Evidence

How accurate are plant identification apps? What the studies found

Three peer-reviewed tests, their samples and their numbers as published, what pushes results up or down, and a short routine for checking any app's answer.

A potted variegated spider plant with a dangling plantlet beside a potted dragon tree with narrow, red-edged leaves on a bare trunk. — Illustration (AI-generated, for orientation only)

In short

Plant identification apps are accurate often, but far from always: in peer-reviewed tests the best apps named the species correctly on the first answer for 57% of British flora samples (2020) and 88% of flower photos of Irish herbaceous plants (2023), while one app tested on toxic plants was right to species 64% of the time. Flowers, sharp photos and common species score best.

Results vary by study, app and plant group: the same app can score very differently on another sample.
Genus is much easier than species: in a 2022 test of toxic-plant images, one app got the genus right 82% of the time and the species 64%.
Flowers beat leaves in the 2023 Irish study (55% vs 43% across six apps); a 2020 British study found no significant difference by plant part.
No study we found tested Plant Identifier, so this page gives no accuracy figure for it.

Plant identification apps get common plants right most of the time and rare ones, cultivars and poor photos wrong much more often. That is the short version of what peer-reviewed tests have found. Below are three of those studies with their samples and results as published, then what moves results up or down, and a routine for checking any answer. No study we found tested Plant Identifier, the app this site belongs to, so this page gives no accuracy figure for it.

What the published studies found

Each study used its own photos, its own plants and its own scoring, so their numbers can't be lined up into one ranking. Read each row on its own.

Study	Sample	Apps	Main result (first suggestion, correct species)
Jones, AoB PLANTS, 2020	38 images, 24 species of the British flora, 5 runs per app	9 apps	Plant.id 57%, Flora Incognita 46%, Google Lens 45%, PlantNet 39%, Seek 35%, PlantSnap 24%; Bing, Candide and iPlant 1–3%
Campbell, Peacock and Bacon, PLOS ONE, 2023	38 herbaceous species, 280 flower and 280 leaf photos, Ireland	6 apps	Flowers / leaves: PlantNet 88.21% / 80.36%, LeafSnap 83.57% / 77.14%, Seek 52.14% / 22.86%
Mahonski, Furlano and Chiang, J Med Toxicol, 2022	369 images of toxic plants (237 species)	PictureThis, free version	Species 64%, genus 82%; repeated photos of the same species consistent 71% of the time

Jones (2020). Hamlyn Jones tested nine apps on 38 images of plants from the British flora, each identified five times per app. Plant.id named the species correctly on the first answer 57% of the time, Flora Incognita 46%, Google Lens 45%, PlantNet 39% and Seek 35%. At genus level or better the top apps did better: Plant.id 70%, Flora Incognita and Google Lens 56%. Averaged across all nine apps, first answers were right to species 27.7% of the time and to genus or better 40.8%. The three weakest apps were nearly useless at species level.

Campbell, Peacock and Bacon (2023). Researchers in Ireland photographed 38 herbaceous species from 15 families at four Irish sites and scored six apps. Counting only a correct species on the first suggestion, PlantNet scored 88.21% on flowers and 80.36% on leaves, LeafSnap 83.57% and 77.14%, Seek 52.14% and 22.86%. iNaturalist scored 3.57% and 6.79%, which the authors explain: it often gives a genus rather than a species, and this scoring gave no point for a genus. The authors concluded that apps "should be considered an aid to plant identification and not assumed to be correct".

Mahonski, Furlano and Chiang (2022). Medical toxicologists ran 369 images of toxic plants from a toxicology handbook through the free version of PictureThis on an iPhone 11 Pro. It got genus and species right 64% of the time and the genus right 82% of the time. When the same species was photographed more than once, the answers were consistent 71% of the time. The authors wrote that the app should not guide decisions in patient care.

Two patterns hold across all three: the right genus is far easier than the right species, and no app was right every time.

What makes a result better or worse

The part you photograph. In the Irish study, flowers were identified to species 55.18% of the time across six apps and leaves 43.21%, a significant difference. The British study found flowers, fruit and whole plants ahead of leaves too, but with uneven samples and no significant difference, and its author called the comparison only indicative. Pl@ntNet's own identification guide asks for several organs (leaf, flower, fruit) to get a better answer. When a flower is there, photograph it.

Image quality. Pl@ntNet lists blurry, badly framed photos missing key organs among its weaknesses. The New York Botanical Garden's guide to photographing plants for identification asks for sharp focus on the target, close-ups of both leaf surfaces and their arrangement on the stem, and soft light: bright sun washes details out, and zooming in on a distant shot turns them into pixels.

Rare species and coverage. An app can only name what it has learned. Pl@ntNet says it covers about 70,000 species, a small fraction of the roughly 350,000 known plant species. A rare plant, or one from a flora the app has few photos of, is more likely to come back as its commonest look-alike.

Cultivars and hybrids. Garden plants are often bred forms with unusual colors, doubled flowers or variegated leaves. A photo may return the parent species, a close relative, or the wrong plant. None of the three studies tested garden cultivars directly, so treat a cultivar name from any app as a guess.

Region. Each study tested one flora: British, Irish, or a toxicology handbook's mix. An app that scores well on European wildflowers may score differently on Californian natives or tropical houseplants.

How to check a result

A plant name is a hypothesis. Five steps turn it into something you can rely on:

Photograph a second part. If you started with a leaf, add the flower, fruit or whole plant. Two organs that agree beat one.
Ask a second tool. Run the same photo through Visual Look Up or Google Lens (see how to identify plants on iPhone). If two independent tools agree, you're on firmer ground; if they don't, keep checking.
Look at the alternatives, where an app shows them. Pl@ntNet shows "a list of possible identifications, ranked by order of certainty". A close second suggestion is a warning sign.
Compare the features yourself. Check leaf arrangement (alternate, opposite, whorled), leaf edges, flower parts and stems against the description and photos for the suggested species. The photo checklist lists the shots that make this possible.
Ask a person when it matters. A local extension office, botanical garden or naturalist group can confirm what no app can.

Plant Identifier shows one name per photo, with no confidence score and no list of alternative species. That keeps the result simple, but it means steps 1, 2 and 4 are on you for any plant that matters. Its follow-up chat can help with step 4: ask how the suggested species differs from a look-alike, then check that difference on the plant.

Toxic plants, children and pets

The toxic-plant study is the clearest warning in this literature. Its authors tested an app on toxic plants and concluded it should not guide patient care. The Irish study's authors called apps an aid, especially where toxic plants are involved. An app name is never a reason to decide that a plant is harmless.

If a person may have eaten part of a plant, call Poison Control at 1-800-222-1222. If a pet may have, call your vet or the ASPCA Animal Poison Control Center at (888) 426-4435. Bring the plant or a piece of it if you can. For houseplants and cats, see plants toxic to cats, which names its source for every listing.

What these numbers can and can't tell you

These studies show the range: from almost no first answers right to species for the weakest apps to just under nine in ten for the strongest, depending on app, plants and photos. They can't tell you how today's version of any app performs, because models change after a paper is published. They can't rank apps that weren't tested, including Plant Identifier. And they can't tell you how an app will do on your plant. For that, the checking routine above is the most reliable method you have.

What a photo can't settle

Studies test a fixed set of photos at one moment; apps update their models, so a 2020 or 2023 score may not describe today's version.
No photo app can settle whether a plant is harmful to touch, eat or keep near a pet; the authors of the toxic-plant study say an app should not guide patient care.
Cultivars, hybrids and rare species are where every app is weakest, and a photo of a leaf alone may never separate some look-alikes.

Questions people ask

How accurate is PlantNet?

It depends on the sample. In a 2023 study of 38 Irish herbaceous species, PlantNet's first suggestion was right to species for 88.21% of flower photos and 80.36% of leaf photos. In a 2020 test on plants from the British flora, it was right to species 39% of the time.

How accurate is Google Lens for plants?

In Jones's 2020 test of nine apps on 38 images from the British flora, Google Lens named the species correctly on its first answer 45% of the time and reached at least the right genus 56% of the time. That placed it third, behind Plant.id and Flora Incognita.

Is PictureThis accurate?

In a 2022 test on 369 images of toxic plants, the free version of PictureThis got genus and species right 64% of the time and the genus right 82% of the time. The authors said it should not guide decisions in patient care.

Which plant identification app is the best?

No study crowns one app for every plant. The leaders changed between samples: Plant.id led a British test in 2020, PlantNet led an Irish test in 2023. Pick an app that fits your plants, and check any answer that matters.

How accurate is Plant Identifier?

No published study we found has tested it, so we give no figure. The app shows one name with no confidence score and no list of alternatives, which means a doubtful result needs your own check against the plant.

Sources

What plant is that? Tests of automated image recognition apps for plant identification on plants from the British flora — AoB PLANTS 12(6), plaa052 (Jones, 2020)
A repeatable scoring system for assessing Smartphone applications ability to identify herbaceous plants — PLOS ONE (Campbell, Peacock and Bacon, 2023)
Validation of a Plant Identification Application Using Digital Images of Toxic Plants — Journal of Medical Toxicology 18(2):159-162 (Mahonski, Furlano and Chiang, 2022), via PubMed Central
Identify a plant — Pl@ntNet documentation
Pl@ntNet's strengths and weaknesses — Pl@ntNet documentation
Photographing Plants for Identification — New York Botanical Garden, LuEsther T. Mertz Library
Poison Control (US) — National Capital Poison Center
Animal Poison Control — ASPCA