Paste a review — we estimate manipulation likelihood from hype, emotional overload and lack of specifics, with no model. A rough guide.
Free with no sign-up, and the check runs inside your browser, so what you enter never leaves your device. The result is a signal to help you decide, not proof for money or legal decisions.
Estimate whether a shopping, app-store or restaurant review reflects real experience or a paid, exaggerated or fabricated one — using the language of the text. It flags the linguistic signals you miss when you look only at the star rating.
Fake reviews pack in emotional adjectives and superlatives ("the best", "life-changing", "highly recommend", "perfect"). People who actually used the product tend to mention small annoyances too (shipping, size, smell) alongside the good.
Lack of specificity is a key signal. "Really good, so satisfied" never explains what about the product is good, or why. And when many reviews reuse the same sentence template, mass templated writing is likely.
The score reflects "likelihood of manipulation/exaggeration", combining hype density, emotional bias, lack of specificity and templated structure. Detected hype phrases are shown as tags.
Watch the pattern more than any single review's score. Extreme 5-star (or 1-star) bursts in a short window, or accounts that only ever praise one product category, are far stronger signals than one sentence.
Reading reviews one at a time is often slower than looking at the shape of the ratings. A product with real sales behind it produces a gently skewed shape: five stars most common, four close behind, a scattering of ones and twos. Products with no unhappy customers at all barely exist.
Manipulation distorts that shape. All fives packed together with the middle empty, or a split into fives and ones only: the signature of a competitor attack met with a defensive push. Reviews clustered into a few days with nothing on either side of them is the same signal in the time axis.
Volume against selling period matters too. Hundreds of reviews two weeks after launch, almost none with photographs, means writers are running ahead of buyers. These layers sit outside what sentence analysis can reach, so they are worth checking even when individual scores come back low.
Fake reviews are usually transactions rather than personal lies. Products given free in exchange for a write-up, cash returned after purchase, accounts and reviews traded outright, and listings whose options are swapped so an unrelated product inherits a successful page — all of it exists as a market.
Many jurisdictions require paid or incentivised reviews to disclose that relationship, in a place the reader can actually see it, at the same size as the body text, in wording that is not ambiguous. Hidden behind a "more" link or slipped between hashtags does not satisfy that.
Turned around, this becomes a practical test. A review that discloses its incentive honestly has something going for it; unbroken praise with no disclosure anywhere is the more suspicious of the two. Checking for the disclosure comes before analysing the prose.
Two weeks after launch the listing had over three hundred reviews, almost all five stars, and photo reviews you could count on one hand. The sentences read naturally, so individual scores were not high. This type shows in the distribution, not the prose. An empty middle, reviews bunched into a few days, and a count out of proportion to the selling period mean writers arrived ahead of buyers whatever the per-review score says. Do not let the tool stand in for looking at the distribution.
Five five-star reviews of the same product were checked in sequence. "Fast shipping, great quality, highly recommend" appeared in each with only the word order changed, and none of them said what was good or why. Superlative density and template similarity rise together, producing a high score. One review alone is ambiguous; the same frame repeating across several is the strong signal.
"Performance is fine but the cable is too short so I bought another one" — praise and inconvenience in the same text, naming a specific part and a real action taken. The score comes out low. Paid reviews rarely volunteer a reason not to buy, so a concrete complaint reads as a trust signal.
A real customer writing "best purchase ever, obsessed, 10/10" pushes superlative density to the ceiling and scores as manipulated. The tool measures phrasing statistics, not sincerity. When this happens, drop the single-review score and look at the account instead: ten reviews in one day, or a burst of five-star ratings in a short window, tells you far more than any sentence-level number.
No. This tool only reads signals of exaggeration in the wording. Genuinely good products also rate highly. Use the score only as a reason to look twice.
One or two sentences lack enough signal and the statistics are unstable. Reading many reviews as a pattern is more accurate.
Not at all. A review written after receiving a product free is still useful when it names the downsides too. The problem is concealment. Ask which is more suspicious — a review that discloses, or unbroken praise that does not — and the standard becomes obvious.
Enter multiple reviews and they are compared against each other, with copy-paste-looking pairs flagged. That is where one person writing several near-identical entries shows up. The text itself is not stored; only the comparison comes back.