PictureThis vs Plantum AI Plant Identifier Accuracy: 234 Photos Tested

Why I Ran This Test

Before I built AI Plant Hub's identifier, I wanted to know how good the existing free apps really were. Every app page claims "over 95% accuracy." That number is marketing. The real question is: given a photo taken by an average person with an average phone, what fraction of identifications are correct?

So I built a test set and ran the same 234 photos through four apps in March 2026. Writing up the results partly for my own benefit and partly because I could not find a single honest accuracy comparison online. Every "comparison" article turned out to be an affiliate piece ranking PictureThis first.

Methodology

The test set:

• 234 photos total, all taken in natural daylight with an iPhone 13.
• 96 houseplants with labels verified from the nursery tag or by a botanist friend. Monstera, Pothos, Calathea, Peperomia, Ficus species, Alocasia, Philodendron, Dracaena, Snake plants, Spider plants, and so on.
• 54 succulents and cacti, species verified from specialist nursery tags. Echeveria, Haworthia, Crassula, various Sempervivums, a few rarer Lithops and Conophytum.
• 42 outdoor garden plants. Roses (6 cultivars), lavenders (3), salvias, camellias, grevilleas.
• 42 weeds, wildflowers, and grasses identified by a local bushcare volunteer in Sydney.

For each plant I took one clear photo of the whole plant, and for ambiguous cases an additional leaf close-up. All photos were uploaded to each of the four apps the same week, in random order.

Scoring: top-1 correct (the first suggestion matches the known species), top-3 correct (the correct species appears in the first three suggestions). I also recorded the app's reported confidence for the top suggestion.

Four apps tested: PictureThis (7-day trial, no login required), Plantum (free with limits), Plant.net (free, academic backing from the French National Research Institute), and Google Lens (free, general-purpose).

Headline Results

Two things jump out. First, PictureThis's reported confidence averages 94% while it's right 78% of the time. That's not a disaster but it is overconfident calibration. Plant.net's 67% average confidence with 68% actual accuracy is almost perfectly calibrated, which is a point in its favour even though the overall accuracy is lower.

Second, the difference between PictureThis and Google Lens is 20 percentage points. That's a big gap. It's worth paying for PictureThis Premium if you use plant ID daily. It's not worth it if you use it twice a year.

For reference, a 2023 paper in AoB Plants (the Annals of Botany open-access journal) discussed accuracy of citizen-science plant ID platforms and noted that Pl@ntNet (Plant.net) reached high accuracy on well-photographed specimens with multiple organ views, but degraded sharply with single poor-quality photos. My single-photo test here is closer to the worst case for Plant.net and the advertised case for PictureThis, which partly explains the gap.

Accuracy by Plant Category

The succulent numbers are the real story. Every app performs worst on succulents. Small morphological differences (Echeveria vs Graptoveria vs Graptopetalum) trip them all up. Variegated cultivars (Echeveria 'Lola' vs 'Lola Variegata') are nearly impossible to distinguish from a single photo.

Plant.net edges out PictureThis on weeds and wildflowers, which fits what the AoB Plants paper suggested: academic platforms built for field botanists have richer reference data for wild species than consumer apps optimised for houseplants and garden centre stock.

Where Each App Fails

PictureThis nails common houseplants and fails systematically on:

• Any plant not in the US or European mass-market retail supply chain.
• Australian natives (grevilleas and banksias were often identified as "Protea species").
• Succulent cultivars with minor variegation differences.

Plant.net fails on:

• Tropical houseplants in a domestic setting (it prefers field photos).
• Very young or stressed plants without flowers or fruit.
• Anything where no close-up of the flower or leaf venation is available.

Plantum fails on:

• Any slightly unusual cultivar (it defaults to the most common species name regardless of variegation).
• Succulents, harder than PictureThis across the board.
• Weeds, where it's noticeably worse than either competitor.

Google Lens fails on:

• Giving you specific species at all. It often returns "succulent" or "house plant" as the top "identification," which is not an identification.
• It's better at showing you similar images than at naming the plant, which is a different use case.

The Confidence Score Problem

All four apps have a confidence display problem. If the number isn't calibrated, it's not useful. Here's what I saw:

• PictureThis's 95%+ confidence results were right 81% of the time.
• PictureThis's 99% confidence results were right 84% of the time. So a 4-point jump in reported confidence bought only 3 points of real accuracy.
• Plantum's 95%+ confidence results were right 68% of the time.
• Plant.net's top-probability score correlates more linearly with correctness, but it's also a much lower average number.

The practical rule: use the apps' confidence as ordering information (if the app is 97% sure, it's more likely right than when it's 82% sure) but not as probability (97% sure does not mean 97% correct). And always check the top-2 or top-3 suggestions, not just the top-1.

Which App to Use When

1. Identifying a new houseplant purchase: PictureThis. Highest accuracy on mass-market houseplants, worth the trial.
2. Identifying a wild plant on a bushwalk: Plant.net. Academic backing, better calibration, free, and better on wild species.
3. Sorting through a succulent collection: None of them alone is reliable. Use PictureThis for the guess, cross-check with a succulent specialist group on Reddit or Facebook for cultivars.
4. Quick "what is this" curiosity: Google Lens. Fast, free, already on your phone. Accept that it will often say "house plant" and move on.
5. Identifying what's taking over your lawn: Plant.net first, PictureThis second. Plantum is not ready for weeds.

If you're someone who buys rare plants or specific succulent cultivars, no AI app is enough. You need the specialist nursery's label, a community group, or a botanist. AI gets you 60-80% of the way there on a single photo, which is useful, but it is not the finish line.

FAQ

Which plant identification app is most accurate in 2026?

Based on 234-photo testing in March 2026: PictureThis leads overall at 78% top-1 accuracy on common plants, but Plant.net is better on wild species and weeds (81% on that subset vs 76%). Neither is reliable on succulent cultivars; expect about 50-60% accuracy there regardless of app.

Is PictureThis worth paying for?

If you identify plants more than once a week, yes. The free trial is enough to try it. If you only use it occasionally, Plant.net is free and accurate enough for wild plants, and Google Lens is fine for "is this a fern or a philodendron" level questions.

Why are AI plant identifiers so bad at succulents?

Succulent taxonomy is a nightmare even for specialists. Echeveria, Graptopetalum, Graptoveria, and Pachyphytum hybrids overlap in appearance. Many named cultivars differ from each other only by minor variegation or leaf shape. Single-photo AI doesn't have enough signal to distinguish them. Expect 50-60% accuracy at best.

Can I use AI plant identifiers for toxic plant identification?

Be very careful. At 60-80% accuracy, one-in-five identifications is wrong. For a pet that may have eaten a suspicious plant, call your vet or the ASPCA poison hotline rather than trusting an app. AI gives you a starting hypothesis, not a diagnosis.

How do I take a better photo for plant identification?

Natural daylight, plain background where possible, single plant in frame (not a whole garden bed), include at least one clear leaf and one flower if there is one, shoot from about 30-50cm away with the phone focused on the most distinctive feature. Multiple photos of the same plant (whole + leaf + flower) significantly improve Plant.net's accuracy specifically.

About the Author

Jim Liu is a Sydney-based developer who runs AI Plant Hub. He spent two weeks testing AI plant identifier apps on a 234-photo test set before starting to build his own. He is not a botanist. For specialist identification of rare species, consult a qualified botanist or your local herbarium.

Last updated: 2026-04-18.