Browser Extensions 


More than just Fake Reviews: Why detects all types of bias

January 5th, 2017

At, we scan hundreds of millions of reviews and help you sort out the trustworthy ones from those that might not be completely honest.  The average consumer might mistakenly label anything that looks suspicious as a fake review, however textbook fake reviews are only a subset of the overall problem of biased reviews.

First, what exactly constitutes a “Fake Review”?

In today’s world, we see this phrase getting tossed around a lot more than it should.  The term “fake” should only really apply to reviews that are completely fraudulent: written by someone who has never used the product, or someone who has written multiple reviews for the same product.

For example, an Amazon vendor (we’ll call him Ben) wants to boost sales of his widget on Amazon.  He hires an underground 3rd party company to write positive reviews on his products.  He tells them which features to highlight in the reviews, how many 4 and 5 star reviews he wants and when he wants them.  He pays them $500 for their services, and a few weeks later has 50 positive reviews for his product.

Obviously these are textbook fake reviews, and something that looks for.  However, not all examples are quite so obvious.

Some reviews that aren’t exactly fake but still obviously biased.

For our next example, our Amazon Vendor, Ben, decides he wants to write a review himself.  He writes a thorough and detailed review based on his own experiences with his widget and publishes the review to Amazon.  He knows he is breaking the Amazon TOS, but he believes it’s still a valid review because he knows the product better than anyone else.

In this scenario, it’s very obvious Ben’s review on his own product has no place in the “customer reviews” section.  The massive conflict of interest is why Amazon doesn’t allow vendors to review their own products.  However, even though this review is clearly biased, would it still be accurate to label it as fake?  After all, ben did have a pretty thorough experience with his own product.

Either way, this is obviously an unnatural review and something we look for here at, but you can see where we’re going with the use of the word “fake”.

Are Incentivized Reviews fake?  No, but Amazon still banned them.

Up next is the highly-debated Incentivized Review.  On September 19, 2016, we published a video showing that consumers who receive a free or discounted product in exchange for a review were much more likely to leave a positive review.  Three weeks later, Amazon banned incentivized reviews, essentially confirming our suspicion.

It would be a stretch to label Incentivized Reviews as fake.  In almost all cases, the reviewer actually received the product (either for free or at a steep discount) and had some real, hands-on experiences with it before leaving a review.  These were real reviews, coming from real customers, however our data and Amazon’s policy both confirmed they weren’t worthy of being included on the platform anymore.

So, fake?  No.  Detected and adjusted for by  Yes.

Even seemingly honest gestures might result in massive bias.

For our final example, we look at something that seems to happen frequently for products with a loyal, cult-like following.  Our Amazon vendor, Ben, has built up a large, dedicated following for his brand on Instagram. One day, he decides to boost his product ranking on Amazon by posting a short video on Instagram asking his followers to review his widget.  In just 24 hours, he has over 50 new positive reviews for his widget on Amazon.

Would it be fair to label these reviews as fake?  Absolutely not.  We can safely assume all of these super-fans are real people with extensive experience with the product.  However, do the super-fans represent a balanced sample of customers?  Definitely not.  It’s highly unlikely that any of Ben’s Instagram followers disliked his product.  Just because Ben invested heavily in social media does not mean his product is any better than his competitors.  At we’re keeping an eye out for anything unnatural like this as well.

Just because a review fails one of our tests does not mean it is fake. is looking for much more than just textbook fake reviews.  We’re filtering out any reviews that appear unnatural: whether it’s from an incentivized reviewer, cluster of super-fans or a purely fake account.  Ultimately, it’s impossible to determine if a review is fake with 100% accuracy, and that’s why we use the term “Unnatural” in our reports instead of the word “fake”.

8 responses to “More than just Fake Reviews: Why detects all types of bias”

  1. HonoredMule says:

    The YouTube channel SmarterEveryDay has started using the term “inauthentic” around social media manipulation which I think does an apt job of capturing this range of problematic content, particularly without specifying any particular underlying underground marketplace mechanics, motivations, or bias.

  2. Theresa Draper says:

    This is a lot of discussion to just say “They are not fake, they are biased”, and, “These reviews were purchased with money or incentivized in some way”.
    Why not just say that?

    • I still think this is too simplified.

      • Biological says:

        I actually appreciate the level of detail in this write-up, and it establishes a greater level of trust in your process by seeing the types of things you guys consider. It shows that you guys have really thought about the phenomenon and ways to combat it.

        For example, the case with the Instagram followers – I didn’t even think of that, but after reading it, I’m assuming your analysis engine considers a sudden flurry of reviews posted within a short period of time (ex: suddenly there were 50 reviews over 4 days, ie. 12.5 Reviews/day, when previously it was consistently only averaging ~1.3 Reviews/day). I’m glad even these types of things have been thought of and, as a consumer, very greatly appreciate it. Thank you guys for all that you do – we all appreciate it!

      • VocalHero says:

        I agree, I also very much appreciate the detail in the explanation of how you guys go about determining what triggers your unnatural review “catcher.” Keep up the honest work!

    • VocalHero says:

      Why complain about a detailed explanation? No one forced you to read it, and what’s the harm in it? It’s not like the internet is running out of space to write words or anything, lol.

    • Tina McElroy says:

      Personally I wish we could use this technology to weed out replies that are only meant to irritate people. I mean really.

      Here I am having a pleasant morning, reading about this remarkable software
      that I knew nothing about. Just you know, people, exchanging ideas about dealing with this new type of scum. One guy admitted that this method was new and he appreciated the job the company was doing, you know just nice intelligent discussion about how our world is changing and just a good vibe you know.

      Then a person only replies seemingly just to insult someone who unlike herself can think outside of the tiny boxes that she lives in.

      Honestly. Can your software weed out this type of energy draining vampiric type replies. That would be very beneficial in humanities upward evolution.

Leave a Reply