Three Small Matches, an AI Assistant, and the Dutch Hudson Valley

A while back I was scrolling through my father’s DNA matches and noticed three women who all shared roughly 20–22 cM with him — small matches, easy to overlook. But they also shared DNA with each other, in uneven amounts (one pair shared 336 cM, another pair only 109 cM), which should mean there’s a triangulated connection sitting somewhere in their trees.

I didn’t have time to work through all three trees side by side. So I did something I hadn’t tried before: I took screenshots of their public trees — just the historical branches, no living people — and dropped them into a conversation with an AI tool, asking it to spot any common surname or branch.

It found one almost immediately. All three trees contained a man named John Bennett Groesbeck, in the same generation. Match A descended through one of his wives. Match B descended through the other. The third tree was sparser but pointed in the same direction. That asymmetric pattern — different cM amounts across half-lineages — fit what I’d see if these were descendants of two different wives of the same man.

I might have skimmed right past John Bennett Groesbeck’s name without recognizing it. But two things jumped out the moment I read it: his middle name was Bennett, and he was born in Livingston County, New York. Both of those have a place in my own tree.

A man I believe was my seventh great-grandmother’s brother moved to Livingston County from Bucks County, Pennsylvania, in the early 19th century. I’ve long suspected that he and my seventh great-grandmother had half-siblings who carried the Bennett surname — though that’s a hypothesis I’m still actively researching, not an established line. Separately, I have reason to believe that my seventh great-grandmother’s mother came from the Dutch Reformed community of the upper Hudson Valley — the same world the Groesbecks were part of. So there are potentially two threads connecting my father’s tree to John Bennett Groesbeck’s: a Bennett line that may not exist, and a Dutch maternal line that probably does. Either could be the answer. Both could be involved.

The catch is the math. Twenty centimorgans is a small match — typical of a fourth-cousin-range relationship, though the actual cousin distance could be much closer or much further. At the level of seventh-great-grandparents (roughly eighth cousin), most DNA companies wouldn’t reliably predict a relationship at all, and many real relatives at that distance share nothing detectable.

But there’s a wrinkle. My father’s second-great-grandparents were first cousins, descended from this same line. That kind of consanguinity changes the math — the same ancestral segments can get reinforced through multiple paths and survive across more generations than they otherwise might. And the Groesbecks were part of the upper Hudson Valley Dutch settlement network — Rensselaer County, Schaghticoke, Albany — communities notoriously endogamous, with a relatively small pool of families intermarrying repeatedly across generations. If my Dutch maternal hypothesis is right, and especially if the Bennett line turns out to intersect with that world too, I’d be looking at endogamy on both sides. Exactly how much that would inflate a connection at this distance is hard to quantify, but the principle is well-established.

The connection isn’t proven. The next step is filling in John Bennett Groesbeck’s maternal line and seeing whether it touches either of those threads in my tree — the possible Bennetts or the Dutch maternal line. Maybe both. Maybe neither.

Small matches like these often go unworked because they’re too small to predict a relationship from. But they can still be solvable, especially if you approach them in new ways. AI-assisted scanning of multiple trees side by side surfaced a name and place I would have missed. My own familiarity with my tree gave that name and place meaning — neither would have stood out alone, but together they echoed something I’d been working on.