I spent a Saturday afternoon testing AI plant identification apps on my balcony. The trigger was a reader email asking which app she should pay for. Honest answer: it depends on what you’re growing, but a few patterns emerged that surprised me.

My balcony has 23 plants, give or take. Mix of indoor tropicals on a covered shelf, herbs on the railing, a few natives in larger pots. Not a botanical garden, but enough variety to put these tools through a proper test.

What I tested

I ran the same plants through four apps over an afternoon: PictureThis, PlantNet, the new Google Lens plant mode, and Seek by iNaturalist. All current versions as of late April 2026. I photographed each plant from the same three angles - whole plant, leaf detail, and any flowers or distinctive features.

I scored each correct identification, partial credit for genus-only matches, and noted any confidently wrong answers (which are arguably the most dangerous).

The results

PlantNet got 19 out of 23 correct at species level. It struggled with two of my variegated cultivars (it ID’d the parent species, fair enough) and confidently misidentified my Pilea peperomioides as a Peperomia obtusifolia. Free, no ads. Backed by a proper research consortium at French institutions.

Google Lens hit 17. Faster than the others, but more prone to confident wrong answers. It told me my Calathea orbifolia was a Banana plant. It wasn’t. Its strength is integration - take a photo in your camera roll, get an answer in two seconds.

PictureThis scored 20 correct. The slickest interface, the best care advice, and the most aggressive paywall. After three free IDs you’re hit with subscription prompts that are honestly insulting in their persistence. The care advice itself is generic and sometimes wrong for Australian conditions.

Seek (iNaturalist) got 14. Lower hit rate but zero confident wrong answers - it just declines to guess if it’s not sure. For a science-leaning user this is the right behaviour. It also pushes you to upload your finds to iNaturalist, which is a genuine contribution to citizen science data.

What I’d actually use

For everyday “what’s this plant” questions: PlantNet. It’s free, it’s accurate, and it doesn’t pretend to know things it doesn’t.

For when accuracy really matters - say, identifying something before pruning or before letting a pet near it - I’d use two apps and only trust the answer if both agree. PlantNet plus Seek is my combo.

I’d skip PictureThis unless you specifically value the care reminders feature. The ID accuracy isn’t worth the subscription pressure.

Where AI plant ID falls down

A few honest limitations worth knowing.

First, AI is bad at sick plants. Show it a leaf with spots, mottling or unusual colour and you’ll often get a wrong species ID because the visual cues it’s trained on are absent. If you’re trying to identify a problem, you need a plant ID first from a healthy specimen, then look up the disease separately.

Second, cultivars confuse everything. There are dozens of Monstera deliciosa cultivars, hundreds of philodendron hybrids, and the apps mostly can’t tell them apart. They’ll give you the species and shrug at the variety.

Third, native Australian plants are underrepresented in training data. My native finger lime got identified as a citrus (kind of true), my correa as a cordyline (not even close), and my hardenbergia as “Wisteria sp.” All four apps had the same blind spots here. If you’re working with Australian natives, PlantNet has a regional dataset for Oceania that helps, but the gap is real.

The bigger picture on AI tools for plant care

I’m cautious about how much we offload to these tools. Knowing your plants by name is half of caring for them. When I started gardening I used apps constantly - now I use them maybe once a fortnight, mostly for plants I’ve just acquired or unfamiliar things in friends’ homes.

There’s also genuine craft in being able to tell a Pilea from a Peperomia by feel - the weight of the leaf, the way the stem holds. AI can’t teach you that. Time with the plants does.

If you’re using AI tools more broadly for garden planning, plant care reminders, or working out what’ll grow in your conditions, Team400 has been thinking about how AI gets used in everyday workflows like this. The takeaway from their work I keep coming back to: the tool is only useful if it makes you better at the underlying skill, not if it replaces it.

My recommendation

Download PlantNet. Use it for a month. See if it answers your questions. Most balcony gardeners won’t need anything else.

If you find yourself constantly stumped on cultivars or want care advice baked in, then trial PictureThis for a week before paying. Don’t subscribe in the heat of the moment when you’ve just bought a mystery plant from Bunnings.

The real skill isn’t picking the right app. It’s learning your plants well enough that you mostly don’t need one.