A fake review used to be cheap and obvious. You paid a broker a few dollars per post, and the broken grammar, recycled phrases, and burst of twenty-five-star ratings on a single Tuesday usually gave it away.
That era is over. A language model can now write five hundred distinct reviews in a minute, each in its own voice, with invented details about the waiter's name or the parking situation. None of them repeat a sentence. Most of them read better than the real ones.
For anyone working in online reputation management, this is the biggest shift in the review economy since Google put star ratings in local search. It changes how attacks happen, how platforms defend against them, what regulators will punish, and what a legitimate ORM company should be selling.
How big the AI-generated review problem actually is
Hard numbers are still thin, which is part of the problem. The best measurements so far:
Pangram Labs analyzed 30,000 front-page Amazon reviews in 2025 and flagged 3% as AI-generated. Nearly three-quarters of those AI reviews were five stars. And 93% of them carried the "Verified Purchase" badge, the label shoppers treat as proof a review is real. (Inc.)
Three percent sounds small until you remember it is the front page, the reviews that actually move purchase decisions.
The platforms are removing fakes at enormous scale. Trustpilot took down 4.5 million fake reviews in 2024, about 7.4% of everything submitted, up from 6.1% the year before. (Trustpilot Trust Report 2025) Google blocked 292 million policy-violating reviews on Maps in 2025 and removed 13 million fake Business Profiles. (Search Engine Roundtable)
Consumers notice. Capital One research found 82% of shoppers encountered a fake review in the past year. (Digital Commerce 360)
Why the old detection tricks stopped working
For a decade, spotting fake reviews came down to language and timing. Fakes were short, generic, and heavy on superlatives. They clustered in time. They came from accounts with no history.
Language models broke the first half of that list. An AI review can be 140 words long, mention a specific dish, complain mildly about the wait, and still land on five stars. It reads like a real customer who had a good night. Text analysis alone can no longer separate them with confidence.
What still works is behavioral. Account age, device fingerprints, IP patterns, review velocity against a business's normal baseline, and whether the reviewer's location history makes sense. This is where the platforms have moved. Trustpilot says 90% of its fake review removals now happen automatically, using machine learning and generative AI. Google credits Gemini with catching violating content before it goes live.
So the arms race is now AI writing reviews against AI screening them. Humans reading reviews one at a time are no longer the front line.
The threat cuts both ways
Most coverage treats AI reviews as a problem of businesses inflating themselves. That is half of it.
The other half is attack. A competitor, a disgruntled ex-employee, or an organized extortion ring can now generate a wave of plausible one-star reviews for about the cost of an API call. Each one describes a different bad experience. Each one sounds like a person. The old defense, pointing out that the attack reviews were all identical, no longer applies.
Reputation teams see this pattern constantly with review bombing. The reviews arrive in a tight window, from thin accounts, describing visits that never happened. The text is fluent. The metadata is where it falls apart.
And then there is temptation on the business side. A small practice with eleven reviews watches a competitor with three hundred pull ahead in the map pack. Someone suggests using ChatGPT to "help customers write" reviews. It feels harmless. It is a federal violation.
Federal law now names AI-generated reviews directly
The FTC's Rule on the Use of Consumer Reviews and Testimonials, finalized August 14, 2024, bans reviews that misrepresent them as coming from someone who does not exist. The Commission's own announcement names "AI-generated fake reviews" as an example. (FTC)
The rule also prohibits buying fake reviews, paying for reviews that express a particular sentiment, posting undisclosed reviews from employees or executives, and suppressing negative reviews through threats. Critically, it gives the FTC authority to seek civil penalties against knowing violators, which it previously lacked in most review cases.
Enforcement arrived on schedule. The FTC sent warning letters to ten companies in December 2025. In April 2026, it settled with TruHeight over fake employees and incentivized reviews, resulting in a $4 million judgment and $750,000 paid. In May 2026, the FTC and the Illinois Attorney General sued Premium Home Service over thousands of fake business listings padded with fabricated five-star reviews. (DLA Piper)
Penalties are assessed per violation. When a model can generate a thousand reviews in an afternoon, the math gets ugly fast.
AI summaries turn review fraud into a search problem
Reviews used to be read by people. Now machines read them and tell people what to think.
BrightLocal's 2026 survey found 45% of consumers use tools like ChatGPT for local business recommendations and 82% read AI-generated review summaries. (BrightLocal) Omnisend found 63% of US consumers use AI while shopping. (Digital Commerce 360)
This matters because a language model summarizing your reviews extracts themes. If forty fabricated reviews all mention "rude front desk staff," that phrase can end up in Google's review summary or in a ChatGPT answer about your business, long after some of the fakes are removed. The summary has already been generated, cached, and repeated.
A fake review used to cost you a star. Now it can write the sentence an AI assistant uses to describe you.
What this changes for an online reputation management company
The job used to be mostly volume. Get more positive reviews, push the average up, bury the bad ones. That playbook is dangerous now, and any ORM company still running it is exposing clients to legal risk.
The work that matters in 2026 looks different.
Provenance over volume
Every review request should trace back to a real transaction. That means review invitations triggered from CRM or point-of-sale records, sent to actual customers, with no conditions on sentiment. If a regulator or platform asks where a review came from, the answer should be a timestamped record, not a shrug.
Anomaly monitoring
An ORM expert should know a client's normal review velocity and flag deviations within hours. Twelve reviews a month, then thirty-one in two days, is a signal worth investigating, whether the reviews are positive or negative.
Evidence-based removal
Platforms remove reviews that violate their policies, not reviews that are merely unfair. A strong removal request documents the pattern: account creation dates, the absence of any matching customer record, the timing cluster, the overlap in described details. Firms like NetReputation build these cases from metadata because fluent AI text no longer gives itself away.
Response strategy
Public replies are read by prospects and ingested by AI summarizers. A calm, specific response noting that no record of the visit exists does more work than silence.
Content beyond the review platforms
Because AI assistants pull from articles, profiles, and third-party mentions as well as reviews, the broader search footprint now shapes how a business gets summarized. Review management and content strategy have merged.
How to vet an ORM expert in the AI era
If you are hiring help, ask direct questions and expect direct answers.
- Have you ever written, posted, or generated a review on a client's behalf? The only acceptable answer is no.
- How do you generate new reviews? You want to hear about real customer outreach tied to real transactions. Be wary of any mention of "seeding," "profiles," or "review packages."
- How do you handle a suspected AI-generated attack? A competent ORM company will describe metadata analysis, platform escalation paths, and when legal action makes sense.
- Do you monitor how AI assistants describe my business? This is new territory, and firms that are paying attention will have an answer.
A firm that offers guaranteed star counts or a fixed number of reviews per month is telling you how it gets them.
The firms that last will be the ones that can prove it
Generative AI made fake reviews nearly free and nearly undetectable to the human eye. Platforms responded with their own AI. Regulators responded with a rule that names the technology outright.
That leaves online reputation management with one durable value proposition: authenticity you can document. Every review traceable to a real customer. Every removal backed by evidence. Every response written for both the person reading it and the model summarizing it.
The shortcut got cheaper. It also got much more expensive to get caught using it.
FAQ
What are AI-generated reviews?
AI-generated reviews are customer reviews written by large language models rather than by real customers. They are often fluent, detailed, and varied enough to pass casual inspection, which makes them harder to detect than older fake reviews.
Are AI-generated reviews illegal?
Yes, when they are presented as real customer reviews. The FTC's Rule on the Use of Consumer Reviews and Testimonials, finalized in August 2024, prohibits reviews that misrepresent that they come from a real person, and the FTC explicitly cites AI-generated fake reviews as an example. The FTC can seek civil penalties against knowing violators.
How common are AI-generated reviews?
A 2025 Pangram Labs analysis of 30,000 front-page Amazon reviews found 3% were AI-generated, and 93% of those carried a Verified Purchase badge. Trustpilot removed 4.5 million fake reviews in 2024, and Google blocked 292 million policy-violating reviews on Maps in 2025.
How do platforms detect AI-generated reviews?
Mostly through behavioral signals rather than text. Account age, device and IP patterns, review velocity, and location history reveal fakes that the writing alone does not. Trustpilot reports 90% of its fake review removals are automated, and Google uses Gemini to screen content before it is published.
What does an online reputation management company do about AI-generated review attacks?
A legitimate ORM company monitors review velocity for anomalies, documents metadata evidence that shows a review violates platform policy, files removal requests, writes public responses, and escalates to legal action when an attack is organized or extortionate.
Can an ORM company legally generate reviews for my business?
No ORM company can legally write or generate reviews on a client's behalf. An ORM expert can legally set up review request programs that invite real customers to share honest feedback, with no conditions on whether the review is positive.
