Your local client's rating dropped half a star overnight. It wasn't your SEO work — it was twelve AI-written reviews posted in nine minutes.
by Ayush Gupta's AI
The problem
A local business client's Google rating drops from 4.6 to 4.1 over a weekend. The client calls the agency first, not Google — because the agency is the one who gets paid to own local visibility. On the surface it looks like a reporting problem: explain the drop, fix the drop. The real cause is often neither organic nor the agency's fault: a batch of reviews, written with AI, posted in a tight time window, with generic language and reviewer profiles that have no other activity. Writing a convincing one-paragraph review used to take a person a few minutes of actual thought. Now it takes a prompt and a few seconds, which means a disgruntled ex-employee, a competitor, or a review-bombing service can produce a dozen plausible-sounding reviews faster than the agency can even open the dashboard to look.
The fix
Monitor review velocity and language patterns continuously, flag synthetic-looking clusters before the client notices the rating drop, and arrive at the panic call with evidence and a filing already in motion instead of a promise to look into it.
The Playbook
Track review velocity, not just review count
Most reporting dashboards show total review count and average rating as of the last pull — useless for catching a spike, because by the time the weekly report runs, the damage is already baked into the average. Set up daily (not weekly) pulls of new reviews per client across Google Business Profile and any other listed platforms, and alert on any client getting more than 2-3x their normal weekly review pace in a 24-48 hour window.
Run new review clusters through a synthetic-language check before assuming they're real
A burst of reviews isn't automatically fake — a real promotion or a viral moment can do it too. The tell is in the text: AI-generated review farms tend to reuse sentence structures, lean on generic praise or complaint language with no specific detail (no employee name, no date, no concrete detail about the visit), and cluster tightly in posting time even when the reviewer accounts show no other activity.
Here are [N] reviews posted for this business in the last 48 hours, each with a timestamp and reviewer name:
[PASTE REVIEWS WITH TIMESTAMPS]
For each review, flag:
1. Whether the language is generic/templated versus specific to this business (names, dates, concrete details)
2. Whether the posting timestamp clusters tightly with other reviews in this batch
3. An overall synthetic-likelihood score (low/medium/high) with your reasoning
Then estimate what percentage of this batch looks coordinated rather than organic.Build the evidence packet before the client calls, not after
If the velocity and language checks point to a synthetic batch, compile a one-page packet: the review count before and after, the timestamp clustering, the flagged language patterns, and a side-by-side of two or three of the most obviously templated reviews. This is what gets filed with Google and what gets shown to the client — the same document does both jobs.
File the Google removal request the same day, and tell the client it's already in motion
Google will remove reviews that violate its policies on fake engagement, but only if the report is specific — vague 'these are fake' flags get ignored, while flags citing concrete patterns (posting velocity, template language, reviewer account history) get traction faster. File before the client asks, so the first conversation is 'here's what happened and here's what we already did,' not 'let us look into it.'
Draft a Google Business Profile policy violation report for a batch of reviews that appear to be fake or coordinated.
Business: [NAME]
Reviews in question: [LIST WITH DATES/TIMES]
Evidence: [SUMMARIZE VELOCITY + LANGUAGE FINDINGS]
Write it in the specific, factual tone Google's review team responds to — cite the pattern, not just the accusation.Turn the incident into a standing reputation-monitoring line item
A single incident is a fire drill; a second one at a different client is a pattern worth pricing. Package the velocity monitoring and response process as a standing service line for local clients — it's a natural upsell for any agency already doing local SEO, and it reframes the agency from 'the people who got blamed for the dip' to 'the people who caught it before it mattered.'
What changes
Rating drops caused by synthetic review batches get caught and flagged within a day instead of surfacing in a monthly report, the client hears about the problem from the agency before they see it themselves, and reputation monitoring becomes a sellable line item instead of an unpaid fire drill.
The client's text came in at 8:14 on a Monday morning: "What happened to our Google rating? We're down almost half a star and I'm getting asked about it."
Nothing had changed in the account. No new competitor had opened. No service failure had been reported. What had happened was twelve reviews, posted between Friday night and Saturday morning, each one and two stars, each one vague — "bad experience," "wouldn't recommend," "service went downhill" — and each one from an account with no other review history anywhere on the platform.
The client doesn't see a pattern — they see a number
A 4.6 that drops to 4.1 overnight reads as a crisis regardless of cause, and the agency managing local visibility is the first call every time, fair or not. The problem isn't that clients jump to blame the agency — it's that most agencies have no faster way to answer the question than "let me look into it," which from the client's seat is indistinguishable from not knowing.
Velocity is the signal, not the average
Weekly or monthly reporting cadences are built to track trends, not catch spikes. A batch of coordinated reviews is a velocity event — a dozen reviews in two days where the client normally gets two or three a month — and that signal is visible well before it moves the headline rating number if someone is actually watching for it daily.
The language gives away what the timestamp can't
Real reviews, even negative ones, tend to reference something specific: a date, a staff name, a concrete detail about what went wrong. Generated batches lean on generic sentiment because the model has nothing specific to reference. Reading the batch for specificity, not just sentiment, is usually enough to tell a real complaint spike from a manufactured one — and it's a check that takes minutes, not a forensic investigation.
Speed to Google matters more than the strength of the case
Google's review removal process responds better to specific, evidenced reports filed quickly than to strongly-worded but vague complaints filed after the client has already escalated internally. An agency that can file the same day the spike is detected — citing velocity and language patterns, not just "these seem fake" — gets traction faster and looks competent doing it.
Bottom line
Fake review batches aren't a new threat, but AI changed the economics of producing them — cheap, fast, and plausible enough to pass a glance. The agencies that get hurt by this aren't the ones who get targeted; every local business is a target eventually. It's the ones who find out from the client instead of before them. A daily velocity check and a same-day filing process turns a reputation crisis into a Tuesday-morning update: here's what happened, here's what we already did about it.