Here is how AI is changing content analytics

Get a sneak peek behind the scenes with the team that’s integrating artificial intelligence into the smartocto tool and see what the future of content analytics looks like.

Describing smartocto.ai is no easy task; it goes beyond a mere feature release. It's akin to opening a new door in your house that leads to an entirely different world – a world that can reach you, the client.

In this blog, various metaphors are employed, but let's start with an example of how an actual article undergoes transformation through the AI-driven machinery of smartocto. Showcasing a BBC article with the headline: 'Why Aperol Spritz is the drink of the summer.'

What follows is an engaging piece on the history of the drink. However, it can be challenging to follow at times, especially due to the abundance of dates and figures. Take this paragraph, for instance:

-------------------

In 1920, brothers Mario and Vittorio Pilla boiled, mashed and distilled 30 aromatic herbs over a period of nine months in Venice and created the drier Select. Campari, also the base of iconic Negroni drinks, with its distinctive floral and orange notes, was invented in 1860 at Gaspare Campari's bar in Novara, not far from Milan. And finally, the most bitter of all: artichoke-based Cynar, was conceived by Venetian entrepreneur Angelo Dalle Molle in a Padova distillery in 1948.

-------------------

If this article had been analysed by the smartocto tool (we can't use real examples from our clients), we know that readers might disengage at this paragraph due to the plethora of names, dates, and figures (e.g., 30 herbs in nine months). If the bounce rate for that section is notably high, smartocto can issue a warning to the client, if that’s what you’d like to be alerted about. This warning might look something like: ‘Be cautious with this paragraph; many readers may disengage, likely due to the abundance of names, dates, and figures.’

The next step, currently built by smartocto's Labs team, involves offering an alternative paragraph with just two clicks. It retains the same facts but presents them in a rewritten form. For instance, the paragraph might begin with: ‘About a century ago…’ instead of ‘In 1920’. Editors can either adopt these suggestions or use them as inspiration for their own improvements.

ChatGPT, or older tools like Hemingway and Grammarly, can also be employed for this purpose. “The key is that our tool combines multiple capabilities,” says Erik van Heeswijk, CEO of smartocto. “In addition to the capabilities of a Large Language Model, we have the training data from our own algorithm and a knowledge base containing various facts from our clients. The combination of machine learning and prompt engineering by the Labs department ensures that the output of foundational models like ChatGPT, Google’s Bard, or Amazon’s Claude aligns with the tone of voice and strategic goals of the medium.”

As depicted in the infographic above, the switch to AI-driven content analytics has resulted in more than just one feature. All other features receive an intelligent boost. Additionally, smartocto plans to provide an API that allows certain functionalities to be integrated into media CMS. This enables them to assess how suggested improvements might enhance their content even before publication.

Goran Milovanovic, senior data scientist from our Labs team describes smartocto.ai as being “an associative machine” for editors and other content creators. “It’s offering a spectrum of headline variations and strategic insights, thereby steering the article towards a desired narrative trajectory while maximising audience engagement and relevance.”

The technical advancements are being tailored to address specific editorial hurdles such as headline optimisation and content rewriting. “But we simultaneously focus on the implementation, security, and scaling of the system to ensure a seamless user experience and reliable performance”, Goran says.

Newsrooms have the final say

Given the novelty and potential excitement of this technology, Erik van Heeswijk believes it's crucial to roll out the technique gradually and always in consultation with clients. In a previous blog, the focus was on headline testing, emphasising that newsrooms have the final say in how they use AI suggestions as brainstorming aids.

Think of it as a highly intelligent colleague with whom you can endlessly brainstorm. This colleague doesn't need coffee and is tirelessly reliable. Importantly, this colleague always keeps the bigger picture in mind, as the considerations and suggestions stem from knowledge about strategy and policy. Erik emphasises that this knowledge is never shared with other media; the architecture ensures that each media outlet fills only its own knowledge base, never that of its competitors.

Some suggestions from the tool might not be journalistic or text-sensitive; they might be based purely on improving conversion rates or knowledge about timing and distribution. The promise of smartocto is that the notifications editors receive are 'actionable.' The Smartify feature is already AI-driven and can suggest the best article to feature in a newsletter at any given time. A yet-to-be-built virtual assistant could go a step further, suggesting, for instance, that it would be worthwhile to create a long read about a popular athlete to meet the goal for new subscribers this month. More on this virtual assistant will be revealed in next week’s blog.

_____________________

It’s smartoctober! Every year we spend a month deep diving into a subject or skill set that we deem to be absolutely essential to newsrooms around the world. This year it’s time to consider AI.

• Follow us on LinkedIn
• CEO Erik van Heeswijk reveals AI-driven improvements at smartocto in a webinar.
• Catch up on the first blog about AI in news media
• Read the second blog about how to start using AI
• And read the last blog: AI as virtual assistant for newsrooms

In summary, the use of generative AI is ushering in a new revolution. Just a decade ago, online media publishers had to interpret data themselves. The tools providing insight were like a map for a driver; you had to figure out where to turn left and right to reach your destination. Thanks to notifications based on machine learning, smartocto has become a navigation system that knows where you're heading and even considers traffic on the way. "If you extend this metaphor, smartocto.ai is an adaptive cruise control with lane assistance," says Erik.

He also wants to clarify what it's not: a self-driving car, meaning it doesn't replace the work of a journalist. You can ignore the lane assistance by steering yourself and are always responsible for the end result. Erik expects online media outlets to embrace the assistance and let AI contribute where they find it important and valuable.