Content Relevance & Personalization

With over 800 news sources, including hundreds of local news sites, Scouted Media needed a way to figure out which articles were relevant to which users. To do this, we looked at the problem in two dimensions:

1. The places mentioned in the article
2. The reach/significance of the article (e.g. “national news”)

If we could assign that metadata to each article, we could then cross-reference the content with the user’s location.

We created an automation which sent article snippets to OpenAI’s gpt-5-nano model via Azure AI Foundry, with a prompt which asked the model to list:

1. Places mentioned in the article (either explicitly, or implicit)
2. GPT’s best assessment of whether the article had global, national, regional or local significance.

We chose gpt-5-nano because of its low cost and rapid response times. By batching dozens of articles at once, we could perform this classification task quickly and for only pennies per day.

The location names provided by GPT were fed into the open-source GeoNames database, which allowed us to associate a latitude and longitude for each place mentioned in each article.

The data enrichment provided by GPT allowed us to then create customized news feeds for every user based on their location. When a user opens the Scouted app, they are automatically presented with:

• World news (Article significance is “world”)
• National news (Article significance is “national” and mentions a place in the user’s country)
• Regional news (Article significance is “regional” and mentions a place in the user’s state/province/etc)
• Local news for their city (Article significance is “local” and mentions a. place within 50km of the user’s location)