How to fix Amazon reviews: let me count the ways

How to fix Amazon reviews: let me count the ways

Amazon introduced its five star customer ratings system over two decades ago as a way to improve clickthrough and sales. At first, each book’s rating was the pure average of all its customer reviews. But no one ever said that system was perfect.

Six years ago there was a widespread scandal of ‘sockpuppetry’ in the world of book reviews. Then, three years ago, Amazon sued four sites selling reviews for pay, and began using machine learning to give books a weighted average biased in favour of new reviews, verified purchases and reviews people had marked as ‘helpful’. But in 2018, despite their huge influence on customers, Amazon reviews still suck.

It won’t surprise anyone that Amazon has opted for a standard tech solution to a human problem, but there are so many simple things they could do with their community, which would turn the fact that it is a group of people into a strength, rather than a weakness. Here are five suggestions:

1. Split reviews of the delivery and the product.

I can’t believe this hasn’t been done already. It’s so depressing to see someone’s gnomic whinge about the meaning of ‘standard delivery’ lumped in with an assessment of a story’s quality. It punishes the writer for the sins of the delivery guy. How can anyone equate packaging with imagination? Audible splits its reviews into story and narrator – why can’t Amazon do something similar?

2. Give reviewers a more transparent rating.

Rather than marking out the ‘Top 500 reviewers’, or marking reviews as ‘helpful’ and ‘unhelpful’, follow Uber’s lead and give the customers themselves a starred average. Don’t show reviews by those who abuse the system.

3. Only show reviews once a reviewer has posted more than three.

Airbnb doesn’t list a host’s rating at all until their third review gets posted, partly to give a fair average, but also to prioritise those who are engaging seriously with the platform.

4. If a reviewer hasn’t ‘made a Verified Purchase’ (or as a human might put it, bought it from Amazon), don’t allow their post to appear automatically.

Perhaps they have to write a substantial review of more than 100 words, as many bloggers do – it would make it harder for online communities to ‘brigade’ books for ideological reasons. Though not impossible.

5. Make reviews more social and sociable.

When Amazon bought GoodReads in 2013 for a reported $150 million it already had 16 million ‘users’ (an oddly narcotic expression for readers) on its site, and it now has 65 million. For all that they must have rejoiced in the extra user data and recommendation algorithms, they missed a really big reason that people like it: as a speculative purchase, books feel like a comparatively huge investment of time and money, and a lot of people will only buy a book that was recommended by someone they know and trust.

Some of these suggestions might actually make it harder to engage with reviews, and the idea of adding friction goes against the fundamental ethos of Silicon Valley (and, apparently, Seattle). It could mean some books have two reviews, rather than twenty. But if we are seeking quality, we should look to sites like Wikipedia, which allow anyone superficial engagement but which put up barriers to participation where it matters, with locked articles, forum discussions, roving admins and so on, to make sure not just that people are doing stuff on the site, but that it is a virtuous cycle, improving the community and the content. Because the worst thing about Amazon reviews is that, well, a lot of them just aren’t very good.