How Wrong Are the Identification Apps?
By IdentifyThese Editorial · Reviewed · 3 min read
The apps are useful, and the specific thing they are least good at is the thing people most want them for.
What the testing has found
Independent evaluations have submitted photographs of known species to popular identification apps and compared the output.
The consistent finding is that top-result accuracy falls well short of what a safety decision requires, and that deadly species are sometimes returned as edible ones. Performance is best on distinctive, well-photographed, commonly recorded species and worst on exactly the ambiguous groups where help is needed.
Accuracy is also uneven by region, because the training data reflects where photographs were taken rather than where the user is standing.
Why photographs are structurally insufficient
Several features that decide an identification are not in a photograph.
The stem base, which is usually underground and holds the volva that identifies a deadly Amanita — the whole point of the Amanita check.
The spore print colour, which takes hours and is frequently decisive, as taking a spore print describes.
The smell, which separates several similar species.
Bruising reactions, which develop over minutes.
The substrate and the surrounding trees.
Microscopic features, which are required outright for many groups.
| What an app sees | What identification needs |
|---|---|
| Cap surface and colour | Plus the stem base |
| Overall shape | Plus spore print |
| Gills, if photographed | Plus smell and staining |
| One specimen | Plus several ages |
| No location context | Plus habitat and host trees |
Where they are genuinely useful
Narrowing a search. An app that suggests three genera gives a starting point for a field guide, which is faster than working through keys blind.
Learning. Photographing and checking builds familiarity over a season, provided the answers are treated as hypotheses.
Recording. Some platforms feed observations into research datasets with human review, which is a real contribution.
The community platforms
Observation platforms with human reviewers are a different proposition from an automated identifier, and better.
They still operate from photographs, they still cannot see a spore print, and their contributors are explicit that an online identification is not a clearance to eat anything. The limitations of any photograph-based route are covered in identifying a mushroom from a photo.
The asymmetry that matters
A false negative wastes a mushroom. A false positive on an amatoxin-containing species causes liver failure, and the symptoms begin six to twelve hours later, by which point substantial damage has occurred — the pattern in the mushroom poisoning timeline.
There is no tool at any price that makes that risk acceptable on the basis of a photograph.
The route that works
A local mycological society, a specimen in hand, and someone who has handled the species in that region for years.
That is slower than an app and it is the only method that produces a defensible answer, and it is free in most places.
Improving the odds
If an app is being used at all, the way it is used changes the result substantially.
Photograph several specimens of different ages. Include the underside and the base with the soil brushed off. Include the substrate and the surrounding trees in a wider frame. Add the location and date.
That converts a single ambiguous image into something a human reviewer could work with, which is what the better platforms actually need. It does not convert an app's answer into a decision about eating.
Why confident output is the problem
An app returns a name with a percentage beside it, and the format reads as a measurement when it is a similarity score.
That presentation is what converts a hedge into a verdict in the user's mind, and it is the reason these tools are more dangerous than a field guide that simply says the group is difficult.
Species referenced
Read this before anything else
- Never eat a wild mushroom identified by a website or an app.That includes this one. These tools narrow possibilities; they cannot see your mushroom, its base, its smell or its habitat, and the difference between a meal and a liver transplant is sometimes a feature buried in the soil.
- Get it confirmed in person by an experienced local forager or a mycological society. Most regions have one, most run free identification sessions, and they will look at the actual specimen.
- Dig, do not pull. The cup at the base of a death cap is often underground. Pulling the mushroom leaves the most important feature in the soil.
- If someone has eaten a wild mushroom and feels unwell, call poison control now — in the US, 1-800-222-1222. Do not wait for symptoms to worsen. Amatoxin poisoning has a false recovery phase, and delayed treatment is what makes it fatal. Save any uneaten pieces for identification.
Keep reading
- Can You Identify a Mushroom From a Photograph? Not ReliablyThe features that separate an edible mushroom from a deadly one — spore colour, the buried stem base, the cut flesh — are frequently not in the photograph at all.
- The Amanita Check: Dig It Up, Every TimeMost fatal mushroom poisonings are one genus. Two features identify it, one is usually underground, and pulling the mushroom rather than digging destroys the evidence.
- Mushrooms in Bark Mulch and Garden BedsMulch is wood, and wood-decaying fungi eat wood. Most are harmless, one is not, and two of them are strange enough to be worth recognising.