AI-generated product reviews now flood Amazon, Etsy, Google Reviews, and every marketplace where five stars move the needle. Some are paid plants on new listings. Some are competitor hatchet jobs in disguise. Some are honest buyers who used ChatGPT to write up a real experience. Knowing which is which changes how shoppers spend, how brand managers respond to a listing dip, and how marketplace ops teams enforce review policy. This guide is the manual playbook. Five fake-review tells, a five-step workflow, marketplace context for Amazon Vine, Etsy, and Google Reviews, and when to paste the review into TextSight instead of guess.
Reviews drive billions in spending. A seller can prompt ChatGPT for ten enthusiastic five-star reviews in two minutes, post them across a week of stolen accounts, and watch a listing climb. The same playbook works in reverse for competitor hatchet jobs.
Shoppers want to buy what they think they are buying, and a fake five-star pool of reviews is the most expensive purchase mistake a marketplace can hand them. Brand managers want to spot rivals planting AI smears on their listings, and respond before the listing tanks. Marketplace ops and trust teams enforce platform policy, and Amazon explicitly bans AI reviews that misrepresent buyer experience.
A real buyer review almost always names how the product was used, where it sits in the home, what problem it solved, or what limit it hit. AI cannot use a product, so it praises the idea of the product instead. The use-context test outperforms generic-phrase tests on long reviews and works on short blurbs that no other test catches.
Eye-only screening on a single review sits around 65 to 75 percent above 150 words, falling fast on short blurbs. That is enough to triage a product page and route the worst offenders into a detector. For a takedown request or a brand-response decision, a calibrated detector with sentence-level highlights does the heavy lifting on top of the manual read.
Any single tell can show up in a real review written by a careful buyer. The signal is when two or more cluster in the same write-up, or when the same tells repeat across multiple reviews on the same listing.
The single most reliable product-review tell. Openers like "Great product!", "Love this!", "This product exceeded my expectations", and "Amazing find!" show up in roughly seventy percent of ChatGPT-drafted positive reviews. Real buyers almost never open this way. A real opener usually names what the product is for, who it was bought for, or a specific problem it just solved.
AI loves a uniform structure: three to five bullet points of capability-and-praise, each one a sentence long, each one starting with a strong adjective. "Easy to use." "Comfortable to wear." "Great value for the price." Real reviews have shape variation: one long anecdote, a short complaint nested inside praise, a tangent about delivery. Structural uniformity inside a review is the structural tell.
The cleanest single screen for any product review. A review of a coffee maker that never names the brew size, the carafe, the timer, or the cleaning cycle. A review of noise-cancelling headphones that never mentions a flight, a commute, or a song. Zero use-context specifics in a 150-word review is a verdict on its own, regardless of the other tells.
"Amazing", "perfect", "fantastic", "incredible", "wonderful", "great" stacked without naming the model, the size bought, or a feature actually used. The adjective stack is what AI produces when prompted to write a positive product review and given no real context. Real buyers anchor adjectives to concrete details a stranger could verify on the product page.
The second bookend. "I highly recommend this product", "I would definitely recommend it to anyone", "this is a must-have for anyone looking for", "you will not be disappointed". The closer cluster is the second half of the AI fingerprint. Paired with the generic opener in tell 1 and the missing use-context in tell 3, the verdict is settled inside a fifteen-second read.
Free covers around fifty typical reviews per day. Brand managers and marketplace ops teams running high volume usually run on Starter or Pro. Trust and safety teams automating screening start on Business for the REST API. Full details on the pricing page.
Billed $89.88/year — Save $30
Billed $179.88/year — Save $60
Billed $359.88/year — Save $120
Yearly billing saves 25%. View full pricing →
A workflow you can run on any product page without slowing the shop down. Two eye steps, one history check, one tool scan on the borderline reviews, one cluster check across the listing.
Open every suspect review with the first sentence in isolation. Generic openers like "Great product!" and "This product exceeded my expectations" are the single strongest fake-review tell. Real buyers usually open with a use case, a problem solved, or a specific feature. The opener carries roughly forty percent of the manual screening weight on its own.
Scan the body for anything only a real buyer could write. A model number, a room it sits in, a recipe it helped with, a flight it survived, a sibling product it replaced. AI reviews praise the idea of the product. Real reviews name how it was used, where, and how often. A single concrete use-context detail in the first 100 words is the strongest single human signal.
Open the reviewer profile. A brand-new account with only five-star reviews, all posted in the same week, on unrelated product categories is the textbook review-farm signature. A real reviewer has a mixed history across years with varying ratings, varying lengths, and at least one negative review somewhere in the past.
For listings that matter, paste the suspect review into app.textsight.ai for sentence-level highlights and a calibrated Authenticity Score. A 20-second scan confirms or overrides the manual read. The free plan covers 10,000 characters per day, which is around fifty typical reviews, and Pro at $19.99 a month removes the daily cap for brand-monitoring desks.
Check whether the same phrases, the same shape, and the same generic adjectives repeat across multiple reviews on the same page. A single AI-looking review is suggestive. Three reviews on the same listing that all open with "Great product!", all close with "I highly recommend", and all post inside the same week from new accounts is the verdict.
Each marketplace has its own version of the same five tells, its own review pool, and its own response decision. The use-context test outperforms across all three platforms, but the surrounding signals shift by venue.
Amazon publishes guidelines that prohibit AI reviews misrepresenting buyer experience and removes millions of suspect reviews each year. Vine reviewers receive free product in exchange for an honest review, and most write well, but a small share use AI for the first draft. On long-tail listings under a hundred reviews, the share of suspect AI text can run high, especially when a product is brand-new or sees a sudden burst of five-star activity.
Etsy runs on a smaller scale and the verified-buyer floor is higher, but new shops are heavy review-farm targets in the first month. Tells 1 and 5 fire hardest because Etsy reviews are usually short and the bookends dominate the read. The reviewer-history check is the strongest signal on Etsy, because Etsy buyers tend to have specific niches and a single five-star review across unrelated categories is unusual.
The hardest category and the highest stakes for local owners. Restaurant and service reviews are heavy targets because the listings drive foot traffic and pay-per-click value. Generic-opener and generic-adjective tells fire hardest here, because AI praise for a restaurant rarely names the dish ordered or the server who showed up. The use-context test is the cleanest single screen for any Google Review.
Different pattern, same tells. G2 and Capterra follow templated prompts, and AI-written entries struggle to name the specific integration tested, the team size, or the pricing tier. Generic-adjectives and missing-use-context fire even harder here than on consumer marketplaces, because real B2B reviewers cite concrete workflow details a stranger could verify.
A five-star review drafted by ChatGPT for a midrange wireless headphone listing with the prompt "write a 100-word positive review." Read it cold, then read the callouts.
"Great product! This product exceeded my expectations in every way. The quality is amazing and the design is excellent. I have been using these headphones and they are perfect for my needs. The sound is great, the comfort is fantastic, and the battery life is impressive. They work really well for everything I do throughout the day. The value is unbeatable at this price point. Overall, I highly recommend this product to anyone looking for a reliable option. You will not be disappointed."
"Great product!" plus "This product exceeded my expectations" is tell 1, the generic-opener cluster. The five short bullet-style claims about quality, sound, comfort, battery, and value with no concrete detail is tell 2, the uniform five-star bullet shape. Zero mention of a song, a call, a flight, a noise-cancelling mode, or actual battery hours is tell 3, missing use-context. "Amazing", "excellent", "perfect", "great", "fantastic", "impressive", "unbeatable" stacked is tell 4, the generic-adjective pile. "I highly recommend this product" plus "you will not be disappointed" closes tell 5, the templated closer cluster. All five tells in 92 words; a real buyer might trip one.
"Bought these for daily commuting on the Mumbai metro. Battery lasts the full week on my hour-long ride each way. Noise cancelling kills the train hum but leaves the announcements audible, which I needed. Bass is heavier than the Sony WH-1000XM4 I replaced; classical sounds a bit muddy but rock and podcasts are excellent. One gripe: the carry case is bigger than it needs to be." Sixty-five words. Three specific use cases, one specific comparison, one specific limitation, one specific routine. Zero tells.
The sister page for inbox triage. Five tells, a five-step screen, and the four inbox categories where the screen pays off.
Read the guide →The companion guide for sellers writing their own listings — fixing AI-flavoured product copy before it ships.
Open the AI rewriter →The main detector page. Paste a review, get sentence-level highlights and an Authenticity Score in one scan.
Open the detector →Full tier breakdown for Free, Starter, Pro, and Business. Yearly billing saves 25%.
See pricing →Free to try, no card. Sentence-level highlights, Authenticity Score, and a verdict you can route on.