• Over 50% of consumer reviews are fake, bias, or unhelpful.

    *Data as of September 21, 2016 based on a random sample of over 1,000,000 Amazon reviews

  • Sponsored and fake reviews are incredibly bias.

    *Data as of September 21, 2016 based on a random sample of over 1,000,000 Amazon reviews

  • Amazon does not filter out these bad reviews.

    But FakeFilter does.

    See How

How It Works


  • 1. Remove all sponsored, fake, low quality, and unhelpful reviews.

    step1
  • 2. Individually score the remaining reviews on their quality and usefulness.

    step2
  • 3. Use weighted algorithms too create the most accurate star rating possible.

    step3

FAQ


Yes. FakeFilter will always be 100% free to use. No registration, email addresses, or anything like that. Just free and easy.

Amazon is under direct pressure from sellers to create a marketplace in which sellers can succeed. That is why Amazon allows reviews that are sponsored directly by independent sellers. Take a look at Amazon's top reviewers. If you browse through their reviews, you quickly notice that nearly all of them have received a product for free in exchange for their review.

While these reviewers claim no bias, our statistics show a very clear bias. So we remove these reviews as well as all likely fakes from our final star rating. Further, our star rating is weighted based on the quality of individual reviews.

These are reviews that are either fake or that provide no useful information to the consumer. Fake reviews are often paid for by sellers on Amazon and can be detected with advanced algorithms. Unusable reviews could include anything from a consumer who reviewed a product that they didn't actually buy, to a consumer who was upset with the seller and wasn't actually reviewing the product. We do not include these reviews ratings in our final evaluation of a product.

There is no black and white with fake or low quality reviews. A custom algorithm is used to score every review we analyze, and reviews that do not reach a certain score are considered unhelpful or fake. This algorithm evaluates a large pool of information. Here are just a few data points it looks at.

  • The spelling and grammar of the review
  • Sentiment analysis: is this a sales pitch or is there an actual critique of the product?
  • Was this review a verified purchase?
  • The review's author, and the author's previous reviews
  • The ratio of up to down votes on a review by consumers
  • The age of a review
  • Spikes of similar reviews published in a close date range
  • The length of a review
  • Aggregate category analysis to identify outlying writing styles and patterns

Authors disclaim that they received products for free in their reviews. Detecting these disclaimers requires a separate algorithm than what we use for detecting possible fake reviews. This algorithm is far simpler as it primarily consists of string analysis.

Unlike fake or unusable reviews, sponsored reviews are easily identified with a high degree of certainty.

After removing fake, unusable, and sponsored reviews out of a pool of reviews for a product, we calculate a rating with the remaining real and high quality reviews. There are two caveats to this.

  1. Reviews are weighted based on their quality
  2. Not all real consumer reviews are created equal. Some have pictures, in-depth analysis, and are based on months of real usage of a product. Others may have been written by consumers that have hardly used the product, and their review may contain very low quality information. For these reasons, we weigh real reviews such that higher quality reviews of any star rating will have a larger influence in our final FakeFilter star rating.

  3. We do not rate products if the sample size of real reviews is too small
  4. After filtering the fake and sponsored reviews, we often do not have a large enough sample size of real reviews to provide an aggregate rating. In this case we still showcase review statistics for a product, but we do not rank it or show a star rating for it.