There's a fine line between the app suggesting what to write and being a robot that just writes for you.

I'm torn between whether this is useful or depersonalizing:

WHEN MY EDITOR Joe told me to write this story, I knew with algorithmic certainty how to respond: “Done. Absolutely. It’s taken care of.” 

I got this advice from Crystal, a site that promises to help you understand how best to talk to any particular person. All you have to do is pick the subject. Crystal will then slurp up public data from around the web, run it through “proprietary personality detection technology,” and spit out a detailed report on that person’s preferred style of communicating. It’s one part oppo research, one part algorithmic astrology. It’s definitely creepy, perhaps useful, and almost certainly a look at how we’ll communicate in the future.

In the case of my editor, Crystal’s dossier was surprisingly accurate: “Joe is an achiever: fast-paced, ambitious, and persuasive, so get to the bottom line and don’t feel insulted by a direct or blunt comment.” When speaking to him, it told me to use words like “done,” “absolutely,” and “it’s taken care of.” In email, it recommended limiting my message to three sentences and stating my purpose clearly in the first line. After all, Crystal informed me, “it does not come naturally to Joe to be accommodating and forgiving with his time.”

Sorry boss—Crystal said it, not me.

Like anything that cultivates an association with magic, Crystal is less impressive once you know how it works. If someone were looking you up, the site would start by examining things you’ve written publicly—social media profiles being a primary source—and analyzing factors like writing style and sentence structure. Then it processes what others have written about you. Using those data points, the site identifies you as one of 64 communicative types, which the company has adapted from well-known personality frameworks. Crystal doesn’t really know you, in other words, it just knows what you’re like.

According to co-founder Drew D’Agostino, that can be enough. “The beauty of these frameworks is that, if you know one bit of data about a person and you’re accurate about it, you can make really good assumptions about how they’re likely to communicate,” he says. Building the model required considerable experimentation, but given a certain volume of writing, “we figured out a few algorithms that really nailed it,” D’Agostino says. In my experience, the results seemed a mix between useful characterizations and fortune cookie-like truisms.