An AI bat survey has uncovered a remarkable amount of hidden activity in England’s woodlands. Researchers analysed huge quantities of overnight audio recorded in forests and detected more than 98 million bat calls. The project does not mean that Britain suddenly has 98 million bats. One animal can call repeatedly, and the recordings capture activity rather than a population census. What the technology has provided is a much bigger, more detailed picture of where and how bats are using woodland habitats.
The national survey described by UCL on 30 October 2026 involved researchers at UCL and Edinburgh, the Bat Conservation Trust and Forest Research. The team examined recordings from 310 woodland sites, collected across thousands of nights. Without automated sound analysis, listening to that much material and identifying the brief calls of different bat species would have been an enormous manual task. AI has not made the animals more visible. It has helped researchers listen at a scale that would otherwise be difficult to manage.
What the AI bat survey actually heard
Bats navigate and find food partly through echolocation, making high-frequency sounds and interpreting returning echoes. Many of their calls are beyond ordinary human hearing, which is why researchers use specialist acoustic detectors. The survey deployed sensors at woodland locations and accumulated about 90,000 hours of recordings, equivalent to years of continuous listening. Its dataset amounted to 217 terabytes. The headline figure of 98 million therefore represents detected calls across a substantial research effort, not the number of animals living in a particular forest.
The project covered 9,734 nights of recording between 2021 and 2024. As the Bat Conservation Trust explained, the challenge was to find the relevant sounds inside that mass of data and work out which kinds of bats were responsible. Acoustic monitoring is particularly well suited to this task because the animals are often active in darkness when visual observation is difficult. It captures evidence that can be gathered without following each creature through the trees.
There is a satisfying inversion here. The forest may look quiet to a person walking along a path, yet the detectors reveal an active environment full of signals. That is what makes this a compelling application of AI. The computer is not inventing a synthetic wildlife scene. It is helping scientists interpret real recordings made in real places. The best part of the project is that the data can be connected to habitat information rather than remaining an impressive pile of recordings.
Why a call is not the same thing as a bat
A careful distinction sits at the centre of the findings. Acoustic detectors register sound events. A bat circling an insect-rich patch may produce many calls, and the same bat can be heard on successive nights. Activity also varies with weather, season and the placement of equipment. A high number of detections can show that a site is busy or attractive to a species, but it does not directly tell researchers how many individual animals were present.
The automated classifications also need scientific interpretation. Species can produce calls that overlap in their acoustic characteristics, while wind, insects, distance and background noise affect the recording. A model’s output is therefore a tool for organising evidence, not a magical label attached to every animal. Researchers must consider uncertainty and sampling conditions before turning detections into claims about distribution or conservation status. The public should be wary of any account that silently converts millions of calls into millions of bats.
This difference is familiar in wildlife science. A camera trap produces photographs, but repeated pictures of the same animal do not automatically measure population size. A sound detector can produce many recordings, but activity is not equivalent to headcount. Our earlier report on AI recognising birdsong shows how useful sound recognition can be when paired with realistic expectations about identification and what a device is actually measuring.
The surprising woodland findings
The study found that common pipistrelles were detected in every one of the 310 woodlands surveyed. Soprano pipistrelles appeared at almost all sites, and several other bat groups were widespread. Broadleaved woodlands showed particularly high activity on average, reinforcing their importance as places to feed, shelter and move through the landscape. Those statements concern the sampled English woodlands and cannot automatically be extended to every forest in Britain or to populations beyond the survey area.
There is practical value in knowing which habitats are being used. Woodland managers make decisions about tree cover, paths, dead wood and the condition of neighbouring land. Evidence about bat activity can help them understand how their choices may affect wildlife. The acoustic results were considered alongside information on woodland structure and the surrounding landscape, allowing researchers to explore patterns rather than simply produce a species list. The information could support better conservation decisions, while not dictating one answer for every site.
Even the widespread detections should not be mistaken for proof that bats are thriving everywhere. A species can turn up at many sites while facing pressures at important roosts or feeding grounds. Survey methods and time periods also influence which animals are heard. The careful conclusion is that the study provides a much stronger baseline for understanding activity in sampled woodlands. Future monitoring would help establish whether patterns change, which is different from declaring that the current survey has settled the conservation outlook.
Where artificial intelligence made the difference
The scale was the central obstacle. Humans can become skilled at reading recordings, but listening through enormous archives and checking hundreds of thousands of possible sound events is slow. Automated classification helps identify likely calls and sort them into meaningful groups for further analysis. That lets specialists spend more effort on interpreting the ecology and checking uncertain cases, rather than processing every second of raw audio from the beginning.
This is a useful contrast with the most theatrical promises about AI. The technology does not need to converse, write poetry or look human to contribute to science. It needs to recognise relevant patterns reliably enough that researchers can learn from a dataset. The same broad principle underlies work on AI listening for whales, although ocean soundscapes, species and recording methods are different. In each setting, listening at scale can reveal observations that manual work would struggle to collect.
There is still no shortcut around the quality of the evidence. Detectors must be placed sensibly, equipment has to function, and labels need to be checked. AI is only one part of a chain that begins with deciding which question matters. If the survey visits only a narrow set of habitats, a sophisticated model cannot make it representative of every environment. If a species is difficult to distinguish acoustically, a very large recording archive does not remove that ambiguity on its own.
What forests can learn from their night-time inhabitants
Bats make conservation questions tangible because they use a landscape in ways that people rarely see. A woodland that appears similar to its neighbour in daylight may support different levels of activity at night. Comparing detections with tree cover and surrounding habitat can offer clues about feeding opportunities and routes between sites. Those clues help researchers identify questions worth investigating and give land managers an additional way to judge whether habitat changes are having an effect.
The study is also a reminder of the value of collecting information patiently. Its recordings were gathered over several years before analysis drew the pieces together. A single listening session might produce an interesting snapshot; repeated measurements across many places can reveal much richer patterns. The role of AI is to make that careful, slow research manageable, not to replace it with an instant verdict. Related projects that extract wildlife relationships from birdwatchers’ observations show other ways computers can help reveal patterns in information collected by people.
The most memorable result remains the sound of a landscape usually thought of as silent. Another value is that recording methods can be described and compared, allowing later teams to repeat measurements in the same kinds of woodland. That continuity is essential if the findings are to guide conservation rather than stand as a one-off technical achievement. More than 98 million detected calls do not give us a neat count of Britain’s bats, but they provide an extraordinary record of night-time activity. If future surveys use comparable methods, conservationists will have a better chance of identifying where conditions improve or deteriorate. That is a far more enduring benefit than simply announcing that artificial intelligence has listened to a great many squeaks.
About the Author
Stuart Kerr is Technology Correspondent at LiveAIWire, covering artificial intelligence, cybersecurity and the social impact of emerging technology. LiveAIWire is an independent, human-led technology publication using AI-assisted research, editorial production and original AI-assisted editorial illustrations under his direction.