The short answer
Garden Sage can show a presentation-only result as a clearly labelled Prediction when the best-ranked species is useful as a lead but not supported for confirmation. That Prediction is not confirmable or saveable and does not unlock plant-specific care or a health diagnosis.
When the evidence supports a match, the app can keep it at the supported species, genus or broad plant-type level. When no useful prediction or result is available, it can ask for a closer view, request a plant name for a care question, or explain that the image or service cannot provide a result.
It does not turn an uncertain identity into species-specific care, and it is not a substitute for toxicity, medical, veterinary, emergency or other high-stakes professional advice.
The result should match the available evidence
| Evidence state | Garden Sage response | What the user should do |
|---|---|---|
| A best-ranked species is worth showing but not supported for confirmation | Label it as a Prediction and keep it out of saving, plant-specific care and health diagnosis | Treat it as a lead to verify, not a confirmed identity |
| Image and context support a plant match | Show the strongest supported identity, confidence or uncertainty, visible evidence and a practical next step | Verify important traits and keep watching the plant |
| Genus or broad plant type is stronger than species | Keep the identity at the supported level rather than inventing precision | Capture a separating trait or wait for flower, fruit or mature growth |
| No useful prediction or result is available and another view would help | Ask for a closer or more useful view | Retake one plant with whole habit or the requested detail |
| Care question is useful but identity is not settled | Limit guidance to visible, broadly safe checks or request the plant name | Check roots, conditions and symptoms without assuming species |
| Image does not contain a supported plant subject | Do not create a plant identity or saveable species result | Choose a clear plant image |
| Analysis provider is unavailable | Return an availability limit instead of fabricating an answer | Try later; do not treat the failure as a plant conclusion |
| Question has high safety stakes | Escalate the limit | Use a qualified local professional or emergency resource |
Confidence is shown as context, not a guarantee
A confidence value belongs to a particular result and system. It is not a universal probability that the name is correct. Garden Sage pairs the match with supporting observations and limitations so you can check whether the proposed identity fits the plant in front of you.
Garden Sage does not publish a universal accuracy claim. A defensible accuracy figure would require a fixed app and model version, a named representative evaluation set, documented regions and plant groups, separate genus/species scoring, and transparent handling of abstentions.
Uncertain identity narrows the care scope
Some observations are useful without a species: whether the root zone is dry or saturated, whether leaf damage is old or expanding, or whether insects are visible under a leaf. Species-specific watering schedules, toxicity conclusions and treatment choices require stronger identification and context.
When a working genus or plant type is the strongest supported level, the app should preserve that uncertainty. The user can then add better evidence rather than building care history on a false species.
A request for another photo is a result
“Take a closer photo” is not a hidden identification. Garden Sage uses that route when there is no useful prediction or supported result and another view could help; a low score alone does not force a retake. A useful request should tell you what to capture: one plant rather than a mixed bed, several leaves attached to a stem, the underside of damage, an attached flower, or the whole growth habit.
Service limits remain separate from plant conclusions
A provider timeout, rate limit or unavailable model says nothing about the plant. Garden Sage is designed to fail as a service state rather than fill the gap with generic care. Retrying later is appropriate; treating the error as a negative identification is not.
Safety boundaries
- Do not use Garden Sage to decide whether a plant or mushroom is edible.
- Do not rely on it for poisoning, allergy, medical or veterinary decisions.
- Do not choose a pesticide from a photo result alone.
- Do not use an uncertain identity for regulated invasive-species reporting or removal without local confirmation.
- Use local Extension, a plant clinic, certified arborist or other qualified professional when the consequence of error is high.
Photo processing and your controls
Plant analysis requires photo processing. Garden Sage’s privacy policy explains machine-learning and computer-vision processing, permission controls, account deletion and retention. The support page provides the current contact and account-help path. Those public contracts—not this editorial page—govern the service.