BirdNET provides superior diversity estimates compared to observer-based surveys in long-term monitoring

• Spectrogram analysis outperforms other methods in 10-minute surveys. • Automated detection underestimates species richness in short surveys. • BirdNET on all-day recordings detects twice as many species as 10-minute surveys. • Species richness across surveys is weakly correlated. • Long-duration automated surveys provide the highest cumulative diversity estimates. Bird population monitoring is crucial for understanding biodiversity trends, with various methods differing in effectiveness. This study compared traditional observer-based methods, including point counts and manual analysis of passive acoustic monitoring (PAM) recordings, with automated species detection using BirdNET during springtime in a European primaeval forest, Białowieża National Park in Poland. We evaluated 97 10-minute surveys from each method and input all-day recordings into BirdNET to explore the differences in alpha richness and beta diversity at both the survey level and cumulative over the entire breeding season. The results show that BirdNET detected the highest alpha richness (80 species) when processing all-day recordings. However, its performance during short-term surveys was lower (49 species) compared to that of human observers, with recording analysis yielding the highest species count among the observer-based methods (61 species). Beta diversity also differed according to the site and method used (p < 0.05). The number of species detected during all short-term surveys showed a weak correlation with the number of species obtained from all-day BirdNET analysis (Pearson’s correlation: 0.34–0.49), questioning the adequacy of short-term surveys in species-rich habitats such as primaeval forests. Our findings demonstrate that monitoring methods should be adjusted to improve coverage. Automated analysis offers scalability and efficiency but still requires significant time dedicated to processing, so manual review remains essential for validating automated classifications and ensuring data accuracy. On the other hand, point counts and recording analysis can provide valuable insights in situations where time is limited or when finer verification of species detection is necessary.