Can AI Help Identify Gemstones & Other Organic Materials In Jewellery?

Recently I have been submitting twenty or thirty items per day for AI identification, without giving any information on prior research, or knowledge about each item.

Some people claim AI is unreliable and inaccurate, which is true, if the photographs are taken with an iPhone on the fly, in poor lighting conditions. AI identification using poor photography and lighting, is less than 40% accurate.

However the accuracy goes way up, if you use a Nikon camera, a macro lens, and shoot raw, high resolution photos in good window light, without glare. It makes all the difference in the world. You also have to use a tripod. In such a case, you can aim for much higher accuracy.

I realize AI is not a substitute for hands on evaluation, and laboratory testing done by certified gemologists. But let’s face it, collectors of vintage and antique jewellery, as well as rock collectors who have large collections, cannot realistically get every stone appraised by a gemologist. The cost of doing so would be astronomical.

A high percentage of coloured gemstones are beautiful, with much to be learned about the unique properties, inclusions, and colours, all of which help to identify what they are. But the value does not warrant paying as much, or more than the item is worth, in order to have it appraised or certified.

So far, I have found the AI assessment on not only gemstones, but other organic materials, to be quite accurate. For instance, I submitted a Victorian era carved bracelet that I had already determined must be tortoise shell, and sure enough AI returned an in-depth description explaining why it was identified as tortoiseshell.

Another antique bracelet that I had studied at length, compared it to bone, and known samples of horn, in order to identify it as horn, AI confirmed this after high resolution images were submitted.

Once the photographs are submitted, AI returns an ID that is most probable, and gives the reasons. It also tells you what else it could be, and what other tests to do to confirm it.

In addition to identifying coloured gemstones, it was able to identify egg yolk amber, Baltic amber, and fake amber. Once again, in some cases I already knew certain pieces were real amber, and certain ones were fake. In fact, with amber, when doing the macro photography, you can see the glow from within if it is real. If it is fake, you can tell by the difference in the photography, and also by the difference in the way it feels.

The AI rock identification site was also able to differentiate coloured gemstones from dichroic glass with surprising accuracy.

It does stand to reason that the AI identification can have a higher degree of accuracy with high resolution macro photos, because there is a huge database of imagery to draw upon. Gemstones are naturally categorized by their colours, inclusions, reflection, and refraction of light. These are things that can be picked up in the images.

AI also successfully identified rock crystal, and camphor glass, which was quite impressive, especially when you consider that camphor glass was used in jewellery a hundred years ago, and is not used in modern jewellery.

On Reddit there are people who posted that they submitted a burnt marshmallow, and even ice, and got back results claiming it was some kind of stone. We should keep in mind that if you try to defraud any system, it is likely to give a “wrong answer”. After all, the websites and apps are set up to identify rocks, and other organic materials, not attempts to fool the system.

We read and hear so much about AI being akin to the antichrist, yet like most things related to knowledge, some of it is good, and some of it is not.

AI also identified heat treated gemstones. When you study these items, and are worried about making mistakes in the identification, this is a helpful tool. When you add this assessment to what you already know about the items, it helps to increase your own percentage of accuracy.

It also helps to narrow down what items are worth sending to a certified gemologist.

Personally, I would stay away from AI if it relates to things we should not dabble in, or to creating some kind of fantasy world blurring the lines with reality.

But in the case of identifying rocks, coloured gemstones, amber, tortoise, horn, bone, and perhaps ivory (although I have not tried submitting ivory yet), it is a valuable, and interesting resource.

For those of us who are not certified gemologists, AI is a very helpful tool when it comes to confirming our own conclusions, as well as helping to identify the many mystery stones.

So far, I have submitted sixty to seventy-five items, and have at least a couple hundred more to assess.

From what I have learned so far, it is very important to submit high resolution photographs. It is also important to view it as an adjunct to other research, and what is known about the items, the settings, and the era each one came from.

Rocks, gemstones, and other materials can be very old, but when it comes to the settings, they too, provide clues to help with identification. For example, Navajo and Zuni will use turquoise from a certain area, and will also often use settings that are unique to their own design, and history.

Amber comes from many different locations throughout the world. As a hobbyist, or collector, we do not automatically know the unique properties and inclusions in the various types of amber.

Although I have many more items to test, and will remain cautiously optimistic, in my experience thus far, I think AI is a very good tool for helping to identify coloured gemstones, and other organic materials used in jewellery.

It is also very interesting to read the detailed description they give on each item. There are so many things I have wondered about. I am well aware of the limitations of my own eyesight, so I have to say, this is one area of AI knowledge that I think is worthwhile.

Update – after a few more days of testing the site I was using started limiting the number of items it would scan. Initially it was 20-30 items per day, then it dropped to 8, and now to about 5. There is not an option to buy a subscription for unlimited use on that site. 

Therefore I will have to find a site with a subscription, in order to submit the remaining items. 

The AI information does help increase knowledge of coloured gemstones, especially when it comes to differentiating quartz from feldspar minerals. 

In most cases when it comes to vintage, and antique jewllery, the pricing is based partially on the setting, partially on the size and number of the stones, and partially on the value of the stones. 

For certified gemologists who are dealing with high value diamonds, emeralds, sapphires and rubies, I can certainly see why hands on evaluation and testing is required. 

For coloured gemstones such as agate, carnelian, jasper, moonstone, tiger’s eye, sodalite, garnet, amethyst, nephrite, lapis lazuli, and so on – the item can be identified for its beauty, size of the stones, and the setting. These items are generally less than $1000, and are not considered high end. 

If you look at new jewellery with coloured gemstones, you seldom see large stones anymore. Agates that are cut ond polished into large bead necklaces, are beautiful. You do not find them in the average jewellery store. 

Once again, the vintage market offers a range of unique and much larger coloured gemstones than you can find in modern day jewellery. 

From what I have learned thus far using AI as an identification tool, it is educational, and confirms what you might have already concluded through research, and gradual learning about coloured gemstones. 

It does have certain restrictions, and is probably limited in how reliable it is. However, when combined with hands on evaluation, research, settings, and how the item is made overall, it does help to improve the description of each item with more detail. 

Copyright Valerie J. Hayes and Quiet West (2026). Unauthorized use and/or duplication of this material without express and written permission from this blog’s author/owner is strictly prohibited. Excerpts and links may be used, provided that full and clear credit is given to Valerie J. Hayes and Quiet West with appropriate and specific direction to the original content.