Auto Review Request + Bad Review Interceptor

The review problem local businesses don't realize they have twice

Walk into almost any small business, roofing crew, dental office, salon, and you'll find the same number sitting on their Google profile: somewhere under 50 reviews, often stuck there for years. Not because the work is bad. Because nobody has a real system for asking. A card on the counter that says "leave us a review!" gets ignored. A follow-up text sent whenever the owner remembers gets sent inconsistently, or not at all during the busy season, which is exactly when the reviews would matter most. The result is a profile that looks thin next to a competitor with 300 reviews, even when the actual work is comparable or better.

That's problem one: volume. Google's local pack rewards businesses with more reviews and more recent reviews, and customers themselves treat a business with 12 reviews as a coin flip compared to one with 150. Under that visibility threshold, a business is invisible in the exact moment a customer is deciding who to call.

Problem two is quieter and more expensive per incident. A single 1-star review, when a business only has 20 or 30 total, doesn't just sit there as one data point. It drags the average down hard, it sits at the top of the profile where every new prospect sees it first, and it stays there indefinitely because almost nobody writes a fresh positive review to bury it. One angry customer with a phone in hand, thirty seconds after a bad experience, can undo months of good service. And the business owner usually finds out about it the same way everyone else does: by seeing it live on Google, already public, already costing them clicks.

Most businesses treat these as separate problems needing separate fixes: a marketing push for more reviews, and damage control when a bad one lands. They're actually the same problem wearing two faces, and they have the same fix. Catch every customer at the exact moment their satisfaction is measurable, before anything goes public, and route them based on what they actually say. Happy customers get funneled toward the platform that helps the business. Unhappy customers get funneled toward the business owner, privately, while there's still time to fix it. That's the entire premise behind the Auto Review Request + Bad Review Interceptor blueprint on Prompt-King.AI, and it's why it works on both sides of the problem with one build instead of two.

Want to build this for a client? The full branch logic, the Base44 build steps, and the sales script are inside the Auto Review Request + Bad Review Interceptor blueprint on Prompt-King.AI.

How the two-stage branch actually works

The trigger is the part people underestimate. This isn't a tool that fires on a fixed schedule or waits for a human to remember to send something. It fires the moment a job is marked complete, and there are several different places that completion signal can come from. It can be manual, the owner or a staff member marking a job done inside the admin panel. Or it can be automatic, pulled straight from whatever system already tracks completion: QuickBooks closing an invoice, ServiceTitan or Jobber closing out a job ticket, a Stripe payment clearing. And notably, it can also come from any other tool already built for that client out of the same prompt library. If a client already has an appointment or job-tracking build running, that build's own "job complete" event can be wired straight into this one. You're not asking the business to change how they operate. You're listening for a signal that's already happening somewhere in their existing workflow.

Once that signal fires, the system doesn't text the customer immediately. It waits 90 minutes. That gap matters more than it looks like it should. Text a customer the second the technician's van pulls away and you catch them mid-transition, distracted, annoyed by the interruption regardless of how the job actually went. Wait 90 minutes and the experience is still fresh enough to rate accurately, but the customer has had a beat to actually notice the outcome: the AC is cooling the house, the paint looks right, the haircut held up through a shower. Ninety minutes is the default because it's the right average across a wide range of trades, but it's explicitly not universal. Restaurants need something closer to two hours, since a meal experience settles differently than a repair job. Dental practices need roughly the same. Medical spas doing Botox or filler need a full 24 hours, because the actual result of that service isn't even visible yet at the 90-minute mark. The blueprint bakes in vertical-specific defaults, but it's on whoever's building it to check the delay matches the actual service before launch, not just copy the default and move on.

The message itself stays deliberately simple: a plain 1-to-5 rating request by SMS. No ten-question survey, no NPS scale, no "tell us about your experience" open text field up front. Just a number, because a number is what the branch logic needs to make its decision, and because response rates on a one-tap rating request are dramatically higher than on anything that asks for more effort. The system also has to handle the customer who doesn't answer with a clean digit and instead texts back "amazing" or "terrible" or "it was fine I guess," which is why the build includes a keyword library sorting common word responses into positive, neutral, and negative buckets before falling back to an LLM call only when nothing in the keyword list matches. That ordering matters for cost too: keyword matching is free and instant, the LLM call is the expensive fallback, not the default path.

From there the branch splits, and this is the actual mechanism doing the work. A customer who rates 4 or 5 gets routed straight into a one-tap Google review submission (or Yelp, Healthgrades, or Zillow, depending on what platform matters most for that vertical), and their own positive SMS reply gets pre-loaded directly into the review text box. If they texted back "great job, on time and cleaned up after," that exact sentiment is already sitting in the composer when the review page opens. The customer isn't being asked to write a review from a blank page, which is the single biggest reason people who intend to leave a good review never actually get around to it. They're being asked to tap post on something that's already written in their own words.

A customer who rates 1, 2, or 3 gets routed somewhere else entirely: a private internal feedback form that never touches Google, Yelp, or anywhere public. At the same moment, the business owner gets an urgent SMS, and the build spec is specific about the speed here: within 60 seconds of the rating being parsed, with the customer's contact information and their complaint attached. That 60-second window isn't cosmetic. It has to fire in parallel with the customer-facing follow-up, not after it, because a business owner who finds out about an angry customer twenty minutes later has already lost most of the window where a phone call can still change the outcome.

And that outcome window is real: a 60-minute rescue period where the owner can personally call the customer and fix whatever went wrong before it ever becomes a searchable, permanent, public 1-star review. Not a canned apology text. An actual phone call from an actual human who owns the business, made while the frustration is still fresh enough to be resolvable and before the customer has had time to sit down and type out a review that will outlive the actual problem by years.

Two customers, two branches, one system

Picture a plumbing company that just finished a water heater replacement. The tech leaves at 2:00pm. At 3:30pm, the customer gets a text asking them to rate the job 1 to 5. She's already run the hot water and confirmed it works, so she taps 5. That single tap fires the positive branch: she gets a follow-up asking for a quick word on what went well, replies "fast, clean, explained everything before he started," and thirty seconds later she's looking at her own sentence already typed into a Google review page with nothing left to do but hit submit. Total time from her side: maybe 45 seconds across two texts. No blank page, no "what should I even say" hesitation. That's a review the business would very likely never have gotten through a paper card on the counter or a generic "please review us!" follow-up email sitting unread in a promotions folder.

Now picture a different customer from the same company, same day, different job: a drain cleaning that took longer than quoted and left a mess the tech didn't fully clean up. At 3:30pm she gets the same rating text and taps 2. That fires the other branch entirely. She's routed to a private form, not Google, so nothing public happens yet. At the same moment, the owner's phone buzzes with an urgent alert: 2-star rating, her name, her number, and whatever she typed into the private form about the mess and the time overrun. He calls her eighteen minutes later, apologizes directly, offers to send someone back out that week at no charge to finish the cleanup properly, and she accepts. The 60-minute rescue window closes with a resolved customer instead of a public review that would have sat at the top of the business's Google profile for the next three years, visible to every prospect who searched "plumber near me" and clicked through before calling anyone.

Same trigger, same 90-minute wait, same one-question text. Completely different outcome, decided entirely by which number the customer tapped and what happened in the minutes immediately after.

What it actually takes to build, and where it bites

This is explicitly one of the simpler builds in the library mechanically. The branching logic is a single fork off one rating value, not a multi-path decision tree, and the blueprint puts the expected build time in Base44 at 10 to 15 minutes. That's not a reason to treat it as trivial to sell or trivial to get right in production, though. The blueprint calls out a specific list of follow-up prompts a builder is likely to need, and they're worth knowing before you start rather than after a client calls confused about why something isn't working. That full prompt list sits inside the build steps on Prompt-King.AI, worth reading before you open Base44 for the first time.

The biggest one by far is the Google Place ID. The one-tap review deeplink only works if the URL is built correctly: https://search.google.com/local/writereview?placeid=[PLACE_ID]&review=[URL_ENCODED_TEXT], and most business owners have genuinely never heard of a Place ID and don't know where to find their own. That's a real onboarding step, not an assumption you can skip. Related to that: the customer's reply text has to be URL-encoded before it's inserted into that link, because apostrophes, ampersands, and line breaks in an unencoded customer reply will break the deeplink outright and send a customer to a broken page instead of a pre-filled review, with no error message telling anyone it happened.

Two more things tend to surface after launch rather than during the build. First, word-based ratings: customers who reply "amazing" or "terrible" instead of a digit need to hit a real keyword library covering positive, neutral, and negative language, matched case-insensitively, with the LLM call reserved as a fallback rather than the default parser. Second, notification speed on the negative branch: the owner alert and the customer's private-form follow-up need to fire in parallel, not sequentially, or that 60-second urgent-notification promise quietly turns into three or four minutes, which matters when the whole value proposition rests on the owner catching the customer while they're still reachable.

Who should build this, and the economics of selling it

The publicly listed pricing on this blueprint is straightforward: a $1,000 setup fee and a $297/month retainer. The stack behind it is deliberately cheap to run: an app builder plan (around $50/month for the backend functions this needs), Twilio for SMS at roughly $1.15/month per number plus about $0.008 per message plus one-time and monthly A2P 10DLC compliance fees, and an optional OpenAI or Anthropic call for parsing ambiguous replies that runs somewhere around $10 to $20/month for a typical business. That leaves real margin on the retainer even before accounting for the setup fee, and the value story that justifies the price to the client is built into the tool itself: 3 to 5x positive review velocity, and preventing over 80% of bad reviews from ever reaching a public platform, against a stated cost of $5,000 to $15,000 a year in bad-review revenue loss for a business that isn't running anything like this. Full pricing guidance, the exact sales script, and the phase-by-phase build steps sit behind a free Prompt-King.AI login, so anyone serious about actually selling this should create the account and read the full listing before pitching a client, rather than freelancing the pitch off the public numbers alone.

Who this fits best matches the "who you sell it to" guidance directly: home service trades like roofing, HVAC, plumbing, painting, landscaping, and pool service, plus restaurants, salons, dental offices, med spas, auto repair shops, and cleaning services. It's a weaker fit for law firms, where bar association rules restrict solicited reviews, and for B2B services generally, where the buyer isn't searching Google for a local provider the same way a homeowner is. The sharpest version of the pitch, per the blueprint, is aimed at owners sitting under 50 Google reviews: they're stuck below the visibility threshold where a searcher trusts the listing enough to click, and this is the tool that gets them past it in roughly 90 days.

Here's the part that makes this worth building even if review management isn't your first pitch to a given client: it's arguably the smoothest cross-sell in the entire library, because of that trigger list from earlier. Almost every other tool in a local-business AI stack, a booking system, an invoicing flow, a job-tracking app, a checkout, ends with some version of "this thing is now done." That completion event is exactly what this tool listens for. If you've already built and sold a client any other tool with a job-complete or purchase-complete moment in it, you don't need a second discovery call to figure out where this one plugs in. You already know, because you built the thing it plugs into. That's a genuinely rare property in a portfolio of local-business tools: most of them stand alone and need their own integration story. This one rides on the back of whatever you've already shipped.

Why this is worth building now

The blueprint itself flags something worth sitting with: this is one of the simpler builds in the library, ten to fifteen minutes in Base44, one branch, one trigger, and it's still explicitly a $1,000-plus-$297/month product because the outcome it produces is measurable and immediate. A business owner doesn't have to take your word for whether it worked. They watch their review count climb over the following weeks, and at some point they get the urgent text, make the call, and personally watch a bad experience turn into a saved customer instead of a permanent black mark on their profile. That's not a promise you have to sell twice. It sells itself the first time it fires.

Every week a business runs without this is another week of reviews going unasked, and another week where the next unhappy customer has a direct, unblocked path to posting publicly before the owner even knows something went wrong. The build is short. The Place ID step is the one real piece of friction, and it's a known, solved piece of friction with a documented walkthrough. There isn't much reason to wait on this one.

None of this needs new infrastructure if a client already has a job-complete signal running anywhere in their stack. Get the Auto Review Request + Bad Review Interceptor blueprint on Prompt-King.AI and have it live before your next client call.