Knowing the evolution of activity even earlier can be crucial. This was especially essential during the Covid-19 health crisis, which caused sudden and large-scale economic movements. Press articles constitute an important source of knowledge to analyse the economic activity. The present work addresses the construction of NLP-based leading indicators of GDP, which uses daily press article tests. It extends S. Combes & al. (2018) who calculated a media indicator from a count of positive or negative words on articles in the daily journal Le Monde.