One day, we will all write for machines
Earlier this month, Facebook announced it has changed its news feed algorithm once again — this time, to weed out “clickbait” posts. “People expect the stories in their feed to be meaningful to them,” Facebook stated in a blog on its Best Practices page. “When the headline of a story is missing information, people tend to find that misleading, sensational and spammy.”
Over at Fortune, a quick retort was prepared. “The problem for Facebook is that, as several of its algorithm-change posts have admitted, clickbait headlines tend to get a lot of clicks. But in surveys, 80% of readers say they would rather see ‘genuine’ posts, and not those with trick headlines,” Mathew Ingram wrote. “So should Facebook watch and alter its behaviour based on what people do, or based on what they say they want?”
This is a good question. We might ask the same thing from the perspective of content creators, and news organizations in particular, as they struggle to maintain readership. That is: should they watch Facebook and alter their behaviour based on what it does or on what it says it wants?
As Ingram notes, Facebook is inherently designed for people to share information — technically, of any kind. That is what it does. And yet, it says it wants to ensure that only certain kinds of information and stories get shared. For those writing the words that appear on Facebook’s news feeds, this clearly poses a challenge.
But there is more to it than that.
This struggle within Facebook highlights a deeper problem. It’s one we might all face very soon when writing just about anything online, whether it’s for news organization or a personal blog, or even when we’re composing an email or text. The real question Facebook is making us ask here (and the real question that has been underlying pretty much every question about online content, viral stories, clickbait, etc. for years) is one about whether we should write for human behaviour or computer behaviour.
In other words, this is about more than just clickbait.
This is actually about whether we — all of us — will in the future write so that humans like reading what we write, or so that an algorithm does. This is about how we will communicate.
Let’s talk about Yoast.
Yoast is a Search Engine Optimization (SEO) plugin. It’s not the only one, but it’s a popular one. Yoast embeds itself into the Wordpress platform. What’s important is not which platform the SEO plugins operate on; what’s important is what they do. SEO plugins analyze content and determine how easily it will be found in a search (read: when people Google something). If you’ve never seen one in action, Yoast looks like this:
As you can see, Yoast uses traffic light colours to note elements of the text that will help its SEO (green) or those that might hinder it (yellow, red).
The point here is not that Yoast automatically makes posts into clickbait; it doesn’t.
Instead, the point rests on some of those green lights you see in that example above — the ones that highlight “keyword density” (how often the keyword appears in the text), the length of the post, and the score Yoast reports the post has received on the Flesch Reading Ease test. Though it’s not shown here, Yoast also commonly notes average sentence length.
The expectations, or recommendations, that Yoast has for these elements in particular will affect how the content appears to a reader. And for our purposes, that makes them the most important.
Because, if you managed to fulfill all of Yoast’s SEO expectations — that is, responded to the behaviour of this computer program (which in turn is only responding to the behaviour of another computer program — Google’s) — you might have an article or piece that a lot of people can find in a search, but that is also weirdly composed. It would likely be very short. It would be fairly repetitive. It would contain no sentences longer than 20 words. And it would be written in a style suitable for an 11-year old, with very few words containing more than two syllables.
I got an email the other day that told me I could write better emails. More specifically, this message told me that I could write the kind of emails people actually reply to. The email was from Baydin, the makers of Boomerang, an email assistant. When it first appeared a few years ago, the selling feature of the Boomerang plugin was that it would allow users to schedule emails. This was particularly appealing to Gmail users, since the service was (and still is) unable to do that.
The message I got this week was about how Boomerang can now do more. It was about how Boomerang is now all about A.I.
“Boomerang Respondable uses machine learning algorithms trained on hundreds of millions of messages to isolate what factors impact response rates,” the email explained. “It understands how different aspects of your writing combine to affect the quality of the emails you send.”
This means Boomerang Respondable will analyze what I’m writing in an email, and encourage me to change things in order to optimize my chances of having someone reply to it. All in real-time, as I type.
Respondable says it focuses on a short list of factors when it considers how likely it is that someone will reply. Those are: Message subject length; message word count; reading level; question count; positivity; politeness; and subjectivity.
When installed, it apparently looks something like this:
Much like Yoast, Respondable offers suggestions if it detects what it perceives to be a problem. The example provided in the email is familiar.
In case you missed it, they’ve highlighted it: “Surprisingly, emails written at a 3rd grade reading level… are most likely to get a response.”
The makers of Respondable say they are “excited about the prospect of using machine learning to provide actionable writing advice based on data,” and claim this is “the first time in history that there has been a way to bring that analysis to bear on your own writing in an accessible way.”
Most brazenly, given that it suggests ways to limit things like word count and and pare down word complexity, Respondable justifies its existence in your email not merely by claiming that you will receive more email responses, but because “it helps you unlock your creativity.”
People read a lot online. Or, at least, they’re online a lot, looking at words. How many of those words are getting read remains a key issue. Back in 2013, Farhad Manjoo asked Chartbeat to show him how many people read past the top of a Slate article. The results were discouraging to anyone writing anything online.
It stands to reason that news organizations and bloggers, or even people who email a lot, would want some assurance that their words are being read. What makes these apps and plugins so appealing is that they profess to reveal some secret code. Yoast, Respondable, Facebook’s news feed (as well as Facebook Instant Articles), and their ilk are popular because they offer as close as we can come to a guarantee that more of what we want to say will make it to the people we want to say it to.
But then, what are we saying? And, just as importantly, how are we saying it?
Facebook has a laudable goal of showcasing posts that are meaningful to people in a way that goes beyond the temptation to merely click through. This may or may not signal that, behind the scenes within the company, someone still believes the quality of language (how it is used and where it directs us) is important, and that it will continue to be. Of course, what matters more is whether the rest of us really feel that way.
Maybe there is no turning back. Maybe there should be no turning back? In any case, it is maybe also worth taking a moment to wonder what communication will look like when we ultimately hand it all over entirely to the machines to, as they might say, unlock our creativity.