How these guides are built, and where they stop

By Karim Diabate

257 guides on this site are built from one shared knowledge base rather than written one at a time. That is worth being open about, because it decides what they are good for and what they cannot do. This page is the second half of that: not how clever the method is, but where it runs out.

Why the answer is a list and not a verdict

Most plant symptoms have four or five plausible causes that look identical in a photograph. A limp plant is either thirsty or drowning, and the two want opposite treatments. Yellow lower leaves are ordinary ageing on one plant and the first sign of root rot on the next. Anything that answers with a single confident cause is throwing away the part that actually helps you: what would be true as well, if it were that cause.

There is a published measurement of how much gets thrown away. A 2022 study in Arboriculture and Urban Forestry ran 440 photographs of 55 tree species through six identification apps. For leaf photos, iNaturalist named the correct species as its answer 69.55% of the time, but the correct species was somewhere in the suggestions it offered 95.91% of the time. Most of what the tool knew was in the part a single answer would have hidden.

What a guide is actually made of

Each guide is composed for one plant and one symptom from two things. The first is a record of what that plant is in the terms that change a diagnosis: how far it likes to dry out, what light it wants, what a healthy root looks like on it, and the handful of quirks that a generic guide would miss. That part is written by hand, per plant. The second is a library of causes, where each cause knows how to describe itself for that kind of plant, because overwatering on an aloe, on an orchid and on a fern do not look the same.

The ranking is per plant too. A jade plant page opens on overwatering and a spider plant page on water quality, because that is the order the causes deserve on those plants. Where two plants genuinely want the same care, their guides will read alike. We would rather repeat ourselves than invent a difference that is not there.

Where this stops being reliable

This is the part most pages like this leave out, so it is the longest section here.

• A guide cannot see your plant. It ranks causes by what is usually true of that kind of plant, not by anything about yours. The order is a starting point for looking, not a diagnosis.
• "Most common" is an ordering, not a measurement. We do not publish percentages next to causes, because we have not measured how often each one is right. A number would look like evidence and be a guess.
• A plant can have two things wrong at once, and a ranked list quietly assumes one. Spider mites arrive on a plant that is already stressed by dry air far more often than on a thriving one.
• The confirmation step is the load-bearing one. Every cause on every guide names a single check that settles it, usually something physical like sliding the plant out of its pot. If you skip it and act on the ranking, you are acting on a guess made by someone who has not seen the plant.
• Scope. These guides are about common indoor plants kept in ordinary rooms. Outdoor conditions, regional pests and diseases, edible crops and anything a plant clinic would want a sample for are outside what they cover.

The photographs are healthy plants, deliberately

Every guide carries a reference photograph of a healthy specimen of that species, credited to its photographer and licensed through Wikimedia Commons, and the caption on every page says it is a reference photo and not a picture of the problem.

The reason is simple. We do not have a library of symptom photography we can verify, and the two ways to fake one are both worse than having none: a photograph of the wrong plant sitting next to a diagnosis, or a photograph of a symptom we cannot confirm is the symptom named. A healthy plant you can compare yours against is honest and useful. A misleading picture on a diagnosis page is the thing a hurried reader remembers.

What the app does, and the one measurement behind it

The app reads a photograph and returns the same shape of answer: causes in order, with the visual evidence it used for each. In August 2026 we ran 28 openly licensed PlantVillage images through the real pipeline to choose between three models. The one that shipped named the right condition first on 71% of them, and had the right condition somewhere in its list on 82%.

What is never published here, whatever the evidence says

Treatment guidance stays at the level of the category. No product names, no mixing ratios, no application intervals. In the United States a pesticide label is federal law, and it overrides anything written on a website. Whatever the label on the product in your hand says is the instruction that counts.

We never tell you whether a plant is safe to eat or touch, and we never tell you that a plant is safe for a child or a pet. Those are questions about people and animals rather than about plants, and the right answer is to route them, not to answer them carefully. If a child or an animal may have eaten any part of a plant, contact Poison Control (US 1-800-222-1222) or a vet immediately.

Sources, dates and corrections

The general horticultural facts behind a guide come from published university cooperative extension services and the Royal Horticultural Society, and each guide links the ones it leans on at the foot of the page. The published and updated dates on a guide are the real ones: a date changes when the page changes, and never simply to look fresh.

Where a claim on this site cannot be traced to a source, it comes off the site. That has already happened once: a pair of accuracy figures used to sit in this project's own notes as the justification for the ranked list, and when we went looking for the study behind them we could not find one. They are not on this page, and the study cited above is what replaced them.

If a guide is wrong, tell us which one at support@yiriba.com. No page here is reviewed by a credentialed horticulturist and none claims to be, which is why the limits above are written out rather than summarised. More about who writes this is on the about page.