Overview: I led a project for a major UK tabloid to evaluate how vulnerable their content is to disruption from large language models (LLMs) like ChatGPT. The goal was to assess which article types are most at risk of being replaced or bypassed by AI-generated responses, and to help inform editorial strategy moving forward.

My Role: I was solely responsible for the technical execution of the project, including data collection, LLM-based classification, survey analysis, and insight generation. This included:

Scraping hundreds of thousands of articles from the publisher’s website while filtering out ads and noise.

Using OpenAI’s API to categorize each article by the user need it fulfills — e.g. factual updates, emotional storytelling, or practical guidance.

Designing and analyzing public surveys to assess how likely users would be to consult an LLM for different types of information.

Calculating comparative risk scores for each content category.

Python (BeautifulSoup, Selenium, pandas, requests) for web scraping and data analysis

OpenAI API (GPT-4o) for semantic classification

Google Forms & statistical analysis for survey insights

Outcome: The findings revealed a clear trend: factual and utilitarian content (e.g. weather, event summaries, product explainers) is at high risk of being replaced by LLMs, while emotional, speculative, or gossip-based articles retain stronger engagement and defensibility. These insights have since been used to guide editorial planning and content investment.