Precision vs Recall: Two Ways an AI Can Be 'Right'

An AI says it is '95% accurate'. Sounds great, right? But accurate at what? Two simple ideas, precision and recall, tell you what that number is hiding.

The fishing net analogy

Imagine you are fishing for tuna. You cast a net and pull it in. Inside are some tuna, and some other things you did not want: boots, seaweed, other fish. Behind you in the sea, a few tuna swam away.

What is precision?

Precision asks: of everything in my net, how much is actually tuna? If the net holds 10 items and 8 are tuna, precision is high. Precision is about avoiding false alarms.

What is recall?

Recall asks: of all the tuna in the sea, how many did I catch? If there were 20 tuna and you caught 8, recall is low. Recall is about not missing things.

Why you usually cannot max out both

A wide-mesh net grabs almost every tuna (high recall) but also brings in lots of junk (low precision). A very selective net keeps only perfect tuna (high precision) but lets many escape (low recall). Improving one often hurts the other, so you choose based on what matters.

Which one matters more?

It depends on the cost of a mistake. A spam filter should favour precision: wrongly binning an important email is worse than letting one spam through. A medical screening test should favour recall: missing a real illness is worse than a false alarm that gets double-checked.

The simple rule

Precision = how clean is what I caught? Recall = how much of what exists did I catch? Whenever someone quotes one AI accuracy number, ask which of the two they mean.

🎁 Use code FREE2026 at maritime.rangalabs.cloud for 1 year free access to MarineRef.

👉 Get one AI tip every day on WhatsApp, free: https://chat.whatsapp.com/DhaGgTuQ9GE67ykGVMgxXb