Content preferences’ is a term that’s used in marketing in several different ways for a variety of reasons. And if you’ve visited our website, you’ve probably noticed us talking about using your audience’s content preferences to predict the most engaging marketing assets.
When someone says ‘content preferences’ they might be talking about what type of content someone prefers to consume more generally, for example, are they more of a reader or a video watcher. They might also be talking about what topics audiences like, whether they like content focused on extreme sports or mindfulness, for example.
At Datasine, we define content in a much more granular way, all the way down to an individual image or sentence. For us, that means content preferences are the very specific elements within that content that users want to engage in marketing assets: what about it motivates them to ‘click’ – or not. In an image, that might be a particular colour that makes it stand out to them, for example, or the inclusion of flowers if they’re a gardening fan. For text, this could be a particular turn of phrase, or a feeling evoked by a sentence.
We gain knowledge into what users’ content preferences are by first understanding what features make up the content.
What are ‘content features’?
Content features are essentially all the parts that are present in each piece of content. These are identifiable in any kind of content (images, video, text, etc.), but we’ll be focusing on images for this article.
These features are how we as humans talk about and understand images: what is actually in it (e.g. physical objects) as well as how we perceive what’s going on in the image (e.g does it seem staged or natural?).
For example, we might identify the following features:
- This image has trees in it
- This image features the colour blue
- This image evokes happy feelings
- This image features a forest environment
By identifying content features (a process we at Datasine refer to as semantic content analysis), we’re able to extract and measure these attributes from each piece of content.
How a computer understands content features
Computers can be trained to identify these very human judgements using modern tools like machine learning, but it requires a bit of extra work. We first have to collect thousands of human judgements – or labels – on diverse marketing assets with text and images. The types of labels we’re interested in, such as whether an image is professional or casual, are informed by surveys with marketers, academic research into brand communication and proprietary research.
Once we have these labels, we can use powerful neural network models called convolutional and recurrent networks, to predict human judgements on new content, and these predictions are our features.
How content features can be used to establish content preferences
Once the content features have been extracted, we‘re able to use them to establish an audience’s or an individual’s content preferences.
To uncover what the user wants to see, we can analyse whether viewers engaged more when the feature was present or not. That way we can understand what is making people click.
If they engaged every time the image contained nature scenes, for example, they’re likely to prefer this kind of imagery in the future. So if you used natural imagery in an ad, that particular user or group of users would be more likely to click because it aligns with their content preferences. On the other hand, if they found lively imagery less engaging, they don’t want to see content containing this in the future.
At Datasine, we establish your audience’s content preferences using our AI model Connect, which offers three distinct levels of insight depending on the version you choose.
How content preferences work for Free
For the Free version of Connect, our AI model has been trained on data from a general audience, which we collect through a series of large online surveys. In this plan, we use our machine learning platform to understand the content preferences of the general audience.
Since the Free version has been trained on generic survey data, it’s not tailored to you or your brand, and it doesn’t consider the specific images you use to sell your product and it’s not based on the audience you’re targeting. Instead it offers helpful insights into what images will give you the highest click through rate for the average person, and is great for marketers primarily using diverse stock imagery to build early traction.
How content preferences work for Pro
With Connect’s Pro plan, the AI model is trained specifically on the engagement data of your audience. Connect combs through all your previous campaigns to see what features your audience have found more engaging, and offers distinct insights for your brand that will enhance engagement.
The suggestions Connect makes at this level are tailored to your brand, your specific imagery and your tone of voice. It helps you find the most engaging content for your average customer based on the content preferences of that specific group.
How content preferences work for Enterprise
The Enterprise version of Connect builds individual content preference maps for each customer within your CRM, and can use additional data such as orders and demographics to further improve performance. With individual-level preferences, Enterprise offers insights for segments of all sizes, from large audiences right down to specific individuals within your customer base. These preference maps are a valuable data source which can be used in downstream business applications.