How to check an AI rock identification before you trust it

Start with the evidence, not the name

Copy the tentative match into a note, but put the name at the bottom of your checklist. Above it, record what you can actually see: grain size, whether grains lock together or sit in layers, surface shine, crystal faces, holes, banding, and any contrast between a weathered outside and a fresh-looking surface. A name is a hypothesis; these observations are the evidence you can revisit.

Keep the specimen dry and photograph it again in even daylight if glare or wetness hides the texture. Add a scale object only when it does not cover a diagnostic feature. USGS field guidance similarly treats a written description and a scale-bearing photograph as early parts of documenting a specimen.

Use this evidence order

1. Rock fabric first. Ask whether the specimen is granular, layered, glassy, vesicular, foliated, or made of visible crystals. This can eliminate broad rock classes before you compare colours.
2. Luster and transparency next. Describe what the surface does in light: metallic, glassy, waxy, pearly, dull, or earthy. Do not turn one shiny spot into a whole-specimen property.
3. Break pattern after that. Look for repeated flat planes versus irregular, curved, jagged, or crumbly breaks. Do not create a fresh break just for an app result; examine an existing edge instead.
4. Only then consider simple physical checks. A streak plate, magnet, or careful scratch comparison can be useful when the specimen and location make the test appropriate. A photo cannot measure hardness, density, chemistry, or crystal structure.

Use a three-candidate rule

If the tool gives a confident-looking name but the visible features conflict, do not rename the evidence to suit it. Keep up to three candidates and write one observation that would separate each pair. For example, write “look for visible grains” or “check whether the break follows repeated planes,” rather than writing a conclusion you cannot support.

When every candidate relies on a property you cannot observe safely, the useful result is undetermined from this photo. That is a better field note than a false positive.

When to stop and ask a person

Do not use a photo match to confirm a meteorite, appraise a gem, decide whether a powder or fibrous material is safe, or make a collecting decision where access or regulations matter. The U.S. Geological Survey notes that rocks and minerals are difficult to identify accurately from photographs and are best examined in person from multiple perspectives. A local geology department, museum, geological survey, rock shop, or gem and mineral club can provide a closer look.

Record the context that a photo loses

Before putting the specimen away, record where it was found, whether it was loose or attached to an outcrop, and what was nearby. A river pebble, a beach stone, a driveway aggregate, and a specimen from a bedrock layer can look alike in isolation but belong to very different stories. Context will not prove an identification, but it makes the next comparison more useful.

Sources for the method

• U.S. Geological Survey: Can you identify my rock or mineral?
• U.S. Geological Survey: documenting rock observations from macro to micro
• U.S. Geological Survey: Collecting Rocks