Manipulating Consumption

Amazon has become the Internet’s product search engine…but those star ratings may not be what they seem.

(originally published on the Data for Democracy blog, and research featured in the Washington Post)

“$259,000: That’s the amount of money Amazon will log in the one minute it will take for you to read this. And in just one day, it will log more than $372 million — roughly 10x the median annual sales of the 1,000 largest online retailers in North America.” — Digital Commerce 360

Amazon influences what millions of Americans buy each day. It’s become the Internet’s product search engine — and because of the vast quantity of eyeballs and dollars on its platform, being on the first page of search results for a product can mean millions of dollars in revenue. Reviews shape consumption; people use the number of ratings and the number of stars as a way to make a quick determination of what products to click on, and what products to buy. The incentive to game the rankings is high: making it to first page of Amazon’s search results for a given product is incredibly lucrative, and it appears to deliver the secondary benefit of pushing a product up in Google search results as well. Manipulating reviews is a common tactic in the new SEO battlefield, since Amazon’s stars weigh heavily in its ranking algorithm — and social platforms make it easy.

The Amazon SEO Game
In the early days of Amazon there was a social, fun culture around reviews. Searchers could delight in discovering slightly strange — or extremely mundane — products that spontaneously became the target of thousands of lulzy comments. The Tuscan Milk Jug. The banana slicer. The book “How To Avoid Huge Ships.” A subreddit even emerged for funny Amazon review stories. But the star rankings and reviews that started out as a perk of the buying experience on Amazon gradually evolved into the currency that fed the search and recommendation engines that suggest products. As Amazon’s platform expanded into ever-more categories, and as it launched its marketplace, the reviews became much more important to sellers…and the spontaneous communing for lulz turned into coordinated brigading for economic, and sometimes ideological, reasons.

Tuscan Dairy Whole Milk

Stars
Amazon recognized that social recommendations were a key component of user purchasing decisions, and that fraudulent reviews undermined trust in Amazon’s brand. It started going after fake review brokers on sites like Fivrr, aggressively suing them. It even went after openly disclosed compensated reviews: in September 2016, independent researchers at ReviewMeta released a study showing that even disclosed incentivized reviews (when the reviewer gets the product at a significant discount or for free) were problematic because they consistently had higher star ratings than non-incentivized reviews for the same product. In October 2016, Amazon banned incentivized reviews. Problem solved? Nope — the incentivized reviews ecosystem simply moved underground.

Sellers determined to rise to the top of Amazon’s rankings needed to reach a captive audience who would continue to leave incentivized reviews under the table. Social platforms made it easy.

Despite the lawsuits and aggressive tactics, one thing Amazon has not done, it seems, is get directly involved in monitoring the social network ecosystem where reviews and brigades are brokered. Twitter groups. Facebook groups. Secret Facebook groups. Subreddits. Private messaging platforms. Even services like like Discord. Despite the fact that the practice is banned, the groups make no secret of what they’re doing. Search for “Review” or “Free Products”, and there they are.

Here’s how the process worked when we joined a few of these groups:

1. The seller, who is almost always operating a sockpuppet profile with a stock photo profile pic, posts a photo of the product along with the terms of the transaction: “Refund after review”, “Refund after order”, “Gift card”, “Bonus” (which means they’ll pay something extra for the review).
2. The reviewer/buyer comments on the post: “Interested”. At that point the conversation moves to direct messages.
3. In the DM channel the seller and buyer will confirm the terms of the review. The sellers frequently have a complicated series of rules: search for the product using specific keywords, add it to your Wishlist for two days, add two other products in the category to the Wishlist, too, to make it look like a real decision. Evade Amazon’s algorithmic fraud detection.
4. The seller will ask to see a link to the buyer’s profile to verify that they’ve left reviews in the past. Some require older accounts, Prime accounts.
5. The buyer completes the transaction according to the terms of the deal.
6. Money is refunded to the buyer’s Paypal account.
7. Other sellers will proactively DM users who’ve shown that they are “good” buyers, sending unsolicited product offers.

We ran through the process with a jacket, sweatpants, and headphones, each with slightly different terms; we initially left five star reviews, but subsequently amended them to levels more representative of the product quality (3 and 4 stars).

The closed Facebook Groups attract tens of thousands of members; Facebook does shut them down for violating standards, but it’s a game of whack-a-mole. There’s also a network of middlemen: accounts that appear to be American, apparently compensated as “finders” by the sellers, who are admins of secondary secret groups that they invite desirable reviewers into after establishing a relationship over DM. Searching keywords like “reviewers” only turns up the overt groups; we have no idea how many secret groups and underground economies are out there.

Navigating Amazon moving forward
How big of a problem is this? The Washington Post extended Data for Democracy’s investigation, looking into a few popular categories. Here’s what they found.

Motivated people are always going to try to game the system if the rewards make it worth it. Sellers from Alibaba who don’t have prominent name brands feel like they have to do it to move merchandise. Reviewers are motivated by getting free things; one notable post in a Facebook Group came from a Top 400 reviewer who boasted of getting $16,000 worth of free goods. Regardless of the justification, this is manipulation. It takes advantage of both honest sellers hoping for an even playing field and real buyers who are duped into buying low-quality junk.

The reality is that Amazon’s rating system and internal algorithms shape modern day consumption. No amount of Facebook Group deletions or Amazon lawsuits or reviewer profile wipes will de-motivate manipulators — with the degree of economic power consolidated on Amazon, the incentives are just too high. It’s a game of whack-a-mole; perhaps the best solution is an ongoing partnership in which researchers who observe these evolving tactics and campaigns report them to the platforms as they occur.

In the meantime, users who are planning to buy products on Amazon should check ReviewMeta to gauge the reliability of reviews, or consult external niche recommendation sites that have a reputation for real, honest feedback. Because the stars often aren’t what they seem.

Originally published at medium.com on June 29, 2018.