ChatGPT (AI) analysis of my US Coin Type Set

Well, I just spend the last few hours with ChatGPT and my ~135 coin Type set. I'm very pleased with the results. It found a few small errors in my type year ranges, suggested some new coin slots, added diameter, weight, edge type, to all my Type slots, among other things. Works remarkably well.

Basically anyone with an excel spreadsheet for their coin collection that wants some insight, and doesn't mind sharing the coin details, it would be well worth you time to upload it. Even if just for catching errors.

I also found the insight into my collection interesting, which is why I am sharing it here. I know one or two of you with think this is cool.

this is some of what it had to say:

1. The strongest pattern is grade versus age
Your collection isn't pursuing a uniform minimum grade. Instead, the grade rises almost continuously as the coins become younger and more available:

Coin date......Owned.....Median numeric grade
Before 1800.........4.........3.5
1800–1849.........23.........30
1850–1899.........40.........51.5
1900–1949.........33.........64
1950–1999.........22.........66
2000-present........9.........69

That's a remarkably clean progression.

It tells me you're essentially practicing era-adjusted condition collecting: early coins are allowed to be heavily circulated because simply owning an attractive, problem-free example is the accomplishment; by the 20th century, your standard naturally rises to Choice/Gem Mint State; modern representatives are essentially superb-grade pieces.

That's quite different from registry-set optimization, where grade points and weighting explicitly drive rankings.

2. Your date choices aren't simply “buy the cheapest common date”
This was the part I found most interesting.

Among representatives whose type has a finite multi-year range:

• 23 are the first year of their type
• 23 are the last year
• 62 are somewhere inside the type's run
• another 18 represent single-year types

So roughly 35% of your applicable representatives are boundary-year coins—either the introduction or final year of the design/type.

That makes the collection more historically expressive than one assembled solely by asking, “What's the cheapest date for this type?”

PCGS's traditional description of type collecting specifically says collectors commonly identify the more common dates and buy the nicest examples they can afford. Your choices frequently depart from that simple model.

3. Your mint mark distribution is:

Mint.......Coins
Philadelphia/no mintmark*.......90
San Francisco.......18
Denver.......13
New Orleans.......7
West Point 2
Carson City 1
Manila 1

*For these purposes I'm treating blank as predominantly Philadelphia, while recognizing that the colonial/token/medal material isn't literally Philadelphia Mint production.

So yes, the collection is overwhelmingly Philadelphia.

But your 19th-century branch-mint coins are disproportionately interesting. You have O-mint Seated material across several denominations, including the 1838-O No Stars dime, 1840-O No Drapery dime and half dime, 1840-O No Drapery quarter and 1860-O Seated dollar.

And there's a CC representative—the 1877-CC Seated quarter.

So you haven't pursued mint diversity for its own sake. Rather, branch mints appear where they make the individual type representative more interesting.

4. PCGS dominates, but NGC is substantial

You own:

102 PCGS (77.3%)
30 NGC (22.7%)

That's about 3.4 PCGS coins for every NGC coin.

I wouldn't interpret that as a judgment about the services. It simply tells us that over years of hand-selecting coins, your collection has naturally become heavily PCGS-oriented.

5. Your remaining holes are now extremely concentrated
This is perhaps the best measure of how mature the collection has become.

You currently have 132 filled slots out of 141 = 93.6% complete.

And only 9 NEED slots remain:

1. Draped Bust Heraldic Eagle Dime
2. 1807–08 Original Design Capped Bust Half
3. Liberty Head $2.50
4. 1903–04 large 10 Centavos
5. 1903–05 large 20 Centavos
6. 1907–29 reduced 20 Centavos
7. 1903–05 large 50 Centavos
8. 1903–06 large Peso
9. Lafayette Dollar

What's striking is their concentration.

Five of the nine are U.S.-Philippines.

So your mainstream U.S. type collection is actually much closer to completion than the raw 93.6% figure suggests. The recent decision to split the Philippines denominations by physical type created a large portion of the remaining holes.

6. What I think the collection says about you as a collector
After looking at the actual coins rather than just the slot definitions, I'd characterize it as a historically oriented advanced type collection with condition expectations scaled to availability.

You aren't chasing registry points. You aren't imposing an arbitrary grade floor. You aren't simply buying the cheapest date that satisfies each slot. And you're not using Proofs to make difficult circulation types easier.

Instead, the pattern is roughly:

Early rarity → accept honest circulation.
19th century → seek increasingly strong circulated/low-MS examples.
20th century → generally Mint State.
Modern → extremely high grade.

At the same time, you seem to have a noticeable attraction to transition points—first/last years, short-lived designs, physical changes, unusual denominations and significant subtypes. That matches what we've seen when deciding which rows deserve separate slots. PCGS notes that type collecting doesn't have one universally correct definition and can be expanded into more technical differences according to the collector's interests; your set is very much in that latter category.


Example question I asked at the end of the analysis:




Filling in the diameter and weight and metal composition for each coin I own was worth the price of admission alone, however it did all suggest some edits / fixes.

An example corrections it found for me:

Liberty Head Nickel, With CENTS — 1883–1913 should be 1883–1912. The famous 1913 Liberty nickels were not a normal circulation issue, so 1913 shouldn't define the range of your business-strike type.

Also, your Silver Eagle Type 2 Eagle Landing row says 2023–Present, but your own representative coin is a 2021-W. The redesign actually began during 2021, with both Type 1 and Type 2 produced that year. The Mint specifically says the new reverse began appearing in mid-2021. That one is an unambiguous correction to 2021–Present.

It talked to me A ALOT about my set, and probably could keep doing so. I am just sharing some key points here on CT. I just wanted to give you AI-skeptics a little real world example of how this can help your hobby.


TLDR: AI doesn't need to be so pie-in-the-sky or an all encompassing menace, it can just be like a handy and polite secretary that tells you when you've made a mistake that you would otherwise not notice.

 

Analysis like this can be fun, can't it. I'm always compiling statistics and stuff in my head. Even though most of it is for my own benefit and never gets shared.

 

Well, I just assume any content you give online is public, even if they say otherwise. Basically, I already participate in multiple coin registries/ collection managers. So this information of my what I own is already shared all over the place. Because of that, I felt comfortable taking this small extra step.

 

You are correct in your assumption that I did not gain some massive new understanding about my own collection, I knew most of it already. I have the stats already, and know how to produce them quickly in an ad-hoc manner as needed.

However, I am also highly organized and sort of geek out on the "cataloging" part of the hobby. I don't know for certain of course, but I would imagine my coin database is beyond the scope of 95% of collectors, even if my coin count is quite small.

I think the real magic would be to upload a collection with limited info, (e.g. only year, mint mark, denomination, or perhaps only cert#, and let it populate the 30+ columns of data for each coin. Essentially replicate my database but with close to zero effort.

ChatGPT basically told me it can take total garbage from someone, like a badly organized collection of coins, malformed excel file, and produce all the fields that I already had, with minimal effort. I did not test it, but based on the filling in diameter, weight, metal composition, for each coin. I have close to zero doubt about it's ability to do so.

Example of how specific it was, it assigned a composition value for every single slot in my Type set, and did so quickly. Here is a summary of that work:



It would have taken me quite some time and effort researching each of these Type slots.

 

I really like expanding the typical Type set to include some non-traditional items like US Medals, Bullion coins, US Philippines, Colonials, private tokens, etc.

I wanted a representative type set of what really happened with coins in this country. It would be hard to do that without including things like hard time tokens, or private and foreign colonial issues.

I agree, I never had any interest in that, and I dont think hardly anyone tries, so my set is not really unique for varying the grades. I do think perhaps my set varies them MORE than the average set, as I was specifically trying to get at least 1 example from every numeric grade. I think perhaps people hover around grade "buckets" more than I do, like mostly AU coins, or mostly XF coins, etc. ChatGPT likely noticed my extreme spread.

I agree 100%. I never attempted to buy the first year, last year, or any any specific mix. I bought what I liked and what I could afford. I "accidentally" bought a lot of first year and last years coins, and the AI read more into it than was true.

Thank you for your comments.

 

I’m 100% in agreement with that, and follow that sort of playbook myself. I’m too eclectic to stay within the traditional boundaries, even if those little sideshows dilute my budget and compete for my attention.

Of course in my case, I’m SO eclectic that one could say the organized type set portion of my collection is actually the sideshow and the rest is the main event. If I were to sell off my World and Ancient collections as well as my Exonumia and commems, I could build a much nicer and more complete US Type set by focusing more exclusively on that. But I’m not willing to do that. There’s just so much other fun and interesting stuff out there.

 

Could AI advisors be the wave of the future? Maybe, and maybe not. I don’t think I’m quite ready for that yet, but I can see where it could become a useful supplemental tool.

 

Yeah, I got ChatGPT giving me a few leads every day. I figure only a matter of time until I pull the trigger on one of them.