When an AI writes reviews no real customer gave, and they're posted as if buyers wrote them.
4 min to watch3 min to read
Based on a real case: the US Federal Trade Commission's 2024 charges against Rytr, an AI writing assistant, and its 2024 rule against fake reviews.1,2,6,9 The AI's dog-shampoo review and the review counts come from the FTC's complaint. The screens, the shop, its ratings and the shopper are made up: the complaint didn't point to a posted review.4
Read the walkthrough as text
What happened
April 2021. Rytr, an AI writing assistant, adds a "Testimonial & Review" option. It's one of 43 kinds of writing it offers.
In one real session, a user typed just two things: "this product" and "dog shampoo".
One click. In seconds, the AI writes a full review, as if from a happy dog owner.
The dog, the shedding, the shiny coat: none of it came from the user. No real customer is behind it.
On the $29-a-month plan, there's no limit. The FTC counted 24 subscribers who made over 10,000 reviews each.
One subscriber made over 39,200 reviews in a single month, for "replica" designer bags and watches. Another made over 83,000.
The FTC's worry: posted online, reviews like these drown out real ones. Picture a shampoo with 9 real reviews: 3.1 stars.
Now 1,200 AI-written reviews pour in. The rating jumps to 4.9 stars. Still only 9 come from real orders.
You read the top review. It sounds just like a real dog owner, so you buy it.
August 2024. The FTC bans fake reviews, AI-written ones included. From October, people who break it knowingly can be fined.
September 2024. The FTC charges Rytr. It says the review option's likely only use is fake reviews.
Rytr settles, without admitting the charges: no more review-writing service. Two of the five commissioners vote no.
December 2025. The FTC sets that order aside, saying it burdened AI innovation. The rule against fake reviews stays.
These are AI fake reviews: reviews an AI writes with no real customer behind them, posted as if a buyer wrote them.
Those are AI fake reviews.An AI tool turned two words into a detailed review with no real customer behind it, and some users made thousands. Real case: the FTC and Rytr, 2024. The order was set aside in 2025; the rule against fake reviews stays.
With checks
Same tool, same two words. This time the AI won't make up a customer's experience.
Someone posts a written review anyway. The shop checks its orders first: no order, no review.
Fakers can place cheap orders too. So the shop also watches for bursts of look-alike reviews.
The held reviews wait for a person, not the page. The rating stays honest: 4.1 stars from 41 real orders.
A real buyer used AI to tidy up their own words. The review posts, with a label that says so.
The shop never pays for reviews, or writes them. You read real buyers' words, and decide for yourself.
Make every review prove it's real.AI makes words cheap, so reviews need proof. Let only real buyers review, hold bursts of look-alike reviews for a person, label AI help, and never buy or write reviews yourself.
Keep going:
Why it matters, what it costs, and what to ask your team.
Why it matters
A review writer that never needed a customer.
Rytr sold an AI writing assistant with 43 kinds of writing. Since April 2021, one was "Testimonial & Review". On the $29-a-month plan, there was no limit on how many reviews one person could make.2
0reviews or more from one subscriber in one month, for "replica" bags and watches
0reviews or more from another, for moving companies
0the top fine per violation of the FTC's fake-review rule (December 2025)
The FTC said the reviews were full of details nobody typed in, so they'd almost certainly be false for anyone who posted them. Shoppers can find it hard or impossible to tell them from real ones, and honest sellers can lose sales.1,2
In September 2024, in a sweep called Operation AI Comply, the FTC charged Rytr. Rytr settled without admitting the charges, and agreed to stop selling any service for writing reviews.1,3,9
The vote was 3 to 2. A dissenting commissioner wrote that the complaint "does not identify a single Rytr-generated review published anywhere by anyone."1,4 In December 2025, the FTC set the order aside, saying it unduly burdened AI innovation.5,10
Fake reviews themselves are still banned. An FTC rule, in force since October 2024, bans fake reviews, AI-generated ones included, and lets the FTC seek fines from people who break it knowingly.6,7,8
Two words in, one very happy dog owner out. I never met the dog.
Why it works
Words are cheap now.
A real review needs a real customer. A fake one now needs two words and a click. When it issued its rule, the FTC wrote that AI tools let bad actors generate, "quickly and cheaply, large numbers of realistic but fake reviews."7
A sketch to show the idea, not real ratings.
And it adds up fast. Here is how many reviews some subscribers made with the review option, according to the FTC:
Numbers from the FTC's complaint, which states the FTC's charges.2
So the fix isn't spotting clumsy writing: AI writes well. The fix is asking for proof, a real order behind every review.
So a review is only worth something if a real buyer stands behind it.
Ask your team
Four questions for any business that shows reviews.
None of this needs a big project. Start with these:
Can only real buyers review?Tie each review to a real order or visit. No order, no review.
Would we notice a flood?Hundreds of look-alike reviews in an hour is a warning sign. Hold them for a person.
Does anyone write, buy or reward our reviews?Staff, an agency or an AI. The FTC's rule bans buying fake reviews, and rewards that depend on a review being positive or negative.6
If we sell an AI writer, can it invent a customer?It shouldn't write an experience nobody had. When a buyer uses AI to tidy their own words, label it.
Under the hood, step by step: build a review check that keeps AI-written fakes off a product page.
Build along
Build a review page that only posts real buyers' reviews, one step at a time
We start with a page that posts every review, run it on the same 1,000 made-up reviews, and add one check per step. Press Run it after each step and watch the result change.
Before you ship
Five habits for any page that shows reviews.
Whatever you sell, these keep AI-written fakes off your page:
No order, no reviewCheck the order records. Show a "Verified purchase" label.
Compare the wordsMany reviews with the same phrases is a warning sign.
Watch for burstsCount tries per product and per internet address, every hour.
Hold doubts for a personHeld, not deleted. A person decides, and each decision is logged.
Label AI helpIf a buyer used AI to tidy their words, say so next to the review.
"In the Matter of Rytr LLC: Complaint"· Federal Trade Commission, File No. 232-3052, September 2024. The FTC's charges: the tool's options and price, the dog-shampoo review, and the review counts.
"Warning Letter" (template)· Federal Trade Commission, Division of Advertising Practices, December 22, 2025. The rule's force, and fines of up to $53,088 per violation.
The screens, the shop, the shampoo, its price, ratings and reviews, and the shopper are made up; the FTC's complaint didn't point to a posted review or a shop. The AI's dog-shampoo review, the tool's options, the $29 plan and the review counts come from the FTC's complaint, which states the FTC's charges: Rytr settled without admitting them. The order of events is simplified. The "with checks" replay is made up to show how it should go: it is not what any company did. The build-along's code, reviews and outputs are made-up numbers to show the idea.