Implementing AI for content generation for a global media company

In this case study, we will describe the implementation of artificial intelligence (AI)-based tools and algorithms for a large media company with more than 10 brands, where the main source of revenue is advertising sales.
We briefly describe some context for this case. The company has several products and services that need to be sold. We can use, for example, the Amazon Deals service to collect the most favorable and interesting offers: it shows which products are currently discounted and for which product referrals can get the maximum benefit.
The task was to develop an algorithm or method that would allow us to create a lot of simple and uniform texts (about 100 texts per week). We also discovered how well an AI can handle such a task.

Project Overview:

• Location: USA
• Product: Website
• Technologies: React.js, Node.js, Scala
• Team: Back-end Developer, Front-end Developer
• Timeline: since 2021

Writing 100 review articles in a week requires a large monetary outlay, so we decided to use ChatGPT, but this approach has several problems that took a significant amount of time to complete:

• Creating texts by working with ChatGPT in a question-and-answer format turned out to be disadvantageous, as it is important to get all the necessary data in one query.
• From the previous point, the prompt becomes more complex, and as a result, the neural network may move away from the required text structure. This requires the clearest possible instructions of what the text structure will be. It should have a good headline and a convenient response format: a product description in a couple of sentences and a brief explanation why this product is useful to the customer. That’s why the first thing we did was adjusting the prompt to generate text with the required structure.
• ChatGPT had a problem with adding the correct link to the right product from the Amazon catalog without changing it. This also took some time and writing clear instructions. The last problem that is still being worked on is the tone of voice of the neural network that generates the texts. The results were filled with excessive praise of the product and filled with cliched vocabulary.
  

These tasks were worked on in co-operation with the product team. An administrative panel was created for them so that they could experiment with the introductory part of the prompts needed to customize the tone of voice. Developers focused on creating clear instructions related to the structure of the response and automating the results checking so that low-quality texts with incorrect structure would not be allowed to be published on the website.

After experiments with integrating OpenAI tools we created an automated solution that can help with writing a lot of simple and uniform content, which started to earn thousands of dollars a week for the media company. Of course, it takes a huge amount of traffic to make that kind of money, but it was there. After implementing this tool, employees can now write about 100 article reviews per week, as well as handle more complex tasks. This case study perfectly shows how a media company can make money on new technologies and, in particular, on AI tools.