The records you browse on this portal are only as good as the identifications behind them. Here is how the Pl@ntNet community keeps them honest.

An AI that needs watching

Pl@ntNet’s identification model now covers around 85,000 species. Many of them look very much alike, and no model is infallible: to keep its performance over time it has to be continuously monitored, evaluated and corrected. The Pl@ntNet team is small — it cannot check every prediction or spot every drift on its own.

Enter the Watchers

Some time ago, a group of expert users spontaneously formed after noticing recurring identification errors. On their own initiative they started correcting them through the collaborative revision system that is open to every user. Today this group works hand in hand with the Pl@ntNet team: they monitor identification quality, raise the alarm when a drift appears, contribute corrections and help set up safeguards that improve the data over time.

In 2026 Pl@ntNet is creating a dedicated status for contributors with advanced botanical expertise, to structure this work, give them a direct line to the team and recognise the effort they put in behind the scenes.

Why it matters for GBIF users

Only observations whose name is considered valid — confirmed by the community, or predicted with a very high confidence score — are shared with GBIF. The Watchers’ corrections therefore flow directly into the datasets you can search and download here. If you spot something odd in a record on this portal, follow the link to the original observation on identify.plantnet.org and propose a correction: that is exactly how the system is designed to work.

Professional or amateur botanist with strong skills? Pl@ntNet would be glad to hear from you: plantnet.org/#contact. Read the original post on plantnet.org.