Merlin, a free app that listens to birdsong and identifies which bird is singing, will soon automatically send its findings to eBird — a massive global database where millions of people log bird sightings. This is the first time an AI-powered bird-listening tool at this scale will feed directly into such a large, centralized bird record. eBird holds over 2 billion bird observations and is used by researchers and conservationists worldwide.

Merlin is already popular. It has been downloaded 40 million times across 240 countries. In just six months — from December 2025 to mid-2026 — it gained 7 million new users. The app uses artificial intelligence (AI) to recognize bird species from their sounds alone, making it easy for casual bird-watchers without formal training to contribute useful data.

eBird, created and run by Cornell Lab of Ornithology, is the world's largest organized collection of bird observations. Scientists, conservationists, and governments rely on it to track whether bird populations are growing or shrinking across entire continents. Adding millions of Merlin's automated identifications to eBird in real-time could fill gaps in bird monitoring — especially in parts of the world where expert bird counters are hard to find.

A Key Concern: How Accurate Is the AI?

However, the plan comes with a real problem that experts have not solved. Bird scientists in Europe have warned that Merlin should not be used for official bird counts because it makes mistakes. Merlin once identified a dachshund — a dog — as a mallard. Background noise and non-bird sounds can fool the AI.

As of July 2026, Merlin can recognize 2,066 bird species. It works well in the US, Canada, and Europe. But it is less reliable in tropical regions and parts of Asia, where it has seen fewer examples of bird sounds during training.

Cornell Lab has not yet explained publicly how it will handle this accuracy problem. When Merlin sends identifications to eBird, will they be mixed in with identifications made by human observers? Or will they be kept separate and marked as "AI-generated"? This choice will decide whether the data is trustworthy enough for serious scientific research.

Why This Matters for Wildlife

Bird populations are crashing. Experts in Britain estimate their country has lost over 70 million birds in the past 50 years. Around the globe, falling bird numbers are one of the clearest signs that ecosystems are in trouble. More bird observations — even imperfect ones — can help scientists spot a local bird population collapse before it becomes catastrophic.

There is a real tension here. When a human submits a bird observation to eBird, scientists know who that person is and can ask them questions if something seems wrong. But AI makes identifications without a name attached. To check an AI's work, you would need to keep the original recording and record which version of the AI software made the identification. We do not yet know if Cornell Lab will do this.

Cornell Lab already links Merlin and eBird through a single user account, so the basic connection is ready. What happens next — whether AI records get clearly labeled, whether they are kept separate, and how thoroughly the system is checked — will determine whether this integration genuinely improves bird science or just adds noise to the database.