might say, and that is the line between timing and gating.
What does the AI add over a basic automation?
Judgment about the record and better writing. A basic rule sends the same text to everyone; an agent reads the job and adapts.
- It reads the job. Job type, tech name and the work summary, so the message says "the water heater Luis replaced on Tuesday" instead of "your recent service."
- It applies the checks. It sees the open callback or the unpaid invoice that a simple status trigger misses.
- It counts. One request, one follow-up, and no more than one ask per customer per year.
- It sorts replies. A reply that describes a problem goes to a person the same day, not into the review flow.
- It traces the result. When a review arrives, it matches the reviewer to the job and tech, so you learn which work earns praise and which job types draw complaints.
The message should also say what it is. Automating the timing isn't deceptive. Signing a machine-written text as a personal note from a tech who is already at the next job is a small lie customers can feel. "Thanks for having Luis out" from the company is honest; "Hey, it's Luis!" is not.
What are Google's rules for automated review requests?
Ask everyone, offer nothing, and never steer unhappy customers away. Those three rules come straight from Google's own pages, as of November 2026.
- Google says businesses can ask customers to visit a Google link or scan a QR code to leave a review, and calls offering free or discounted goods or services in exchange for reviews strictly prohibited.
- Google's Maps content policy says merchants may not discourage or prohibit negative reviews, or selectively solicit positive reviews from customers, and lists asking staff to solicit a certain number of reviews as a problem too. No per-tech quotas.
- The FTC's consumer review rule took effect October 21, 2024. Its staff answers say asking only satisfied customers could violate the FTC Act, and the rule makes an exception for reviews that come from generalized requests sent to purchasers.
- Texts need consent, and under 47 CFR 64.1200 a stop request must be honored within 10 business days. Treat stop as same-day.
A "rate us 1 to 5 first" screen that sends fours and fives to Google and everyone else to a private form is review gating. Don't build it, and don't buy a tool that does it. This is not legal advice; ask your attorney about your consent language.
Should you buy a tool or build an agent?
Start with what you already pay for. Many field service platforms include some form of review request, and a reputation platform may already sit in your stack.
| Option | Good at | Falls short when | Fits |
|---|---|---|---|
| Built-in feature in your job software | Fast setup, uses your customer list | Triggers only on a status, with fixed wording | Shops with clean job statuses and one location |
| Reputation or listings platform | Monitoring many directories and locations | The problem is upstream: requests never go out | Multi-location businesses watching many sites |
| Custom agent in your software | Reads the job, applies the checks, traces reviews to techs | Costs more to set up and needs clean records | Shops where the built-in tool can't see callbacks or payment state |
If your built-in feature already asks every customer at a sensible time, keep it. Watching your reputation and building it are different jobs, and a monitoring platform is the right purchase when the first is your real problem.
Where automated review requests fall short
- Nothing beats the tech's in-person ask. At handoff, standing next to the finished work, a tech asking plainly is the best request there is. Automation catches the ones that get missed.
- It can't fix bad work. If tracing shows one job type drawing complaints, the fix is in the field.
- It needs clean records. Jobs closed late, or callbacks logged as new jobs, break the checks.
- It can't promise a number. It removes the forgotten ask; customers decide what to write.
- Replies stay human. An agent can draft a reply, but a public response to a complaint is the owner's call.
How do you set it up?
- Count last month's completed jobs and how many got a review request. That gap is your baseline.
- Define "finished" for each job type and set the settling time.
- Write two or three example messages in your voice. Our review request text template is a starting point.
- Turn on the checks, one follow-up and the one-per-year cap.
- Name the person who gets complaint replies.
- Approve the first 20 sends by hand, then let routine ones go.
- Review the tracing report monthly: requests sent, reviews received, techs named.
For how this looks at an HVAC company, see AI review requests for HVAC. The review requests workflow shows the same pattern in brief.
Where Cortex fits
Cortex builds review request agents inside the software you already run, with their own login and a person approving messages until you trust the pattern, as described on our trust page. If your current tool already asks every customer at the right time, you don't need us for this. If you can't say how many finished jobs got an ask last month, the free AI Operations Audit counts it and shows the gap.
