Reviewing for Robots

This weekend, I happened to visit with quite a lot of good friends and family, and conversation turned (as it does) to Andrea's New Book, and how closely I'm watching my reviews. Why, they wondered, did I care about it so much? And lo I discovered that there is a common misapprehension about the nature and purpose of reviews on Amazon, Barnes & Noble, and Goodreads.

It's intuitively obvious that leaving a good review is helpful to a book and its author. If someone should find the book, the logic goes, all of those reviews will give a reader confidence and incentive to buy, right?

And that's true, to some extent... but that's not actually the big, valuable service you're performing for an author when you leave a review for their book. You're not reviewing for the benefit of other readers. The pivotal core audience for your review is computer algorithms.

Bookstore Time Machine

Back in the old-fashioned days, when you wanted to read a book, you'd hit a bookstore. You'd be tempted by various methods a book might have for spreading its fancy tail feathers and strutting its stuff: end caps filled with the new hotness, displays curated by the store's employees (if you like this, you'll also like...), standing displays and window displays and table displays. The books with the most prominent placement tended to sell the best. That's because visibility is a self-fulfilling prophecy. You buy the books you see and know about. ...We'll get back to that.

And then the shelving system itself would also guide you toward books you'd probably like. You'd make your way to the section of books similar to the books you already tended to enjoy reading, and make choices (consciously or not) based on the cover's color scheme, art style, and font choices. These all signal even today what to expect from a book, to help you decide if you'd be a happy reader or not. When you found some likely contenders, you'd pick up the book, read the back or inside flap copy, and decide whether it might be worth your time and money.

I daresay most book-buying doesn't happen like that anymore. I rely almost entirely on word of mouth from my internets. And I sell books that way, too -- not mine! But I've nudged many a friend into taking a close look at Naomi Alderman's Liar's Gospel, or Max Gladstone's Craft books, or Chuck Wendig's Atlanta Burns, or NK Jemisin's Inheritance series. I'm talking about The Grace of Kings so much I'll probably have sold a dozen copies for Ken Liu before I even finish it myself.

But that's not the only way books are sold. Not by a long shot.

Amazon's Algorithms

Today, the biggest obstacle any author has is obscurity. I can't sell you my book if you never, never see it or hear of it. And who decides whether a reader sees a book? Amazon, mostly. More specifically, the mysterious computer mind that is Amazon's recommendations engine. Goodreads, too; people do go there to find recommendations. But it's my understanding that there is some synchronization of reviews, since Goodreads was purchased by Amazon a few years back. And Barnes & Noble surely uses many of the same tricks.

But for an indie book or a small press, Amazon is the name of the game. That's 80% of your sales. So if nobody sees you on Amazon... basically nobody sees you.

I'm going to speculate here about the nature of Amazon's secret sauce. I don't know, of course; only people who work at Amazon specifically on the recommendation engine can know this, and they're at pains not to talk about it or lose their jobs and quite possibly be sued into oblivion for revealing trade secrets. But given what I know about information systems, metadata, and about how books sell and behave online, I think I can make a few really solid guesses.

So readers, think of it this way: when you leave a review, you are training Amazon as to what kind of book it is. And my guess is that it takes into consideration not just the stars you award, but your own buying and reviewing history, and keywords left in the review itself. Even a bad review is, I suspect, helpful to the book overall, because it means it's more likely to be shown to would-be readers who might enjoy it going forward, and less likely to be shown to readers with a history very much like the unhappy reviewer's.

I further speculate that none of this does much until the book reaches a critical mass of reviews -- there have to be enough data points for the algorithm to reach a solid conclusion. The system couldn't have much confidence in two five-star reviews from people with a history of only buying books from the one author, right? Even beyond that, there's a good shot that it's got some secret metric of reviewer credibility. We know Amazon prunes reviews left by authors writing in the same genre to prevent gaming for good or evil; so reviewer credibility is definitely on the Amazon radar. 

We also know for a fact that it tracks other books bought by the same readers. Right now, Revision has been bought by readers also interested in Myke Cole books, Jews vs. Zombies, and Vermilion. That's great news for those books, but the flip side is crucial, here -- it doesn't mean my book is being shown on those pages as an also-liked. And that is why books like mine need reviews: so they show up on the shelves next to the books kinda like it, so to speak, so that readers who enjoy that kind of thing know it even exists. It works. It really, really works. When my book A Creator's Guide to Transmedia Storytelling was paired with Spreadable Media, sales skyrocketed back to launch week levels for a while! 

That's because visibility is a self-fulfilling prophecy. So if you've read a book in the last couple of years that left you with a strong impression, tell people -- and leave reviews about it on the retailer of your choice. (But let's be honest, mostly Amazon). It's not just a nice thing that makes a writer feel good (or terrible).

A book lives or dies by the algorithm. And the algorithm can only know what it's been told.

Like my blog? Buy my books!

Get the Serial Box App for iOS | Android