AI models can do a lot of the hard work for content creators
Imagine that I have a website with three main audiences. People who love swimming, running, and hiking.
Also imagine that I have a customer data platform, so I can identify individual users and keep tabs on what content they like.
Since I’m able to do that, I can put people into audiences – e.g., swim fan, running fan, hiking fan. Of course there might be some overlap between audiences, which is fine.
Within each audience, I track what’s popular and what’s trending. But not only that. I can check to see if there’s a difference in content preference between the highly engaged visitor and the drive bys and the casual readers.
At this point I might have six collections of content – what the highly engaged fan is reading in each of my three audiences, and what the not-so-engaged are reading. Or watching or listening, right? Obviously I can divide this further – like between video and text content – but let’s leave that aside for now and just focus on the content as content.
I’m going to pick the highly engaged swim fan as my example.
I grab all that content and pull out the keywords, tags, categories, and so on. I’m not going to restrict myself to the keywords and tags assigned by the editors. I’m going to use natural language processing to scan these articles and come up with another type of categorization. Additional words to add to the mix.
At this point I need to go on a slight tangent and talk about how AI views words.
A word like “king” is represented in an AI system by a multidimensional vector. Something like [0.2, -0.4, 0.7, …] There might be hundreds of dimensions, and the values are assigned so that words with similar meanings are located close to one another in this multi-dimensional space. But they might be close along one axis and not on another. “King” and “queen” are close in the context of being a ruler, but they’re not close in the context of sex, while “king” and “duke” are close in the context of sex, but a little farther apart in rank.
These sorts of vectors allow math, like king – man + woman = queen.
AI models don’t understand anything. They just have a complicated mathematical representation of words and phrases that are derived from processing huge amounts of text.
Let’s get back to swim fans. Once I have a collection of all the keywords that are popular with my swim fans this week, AI can find similar words and concepts and start to build a pretty interesting model of what sorts of topics swim fans care about right now.
What’s next?
You could just generate a report for the writers showing what types of concepts are playing well with the swim fans. The benefit of using AI here is that this report isn’t just a matter of keywords, like goggles and freestyle and sample workouts. You can get sentiment analysis, like “fun, light-hearted articles are more popular,” or “challenging articles are more popular.” There’s any number of factors you might want to consider.
This is analytics on steroids.
Your writers can use this analysis to come up with ideas for next week’s articles.
Or we can take this all the way to the scary level and ask AI to write the articles based on the information you’re getting from your data.
Realize what we’ve done here. We’ve come up with a technological way to listen to our audience, discern what interests them, and then give them more of what they want.
While I was going through all this you might have had the thought that there’s a danger of getting into a spiral, where you go deeper and deeper into a narrower and narrower set of concepts. That’s a possible consequence of this kind of work, so it might be a good idea to build mechanisms to prompt people outside of their stated preferences – at least to some extent.
For example, Spotify might get the idea that I only like songs similar to the ones I’ve listened to. That’s usually how recommendations work. But it’s a good idea to toss in the outlier from time to time.
Anyway, there you have it. A system like that could revolutionize content creation. Your weekly e-newsletter could express exactly what your audience is interested in.
