Plant identification? There’s an app for that—actually several!

Erin Hill<hiller12@msu.edu>, Michigan State University Extension, Department of Plant, Soil and Microbial Sciences - May 04, 2026

2026 update: three AI platforms added! Plant identification apps for smartphones have seen significant improvements over the past several years. Now AI offers to take a photo and get an instant identification, but are they reliable?

Students working on fact-checking plant IDs — Figure 1. Students working on plant identification assessment of photos taken around MSU’s campus. Photo by Erin Hill, MSU.

We are driven to identify plants for many reasons; sometimes it is a curiosity about the world around us, other times it is out of the desire or need to manage areas like gardens, agricultural fields, restored habitats, and/or natural preserves. Plants are the foundation of food webs and they are tied to our understanding of how ecosystems function. Plant identification has been and continues to be a matter of familiarity, knowledge passed down through mentorship by family or friends, or perhaps something learned in school. One can also seek expert advice.

Plant identification is one of the many services offered at MSU Plant & Pest Diagnostics and the MSU Herbarium. Help is also available through Michigan State University Extension via the Lawn and Garden Hotline (1-888-678-3464) or Ask Extension.

A bit of history

There are several smartphone apps available to assist with plant identification. I began evaluating plant identification apps in 2018 for use in the Weed Science Laboratory class at MSU (i.e., CSS226L, now CROPS226L) and for presentations to various garden and commodity groups.

From 2018–2020, I evaluated a minimum of six apps (available for both Android and iOS smartphones) using 10-12 plants, with the best performing apps carrying over to the next year’s evaluation. In the fall of 2021, I drastically increased the number of identifications used to rank the apps by involving groups of university students enrolled in the lab.

The 2025 test

Since 2018, the students and I have evaluated a total of 17 apps (see complete list at the bottom of the article), and in 2025, we added three AI platforms. In this most recent test, all the apps and AI used photo recognition and often geolocation information to identify plants, but some tested in the past required more descriptive input from the user, similar to traditional plant keys. Most are free or have a free version. It is important to read all terms before downloading or purchasing apps and using AI platforms. This assessment is for educational purposes only. Reference to commercial products or trade names does not imply endorsement by MSU Extension or bias against those not mentioned.

The students in the weed science lab learn plant identification skills as an integral part of the curriculum, and the addition of the app/AI evaluation supplements traditional methods. Lab groups were asked to photograph 10 plants during our field trips on MSU’s campus to Beal Botanical Garden and the Hancock Turfgrass Research Center (Figure 1, USDA Hardiness Zone 6a). The identity of each plant photographed was confirmed with me prior to completing the exercise.

Photos were not edited and aimed to realistically replicate the average person’s input. It should be noted that many of the apps provide detailed descriptions or videos on how to best take photos to increase the chances of success. We did not discuss specific instructions prior to the class testing the technologies, reflecting the average user’s likelihood of reading instructions.

Identifying weeds at all life stages and various growth habits is important for our students when making management decisions. Therefore, groups were required to photograph one plant from each of the following categories: flowering broadleaf ornamental species (Figure 2), flowering and vegetative broadleaf weeds (Figures 3 and 4), flowering and vegetative grass (or grass-like) weeds (Figures 5 and 6), an Amaranth or grass seedling (Figure 7), a winter annual seedling (Figure 8) and one deciduous and one evergreen tree or shrub species (Figures 9 and 10, respectively). The last plant was their choice, but it was recommended they use common agricultural and turf weeds.

Daylily flowers blooming along a white fence. — Figure 2. Daylily (Hemerocallis spp.) was an example of a flowering ornamental. Photo by Kendall Dumaw and Micah Long.

A White clover flower. — Figure 3. White clover (Trifolium repens) was an example of a flowering/fruiting broadleaf weed. Photo by Dion Myers, Elizabeth Othmer, Rayana Slack.

Buckhorn plantain growing from the ground. — Figure 4. Buckhorn plantain (Plantago lanceolata) was an example of a vegetative broadleaf weed/plant. Photo by Liam Mehren, Elizabeth Smith and Jake Smith.

The flowering grass Giant foxtail. — Figure 5. Giant foxtail (Setaria faberi) was an example of a flowering grass. Photo by Henry First, Drew Morr and Derek Schornack.

Barnyardgrass growing from a sandy ground. — Figure 6. Barnyardgrass (Echinochloa crus-gali) was an example of a vegetative grass. Photo by Nicole Daniel, Alex Hull and Adam Sanborn.

Two side by side images of palmer amaranth and creeping bentgrass. — Figure 7. Depending on the class, students either took a photo of a seedling amaranth, like Palmer amaranth (left, Amaranthus palmeri, photo by Kimberly Abrego, Grace Beem and Seeun Sohn), or a seedling grass, like creeping bentgrass (right, Agrostis stolonifera, photo by Gage Koso, Dan Miller, Patrick Neustadt and Austin Schafer).

Hairy bittercress growing amongst ground debris. — Figure 8. Hairy bittercress (Cardamine hirsuta) was an example of a winter annual seedling. Photo by Gage Koso, Dan Miller, Patrick Neustadt, Austin Schafer.

A Sassafras tree. — Figure 9. Sassafras (Sassafras albidum) was an example of a deciduous tree. Photo by Robert Knapp, Gabriel Schnieder, Carl Seidelman.

A Yew evergreen tree. — Figure 10. Yew (Taxus spp.) was an example of an evergreen tree. Photo by Carter Brandt, Dillon Morlock, Kaleigh Verhage.

All groups were assigned the same three smartphone apps and three AI platforms to evaluate in 2025 (Table 1); sometimes, they only chose two of the AI platforms in the interest of time. Within the categories mentioned above, the students entered the same photos into each app or AI platform and recorded the resulting identifications as correct, partial or incorrect compared with our previously agreed-upon identifications.

Table 1. Plant identification technologies evaluated and ranked by the MSU weed science laboratory students in fall 2025 (1= best performing based on percentage of “correct” identifications).

App name	Placement
Seek-App	6
Copilot-AI	5
Gemini-AI	4
Google Lens-App	3
ChatGPT-AI	2
PictureThis-App	1

Correct identifications were 100% accurate for the scientific name (i.e., Latin genus and species, e.g., large crabgrass is Digitaria sanguinalis). Partial was assigned when a technology correctly identified the plant to the genus level, but not to species (i.e., a photo of green foxtail was identified as Setaria faberi, or giant foxtail, instead of the correct Setaria viridis, or green foxtail) or the correct answer was not the highest suggestion on the list for technologies that make multiple suggestions. These answers may still be helpful towards successful identification but are not entirely correct. Incorrect was designated if neither of the previously mentioned were true.

Using this process, each of the apps and AI platforms was evaluated 136 to 203 times in 2025, up to 21 times per category. The apps and AI platforms were ranked based on the comparative percentage of correct identifications across all categories.

The results

The top performing plant identification technology in the 2025 evaluation was PictureThis, with 76% of the suggested identifications being correct (Figure 11). If the partial ratings (11%) are added to the correct ratings, we see that the app was helpful 88% of the time, averaged across all plant categories. Following this lead was ChatGPT (54% accuracy, 68% if you add the partially correct answers). The remaining two apps and AI platforms fell between 28-48% in accuracy in providing correct answers in the following order from most to least accurate: Google Lens, Gemini, CoPilot and Seek. The addition of the partial ratings increased that range to 46-63%.

Horizontal stacked bar chart titled “Overall accuracy %” comparing six apps: PictureThis, Google Lens, Seek, ChatGPT, Gemini, and Copilot. Each bar is divided into green (Correct), gray (Partial), and blue (Incorrect). PictureThis has the highest correct rate (around 75–80%) with small partial and incorrect portions. Google Lens shows moderate correct (~45%) with a large incorrect (~35–40%). Seek has lower correct (~30%) and the highest incorrect (~50%+). ChatGPT (~55% correct), Gemini (~40%), and Copilot (~30%) show mixed performance with substantial incorrect portions, especially for Copilot. — Figure 11. Overall performance of plant identification apps evaluated in fall 2025; smartphone apps on top and AI platforms on the bottom, both in order of decreasing accuracy from top down.

When focusing on seedling, vegetative and flowering grass identification, PictureThis got the correct answers 36%, 57% and 50% of the time, respectively. When it comes to grass identification, it is unclear whether partially correct answers are actually helpful. All other apps and platforms were correct 6-40% of the time. It is not surprising that these technologies struggled with grass identification, as it usually involves examining several small features (e.g., ligule, auricles, whether the leaf unfolds or unrolls as it emerges, and vegetative structures such as underground rhizomes) that cannot be captured in a single photograph. If identifying a grass is of particular importance, it would be more reliable to send multiple photos to a professional. Examples of the types of detail needed in these photos can be found in the article, “Grass photography tips for ID: Help us help you!” Even with the best photos, a physical sample may be required.

In 2025, there was an insufficient number and variety of photos taken of Amaranthus seedlings. Based on my experience with using it for Amaranthus plants, it is unreliable, even when plants are fully flowering, a much easier stage for identification. Distinguishing Amaranthus species is of particular importance for herbicide resistance management, so consulting other resources, like “Keys to distinguishing Palmer amaranth from other species” by Christy Sprague, PhD, remains important. Looking back at the 2023 data, Amaranth seedlings were correctly identified 46% of the time (Figure 12). Evergreen tree and shrub identification was another weak category for PictureThis in 2025.

Horizontal stacked bar chart titled “PictureThis evaluation by category – 2025.” Categories include ornamental flowering, broadleaf weeds (flowering and vegetative), grassy weeds (flowering and vegetative), seedlings (grass, amaranth, winter annual), and trees/shrubs (deciduous and evergreen). Each bar shows green (Correct), gray (Partial), and blue (Incorrect). Most categories have high correct rates (often 70–95%), especially ornamental flowering, broadleaf weeds, and winter annual seedlings. Lower performance appears in grassy weeds and grass seedlings, where incorrect portions are larger. Evergreen tree/shrub also shows a noticeable incorrect segment compared to other categories. — Figure 12. Performance of PictureThis in the outlined categories. Note, seedling amaranth data is displayed from 2023 due to low sample size, all other data is from 2025.

Nineteen of the groups participating in 2025 provided some additional feedback, with 17 saying PictureThis was the tech option they liked the best. They commented that it was their favorite primarily because of its accuracy, but a few also mentioned its ease of use and the ability to try it out for free. One group mentioned they liked all the additional information provided with the identification. PictureThis offers a free version of the app (tested here) and a paid version. The app is quite persistent in advertising the paid version, which some students and others I have interacted with find off-putting. Those who wish to trial the app for free need to pay attention and exit screens as needed. Since including the student’s data in 2021, we have seen increases in overall PictureThis accuracy of around 11%; about a 14% increase in vegetative grasses; and about a 15% increase in winter annual seedlings.

Including AI in 2025 presented some interesting instructional challenges. These platforms can read file names and any signage in the photo, so through trial and error, we had to remove these clues. Our intent was to only test their ability to use the plant’s image for identification.

In 2024, Visual Lookup, the plant-identifying application integrated into iOS cameras (e.g., iPhones and iPads), was included in the test. The students did not find this competitive with other options that year (e.g., PictureThis and Google Lens), with 36% correct answers across categories.

It’s always good to double check

When using various technologies to identify plants, it is always advised to check the identification with a reputable source (e.g., government or university-affiliated sites). Searching by the scientific name (Latin genus and species, e.g., Amaranthus retroflexus) will yield the most accurate results as common names can differ by region, environment, etc. (e.g., redroot pigweed vs. rough amaranth).

The USDA PLANTS Database is a useful tool for verifying the known distribution of an identified plant and confirming whether it is documented to occur in your area. Likewise, Michigan Flora is a good place to look, showing county level distributions based on specimens submitted to herbaria since the 1800s. Numerous herbaria have begun to digitize their collections into searchable databases with images, such as the Consortium of Midwest Herbaria and Integrated Digitized Biocollections. To confirm the identification of an ornamental plant, a botanical garden resource such as the Missouri Botanical Garden’s Plant Finder may be more appropriate (note the Missouri garden is in USDA Hardiness Zone 7a, Michigan ranges from 4a to 6b).

All apps tested since 2018: FlowerChecker, Garden Answers, Google Lens, ID Weeds, iNaturalist, LeafSnap, PictureThis, Plantifier, PlantIn, PlantNet, PlantSnap, PlantStory, Seek by iNaturalist, Turf Doctor, Visual Lookup, Weed ID and Xarvio. AI platforms tested since 2025: ChatGPT, Gemini and CoPilot. Note that all were not tested each year.

Thank you to Angie Tenney for reviewing this article, as well as reviews in previous years by Matthew Chansler and Alan Prather.

This article was published by Michigan State University Extension. For more information, visit https://extension.msu.edu. To have a digest of information delivered straight to your email inbox, visit https://extension.msu.edu/newsletters. To contact an expert in your area, visit https://extension.msu.edu/experts, or call 888-MSUE4MI (888-678-3464).