One of the challenges of taking lots of photos of different animals (i.e. mammals, birds, reptiles, amphibians, fish, insects) is later identifying all the species you photographed. This can be a time-consuming task. On my Flickr page I have uploaded over 6,500 photos, mostly of wildlife, and I have attempted to identify all of them at the species-level. Of course, many species are easy to identify because they are common and/or easily recognizable, but I find there are often some that prove challenging. This can be the case for several reasons including i) not being familiar with the particular group of organisms, ii) several similar-looking species inhabiting the same region, iii) within-species variation in appearance complicating identification, or iv) key identifying features not being visible in the photo. This post details my strategies for identifying unfamiliar species using several real-world examples. Finally, I end by attempting this task using several Artificial Intelligence (AI) tools.
First off, why identify the species in your photos? For people who keep a life list, such as many birders, identifying the species that are observed and photographed is of the upmost importance. For scientists, correctly identifying species is important for documenting geographic range, conserving appropriate habitat, and even estimating population size. Finally, for those interested in natural history and wildlife there is the curiosity-driven urge to identify what is observed, thereby making better sense of the natural world.
Depending upon taxonomic group, my starting point is usually a trustworthy identification guide. I live in Florida, and for both freshwater fishes and amphibians and reptiles there are excellent guide books which contain high quality color images of every species in the state as well as range maps. Range maps can be really useful for distinguishing between similar-looking species, or narrowing the candidate species list – because of course you know where your photo was taken and can therefore compare this location to that of the range maps.
Pig frogs and Bullfrogs look similar but have somehwhat different ranges in Florida. Photo and map credit: Krysko et al. (2011). Specifically, Bullfrogs are not found in the southern part of peninsular Florida. This fact allowed me to more confidently identify the below photo as that of a Pig Frog. I took this photo in Big Cypress National Preserve at the very southern tip of Florida, where only Pig Frogs are found (and not Bullfrogs).
A Pig frog (Lithobates grylio) photographed in Big Cypress National Preserve, Florida.
Nonetheless, there are still occasionally cases where I am not able to identify at species level because two or more species look similar and overlap in range. One example of this is with freshwater turtles in Florida. Florida has 29 species of turtles. The majority are easily distinguishable but there are 3 species of Cooter turtles in the genus Pseudemys (the River Cooter, Florida Cooter, and Florida Red-bellied Cooter) that look quite similar and have broadly overlapping ranges. The primary criteria used to distinguish these large aquatic turtles are shell coloration differences that can be hard to identify in photos (for example, because the bottom of the shell isn’t visible and/or the top of the shell is obscured by algae growth). In this case, the best that can be done is to identify the turtles at the genus rather than species level.
Cooter turtles (genus Pseudemys) photographed in Florida. Since three species look similar and have broadly overlapping geographic ranges, I am not confident in identifying the pictured turtles at the species-level.
If I don’t have a detailed identification guide, I like turning to reference websites. For birds a good one is the All About Birds website, by the Cornell Lab of Ornithology. I’ll also usually try a google image search using relevant keywords such as ‘Florida hawk species’ if the bird is a hawk or ‘Florida heron species’ if the bird is a heron, or an even more general search in which I specify the taxonomic group (i.e. bird), the location (i.e. Florida), a distinguishing feature such as coloration (i.e. red), and the size (i.e. small). With the use of google image search I then simply compare the animal in my photo to those returned by the search and come up with a likely candidate or candidates. Once I have a likely candidate, I search specifically for this species in order to compare a variety of photos, and I also examine the range map to make sure it overlaps with where my photo was taken (Wikipedia is good for this). This process can be repeated iteratively for each candidate, thereby honing in on the likely species identification.
Another identification option, which I have not used, is to upload the photo to a website called iNaturalist, along with details such as the date the photo was taken and location. Other iNaturalist users will then identify the species in your photo. This identification will then be confirmed or challenged by other users, until a consensus is reached. Essentially, you are crowd-sourcing the species identification using (presumably) knowledgeable people with an interest in that particular group of animals.
Finally, a relatively new way to approach species identification is with the aid of Artificial Intelligence (AI). I decided to try this method using several popular AI Chatbots (ChatGPT, Grok, and Claude). I chose 5 recent photos as test cases. In one photo, I am 100% certain of my identification, in two I am quite confident in my identification (~90% certainty), and in two I am uncertain of the species-level identification as there are several similar looking species in Florida. I simply uploaded each photo and asked the AI Chatbot to identify the species. I also provided the geographic location (i.e. county in Florida) and time of year the photo was taken. My goal was to see what answer the AI Chatbots gave, what was their reasoning (if any), were these consistent with my identifications, and were these consistent with each other.
Photo 1
My identification: Hawk (likely genus Buteo).
Identification notes: There are 8 species of hawks found in Florida. Having looked at photos of the different species I think this is likely either a Red-shouldered Hawk (Buteo lineatus) or Red-tailed Hawk (Buteo jamaicensis).
Photo details: Taken May 31, 2026. Paynes Prairie Preserve State Park, Alachua County, Florida.
ChatGPT identification: Juvenile Red-tailed Hawk (Buteo jamaicensis)
Grok identification: Juvenile Red-shouldered Hawk (Buteo lineatus)
Claude identification: Juvenile Red-shouldered Hawk (Buteo lineatus)
Here we have a disagreement amongst the different AI Chatbots, so we need to take a look at the underlying reasons for the identification.
ChatGPT response:
Grok response:
Claude response:
All three Chatbots gave specific reasons for the species identification and it is hard to distinguish among these, as a non-bird expert. Therefore, I simply prompted ChatGPT with the follow-up message: ‘Grok and Claude identify this as a Red-shouldered Hawk, how confident are you this is a Red-tailed Hawk?’ And conversely, I asked Grok and Claude: ‘ChatGPT identified this as a Red-tailed Hawk, how confident are you this is a Red-shouldered Hawk?’
ChatGPT revised (and reversed!) its earlier assessment and now said the photo is likely a Juvenile Red-shouldered Hawk (Buteo lineatus) with ~65% confidence. Grok and Claude supported their earlier assessment that this is a Juvenile Red-shouldered Hawk (Buteo lineatus) with 90%+ confidence (Grok) and 75-80% confidence (Claude). All three Chatbots gave lengthy responses supporting the identification of Juvenile Red-shoulded Hawk, so this appears to be the most likely species identification.
Photo 2
My identification: Skink (genus Plestiodon).
Identification notes: This could be one of three species of similar-looking skinks in the genus Plestiodon found in Florida: Common Five-lined Skink (Plestiodon fasciatus), Southeastern Five-lined Skink (Plestiodon inexpectatus), or Broad-headed Skink (Plestiodon laticeps). Although, I think this is most likely a Broad-headed skink based on the number of scales on the upper lip and the large (broad) size of the head.
Photo details: Taken May 13, 2026. Sweetwater Wetlands Park, Alachua County, Florida.
ChatGPT identification: Broad-headed Skink (Plestiodon laticeps), most likely an adult female or subadult
Grok identification: Adult male Broad-headed Skink (Plestiodon laticeps)
Claude identification: Broad-headed Skink (Plestiodon laticeps), and specifically a breeding male
All three Chatbots gave the same species identification and similar reasons for the identification. This is most likely a Broad-headed Skink. Interestingly, ChatGPT gave an unprompted estimate of 98–99% confidence in its identification, which seems too high.
Photo 3
My identification: Southern Toad (Anaxyrus terrestris)
Identification notes: I am fairly confident this is a Southern Toad (Anaxyrus terrestris) based on general appearance, coloration, and geographic range.
Photo details: Taken May 12, 2026. Alachua County, Florida
ChatGPT identification: Eastern Spadefoot (Scaphiopus holbrookii)
Grok identification: Southern Toad (Anaxyrus terrestris)
Claude identification: Southern Toad (Anaxyrus terrestris)
Here again, as in the case of the hawk identification above, we have a disagreement amongst the different AI Chatbots. The identification of Eastern Spadefoot by ChatGPT is really surprsing given that the toad in the above photo looks nothing like an Eastern Spadefoot. Furthermore, the reasons given by ChatGPT for this being an Eastern Spadefoot are totally bogus and don’t match what is in the photo! In contrast the reasons given by Grok and Claude for this being a Southern Toad make sense and are similar to the features I used to identify this species. What is even more worrisome is that unprompted ChatGPT stated it has 99% confidence that this is an Eastern Spadefoot. This is undoubtedly a Southern Toad, and the fact that ChatGPT got this identification wrong, and with such high confidence, suggests the need to be very careful in trusting its output.
Photo 4
My identification: Least Killifish (Heterandria formosa)
Identification notes: I am certain that this is a female Least Killifish (Heterandria formosa).
Photo details: Taken April 13, 2026. Fanning Springs State Park, Levy County, Florida
ChatGPT identification: Eastern Mosquitofish (Gambusia holbrooki) female
Grok identification: Eastern Mosquitofish (Gambusia holbrooki), likely juveniles or subadults
Claude identification: Eastern Mosquitofish (Gambusia holbrooki)
All three Chatbots got this species identification wrong. ChatGPT did so while specifying it had 85-90% confidence, while the other two stated the purported species identification without specifying a confidence level. Interestingly, they all identified this as the Eastern Mosquitofish, another livebearing species found in Florida that is in the same family as the Least Killifish. The coloration pattern of Eastern Mosquitofish and Least Killifish is quite different, so it is disappointing that the species was mis-identified – especially given that each Chatbot listed coloration pattern as one of the ‘key features’ it used to identify the species. Eastern Mosquitofish are an extremely common and easily observed species whereas Least Killifish are both less common and surely much less photographed. Perhaps these Chatbots are biased toward the identification of more common species for which orders of magnitude more reference photos / information is available on the internet.
Photo 5
My identification: Marsh Rabbit (Sylvilagus palustris)
Identification notes: There are 2 species of rabbits found throughout Florida: the Eastern Cottontail (Sylvilagus floridanus) and the Marsh Rabbit (Sylvilagus palustris). These two closely related species differ in several ways: Marsh Rabbits are slightly smaller, have smaller ears, and have a small gray-brown tail that is dingy white on the underside (versus a highly-visible white puff tail in the Eastern Cottontail). I am fairly confident this is a Marsh Rabbit (Sylvilagus palustris).
Photo details: Taken May 28, 2026. Paynes Prairie Preserve State Park, Alachua County, Florida.
ChatGPT identification: Eastern Cottontail (Sylvilagus floridanus)
Grok identification: Marsh Rabbit (Sylvilagus palustris)
Claude identification: Marsh Rabbit (Sylvilagus palustris)
Here, once again we see a discrepancy in the species identification between the different Chatbots, and once again ChatGPT gives an absurdly high confidence estimate in its identification (98-99% certain). The explanation and key identifying features provided by Grok and Claude better match what is seen in the photo and are mostly consistent with the features I used to identify this as a Marsh Rabbit.
Overall, I was not particularly impressed with the ability of these AI Chatbots to correctly identify species in wildlife photos. I think one problem is that if you upload a photo and ask them to identify the species they seemingly always give an answer, along with definitive sounding reasons for this. However, I think they should (ideally) only return an answer if they are confident the answer is correct. If they are unsure about the identification they should either not give an answer (i.e. state that this cannot accurately be determined) or give an anwser and say they have very low confidence in the identification.
I don’t claim to know exactly what these AI Chatbots are doing ‘under the hood’ in order to make the species identifications that they do (or how they estimate the level of confidence in these identifications). But there is a platform called Wildlife Insights that claims “Our AI model is trained to recognize 1,295 species and 237 classes higher in the taxonomic tree from around the world”. This AI model has mostly been used to process and identify the wildlife on camera trap images. The platform states that their AI model is trained using a large number of wildlife images as a training data set. If the AI Chatbots that I tested (ChatGPT, Grok, Claude) are doing something similar, then perhaps their species identifications can only be as good as the reference wildlife images used for training them (i.e. presumably those found by scraping what is available on the internet). Maybe, in practice, this means they might be better at identifying (or baised towards) very common species for which there are thousands and thousands of high-quality photos on the internet. In the future, as reference libraries become better, perhaps AI-driven species identification will improve. Also, it seems likely that custom-built AI models for identifying wildlife in photos will greatly outperform generic AI Chatbots. Anyways, for the time being I’ll keep identifying the species I photograph the way I always have.
Reason and Nature 