09/29/2026
Today I want to tell you a little about how I arrive at the dates, attributions, origins and values I give to the objects we find.
The first thing I do is photograph the most obvious details: marks, construction, materials, wear, fittings, decoration, anything that might provide a clue. I then run those images through several AI programs — usually Gemini, ChatGPT and Claude.
If their answers are broadly similar, I move on. If they disagree, things become more interesting: I give each program the others’ conclusions and ask them to challenge them.
Sometimes the reactions are quite funny. ChatGPT, usually the least patient of the three, is perfectly happy to criticize my theories and occasionally seems to take disagreement almost personally. To be fair, it is also the program I have tried hardest to “train” to be direct and critical with me. The problem is that, once I make a decision, it can suddenly become very accommodating and find a way to justify almost anything I have decided — which is precisely why I don’t trust it blindly.
Once I have extracted the information that seems most credible, the real research begins: antique dealers’ archives, museum collections, auction records, image databases, old catalogues, specialist websites and reference material.
This is also where personal knowledge becomes important — in my case, my background in architecture and art history, being European, and simply having spent much of my life looking at old things. AI is useful, but it does not replace experience. It gives you clues; you still have to know which clues make sense.
Once an object is reasonably well defined in terms of date, origin and use, I go back to AI and ask it to help search for comparable sales from recent years.
And this is where you have to laugh.
Prices for apparently similar objects can range from a few dollars to several hundred. One of the most important things is to specify that you want sold prices, not asking prices. We all know that everyone likes to believe there is a hidden treasure somewhere in the house.
So this is really just a small warning about identification apps and instant AI valuations. Many people use them — including customers in our shop — and there is nothing wrong with that, provided they are used cm grano salis: with a grain of salt.
As a starting point, they can be very useful. But if the object appears to be important, unusual or potentially valuable, it is worth doing considerably more research.
And then, of course, the market occasionally makes fools of all of us.
That biscuit jar we might have thrown away without a second glance turns out to be highly desirable to a very specific group of collectors.
The difficult part is not discovering that the collectors exist.
The difficult part is getting one of them to walk through our door in Saskatchewan.