Which tool to choose
Three different jobs, three different tools.
- Identifying an unknown plant: a dedicated identification app, like PictureThis. In tests it recognizes many more plants than general-purpose AIs, which often get this wrong. Confirm anyway by looking at the flower and leaf, don't stop at the first name.
- Understanding why a plant is struggling (yellowing leaves, spots, pests, soil): Gemini from your smartphone. It reads the photo of the sick plant and diagnoses with good reliability, and it's strong on advice tied to your climate zone.
- Care plan and products (which fertilizer, how much, how often to water): ChatGPT, which gives concrete brands and quantities, or Claude, good at giving you tidy schedules without overpromising from a small balcony.
Buttons and photo-upload features often move around. If you can't find the command to attach the image, write the request in words in the text box: the model understands all the same.
How to do it
A struggling plant almost always has two or three possible causes. The job is to narrow them down, not to guess. From browser or app, the path is the same.
- Take a focused photo, in natural light, showing the sick part and the whole plant.
- Upload the photo and describe the real conditions. The working syntax:
This is my basil plant, I keep it on the south-facing balcony. The leaves are yellowing from the bottom and some are mushy. I live in Turin, I water it once a day in the evening. Tell me the 2 most likely causes and, for each, what I should do this week and how I'll know if I got it right.
- Apply one change at a time and wait a few days: if you change water, light and fertilizer all together, you won't know what worked.
- For treatments, ask for the exact doses and a natural alternative; on chemical products, the label remains the final authority.
A concrete example
Luca had a basil plant on the balcony that was yellowing and losing leaves. He photographed the plant and uploaded it to Gemini, writing that he watered it every evening. Gemini pointed to two likely causes: too much water with stagnation in the saucer, or too much direct sun at midday. It told him how to tell them apart: soil that's always soaked means the first one. That was it. He switched to ChatGPT for a watering schedule reduced to every other day and for the right amount. In ten days the new leaves were healthy.
When it does NOT work (and how to fix it)
If it gets the plant's name wrong
General-purpose AIs guess the species correctly less often than they seem sure. For identification use a dedicated app and confirm with a second close-up photo of the flower or leaf. Don't make care decisions based on a name you're not sure of.
If the advice is generic and not suited to your home
Without context the AI gives the textbook care, which isn't the same in Palermo and in Bolzano. Fix: always state your city, exposure (north, south), whether the plant is indoors or out, and in a pot or in the ground. The advice calibrates to your case right away.
If it recommends a product you don't know how to dose
A fungicide with no dose is useless or harmful. Ask for the exact quantity per liter of water and a gentler alternative (soft soap, neem oil) before moving to chemistry. And on plant-protection products, read the label: that's where the binding doses and precautions are.
A tip from someone who really uses it
Keep a small photo diary: one photo of the plant per week, always from the same spot. By re-uploading them, the AI compares the progression and understands whether the care is working, which a single photo doesn't allow. And every time it gives you a diagnosis, also ask how to verify it with your own eyes: a cause you can't check is just a hypothesis.
Frequently asked questions
Can it be done with the free versions?
Yes. Uploading a photo of a plant, receiving a diagnosis and a care plan falls within what the free versions handle. With many photos analyzed in a row a daily limit may kick in, but for a few houseplants you won't hit it.
Does it replace the nursery owner or the agronomist?
For everyday doubts it's a quick help. For a plant you care about that's getting worse despite care, or for a serious vegetable garden, the opinion of a nursery owner who touches the plant is worth more than any photo diagnosis.
Can I trust it if it tells me a plant or a berry is edible?
No, and here the mistake costs dearly. On identifying wild plants, berries and especially mushrooms, AIs get it wrong often enough to make them dangerous for this purpose: an edible species and a poisonous one can look alike in a photo. For anything that ends up in your mouth, the only source is a flesh-and-blood expert. The AI helps you grow basil, not decide whether a mushroom is safe to eat.