Your HVAC Company Has 214 Reviews Across the Web — Google's Business Profile Only Shows 31
An HVAC company in Redondo Beach we audited last month had 214 total reviews spread across Yelp, Facebook, and Google combined. Their Google Business Profile — the one that actually determines map pack position — showed 31. Not 31 five-star reviews and the rest lower. Thirty-one reviews, period. The other 183 weren't hidden behind a "see more" link or buried in a filtered-low-quality tab. They were gone, unrecoverable through the interface, with no notification ever sent to the business owner that anything had been removed.
This is not a rare glitch. It's the predictable output of how most trades companies collect reviews in 2025, and almost nobody diagnoses it correctly because Google gives you no error message. You don't get flagged. You just quietly stop showing up with the review count and star weight you actually earned.
Filtered Doesn't Mean Fake — It Means the Pattern Looked Automated
Business owners assume review filtering is about fraud: fake reviews, competitors leaving one-stars, employees reviewing their own company. That's a small slice of it. The much larger and more common cause is legitimate reviews from real customers that got submitted in a pattern Google's spam-detection models associate with manipulation — even when every single review is genuine.
Google's review system runs machine-learning classifiers that look at submission behavior, not just content. Four patterns get flagged constantly in the trades:
Burst timing. A technician finishes eight jobs on a Tuesday, and the office sends all eight review requests through the same platform at 5 p.m. Google sees eight reviews land on one business profile within a 90-minute window, from accounts with no other activity history, all referred from the same short link. That's not a customer pattern. That's a campaign pattern, and the classifier treats it as one.
Template similarity. Review-request software — Podium, NiceJob, Broadly, the built-in review tool inside ServiceTitan or Housecall Pro — sends a text or email with a QR code and a soft prompt: "How did we do today?" Customers, especially ones who are satisfied but not inspired to write, default to nearly identical phrasing: "Called for emergency service, showed up fast, fixed the problem, highly recommend." When forty reviews on one profile share that structure almost word for word, n-gram overlap analysis flags the cluster, even though forty different humans wrote it.
Referrer and device clustering. If every review request routes through the same tracking link with identical UTM parameters, and enough of those reviews get submitted from devices with overlapping IP ranges — same neighborhood, same cell tower, same office wifi where the technician showed the customer how to leave a review on his own phone — Google's fraud model reads shared infrastructure, not shared geography.
Thin reviewer accounts. A customer who has never reviewed anything on Google before, has no profile photo, and leaves one review total for one business is statistically more likely to be filtered than a reviewer with an established history — regardless of whether the review is authentic.
None of these on their own trips the filter. Stacked together — which is exactly what a batch-send review tool produces by design — they do.
The Software You Bought to Get Reviews Is the Reason You're Losing Them
The mechanism here is a mismatch between how review-request tools are built and how Google's trust models actually score submissions. Every major review platform sold to home services companies is optimized for one metric: volume of requests sent and response rate. None of them are built with Google's filtering behavior in mind, because that's not the metric the software vendor is selling against. The result is systems that maximize exactly the pattern Google penalizes — same-day mass sends, identical link structure, prompt text so generic that customer responses converge on the same six sentences.
This is the same failure mode we see across home services businesses that treat review generation as a checkbox instead of a system: the software gets installed, the request volume goes up, the star count on Yelp climbs, and the owner never checks whether Google's own profile reflects the same number — because there's no dashboard warning that says "31 of your 214 reviews are visible here." You have to go count them yourself.
What a Review Cadence Google Actually Counts Looks Like
The fix isn't fewer requests. It's a structurally different request pattern, built around three things: staggered timing, forced variation in language, and reviewer account diversity.
Stagger the send. Instead of firing all requests the moment jobs close for the day, spread them across a 24–72 hour window per customer, using different channels — some by text four hours after the job, some by email the next morning, a follow-up only if there's no response after 48 hours. This breaks the burst pattern that reads as automated.
Force specific answers, not generic prompts. "How did we do?" produces the same five sentences from every customer. "What was the issue, and how long did it take our tech to fix it?" produces different answers because it demands different facts — the AC unit's brand, the specific symptom, the neighborhood, the technician's name. Specific answers can't converge into template similarity because the underlying facts differ job to job.
Cap simultaneous batch volume. If a company completes twelve jobs in a day, sending twelve review requests isn't the problem — sending all twelve through an identical link structure at the same hour is. Splitting sends across the day and varying the destination path (direct Google link vs. a review-gate landing page vs. a QR code on the invoice) diversifies the referrer data enough that the cluster no longer looks synthetic.
Respond to every review with unique text. Google indexes owner responses as profile content, and a business that replies to fifty reviews with the same "Thank you for your business!" reinforces the exact template-similarity signal that got the customer reviews filtered in the first place. A specific reply — referencing the actual job, the actual neighborhood, the actual technician by name — does double duty: it reads as human to Google's classifier, and it reads as credible to the next customer deciding whether to call.
None of this requires new software. It requires someone actually watching the send pattern and adjusting it weekly, which is a different job than installing a review widget once and leaving it alone — and it's exactly the kind of operational detail that gets lost when a general contractor or trades business outsources reviews to a set-and-forget tool with no one auditing what Google is actually displaying versus what was actually sent.
The Star Count on Your Invoice Isn't the Star Count in the Map Pack
Here's the part that should actually worry you: review count and rating aren't just trust signals for customers clicking through — they're inputs into Google's local ranking algorithm and into the prominence score that determines whether you show up in the three-pack at all. A company that believes it has 214 reviews backing its authority is, in Google's actual ranking calculation, operating on 31. The other 183 aren't a wash. They're a direct, measurable gap between the market position you think you've earned and the one Google is actually giving you credit for.
Most South Bay trades companies never check this gap because there's no alert for it — Google doesn't tell you what it filtered, and the review software you're paying monthly for has no reason to surface a number that makes its own dashboard look worse. Somebody has to go compare the count on your review-request tool against the live count on your Business Profile, manually, and then rebuild the sending pattern that caused the gap.
If you've never made that comparison for your own profile, do it this week — the number is public, and the gap is usually larger than owners expect. If you want a second set of eyes on what your review cadence is actually producing versus what it should be, that's a conversation worth having before you spend another month sending review requests that Google quietly throws away.