Spotted correctly? Human expertise remains essential in <scp>AI</scp> ‐assisted identification of Eurasian lynx

Abstract Identifying individual animals from camera trap photos is crucial for estimating population size and monitoring endangered species such as the Eurasian lynx ( Lynx lynx ). However, this process can be time‐consuming and prone to human error. Artificial intelligence (AI) platforms such as Whiskerbook offer automated solutions, but their effectiveness in real‐world conditions is not yet sufficiently evaluated. Therefore, we tested the performance of Whiskerbook's detection algorithm, compared the accuracy of identification algorithms (MiewID and HotSpotter) and evaluated the differences in the identification process between observers with different experience levels and the quality and colour of camera trap photos. We also tested Whiskerbook performance in identifying matching individuals across two monitoring programs within the same population. MiewID clearly outperformed Hotspotter as more encounters were successfully matched (64.0% vs. 45.5%). However, Hotspotter provided a lower average number of potential matches per query and was particularly successful in providing 100% correct matches among the first five suggestions. The identification accuracy of the MiewID algorithm operated by an expert was below the capabilities of an experienced observer performing manual identification (71.7% vs. 92.7%), while novices faced challenges in identifying felids even with the help of AI. Image quality significantly affected identification by all observers and the AI programs, and surprisingly, the AI performed better on black and white images than on colour images. Notably, Whiskerbook identified a transboundary individual that had previously been misclassified by the expert, demonstrating the value of shared AI databases for conservation across national borders. Our results highlight that while AI can support large‐scale monitoring, expert oversight remains essential to ensure accuracy. This is particularly important for conservation, where misidentification can lead to over‐ or underestimation of populations. We recommend the integration of AI with trained expert validation for sound wildlife research and management.