Elena had been chasing her mother's line for years when the DNA hint finally seemed to click. A "potential ancestor" suggestion, stitched together from the trees of forty-some matches, all agreeing on the same seventh-great-grandfather. She added him to her tree and moved on. Two years later, a distant cousin agreed to take a Y-DNA test to settle a different question entirely, and the result didn't just fail to confirm the ancestor. It ruled him out. Completely. The real common ancestor sat three generations further back, a man none of those forty trees had ever mentioned. The matches were real. The DNA connection was real. The ancestor everyone had agreed on had been invented by consensus, not evidence.
Pick the test that answers your actual question
Before any of this works, you need to have run the right test in the first place, and a surprising number of people haven't. The test most people mean when they say "DNA testing" is autosomal: it looks at DNA inherited from both parents in roughly equal measure, and it's the test behind every match list, every centimorgan figure, and every hint engine on the market. It's strong for finding cousins within roughly five to seven generations. Past that, it gets weaker, because each generation of recombination shuffles and dilutes the signal until a real relationship stops leaving a detectable trace at all.
Two other tests answer a completely different kind of question. A Y-DNA test only works for men, since only men carry a Y chromosome, and it passes down the direct paternal line (father to son to son) largely unchanged for many generations. That makes it the right tool for surname studies and for testing one specific paternal line against one documented man, which happens to be exactly the question Elena's cousin had. Mitochondrial DNA runs the other direction: everyone inherits it from their mother, but only daughters pass it on, so it traces a single direct maternal line with that same kind of stability. Neither test tells you anything about the dozens of other lines feeding into your tree. That trade-off is the point, not a flaw. A narrow, stable answer to one lineage question beats a wide answer that fades with distance once you're several generations out.
Elena's cousin didn't need a second autosomal kit. Autosomal DNA had already said everything it was going to say about a relationship that far back: a scatter of shared segments too thin to pin down a specific ancestor. A Y-DNA test asked a sharper question of a line that inherits differently, and it answered it.
A centimorgan is a range, not an answer
Every DNA match comes with a number: shared centimorgans, the amount of DNA you and a match have in common. It's tempting to treat that number like an address, one relationship per figure. It isn't. The Shared cM Project's data, built from thousands of confirmed relationships, shows the same amount of shared DNA turning up across a genuine spread of possibilities. An amount that looks like a first cousin once removed could just as easily be a half first cousin, a great-aunt, or a great-great-grandparent showing up at the outer edge of what identical-by-descent DNA can produce. Recombination isn't distributed evenly either, so two full siblings can share noticeably different amounts of DNA with the same third cousin. The number narrows the field. It doesn't hand you the relationship.
Identical by state versus identical by descent
There's a second reason small centimorgan amounts deserve suspicion, and it sits underneath the ranges the Shared cM Project documents. Every shared segment of DNA is either identical by descent, meaning it was passed down intact from one real, shared ancestor, or identical by state, meaning it happens to match by chance because human DNA only has so many possible sequences to go around. Above roughly twenty centimorgans, a shared segment is almost certainly identical by descent. Below about seven, the odds shift meaningfully toward coincidence, especially on a single small segment with nothing else backing it up.
This is exactly why a distant relationship claim resting on a handful of centimorgans deserves more scrutiny, not less. A five-centimorgan match presented as a fourth cousin might be exactly that. It might also be two strangers whose DNA happens to overlap at one stretch of one chromosome, with no shared ancestor within any genealogically useful timeframe at all. Testing companies set matching thresholds to filter out the most obviously coincidental noise, but plenty of small, borderline segments still make it into a match list and get treated with the same confidence as a fifty-centimorgan match that's almost certainly real. The number on the screen doesn't tell you which kind of segment you're looking at. Only more evidence does.
How a hint engine invents an ancestor
The "potential ancestor" feature behind Elena's mistake works by scanning the trees attached to your matches and looking for a name every path seems to converge on. If forty people's trees name the same great-grandfather, the tool reads that convergence as agreement. But those forty trees usually didn't do forty independent pieces of research. Most were built by copying a match's public tree, importing a GEDCOM that already carried the error, or clicking the tool's own one-click "add to tree" button on an earlier, equally unverified suggestion. Genetic genealogists have documented cases exactly like this in the wild: one "potential ancestor" hint backed by ninety-eight separate matches was later disproven by a single Y-DNA test, and a sample of a hundred and fifty trees carrying that same ancestor turned out to have copied the identical, disprovable error from each other. Volume of agreement isn't evidence. It's the same unsourced guess, pasted a lot of times.
Sorting matches into branches before you trust any of them
Before leaning on any hint, it helps to know which branch of the family a match even belongs to. This is what genetic genealogists mean by a genetic network, or clustering: grouping matches by which other matches they share, so people who all descend from one great-grandparent end up in one cluster and people from a different line end up in another. You don't need the paper-trail relationship yet. You just need to notice that match A, match B, and match C keep showing up together, and that cluster looks like your mother's side, not your father's. Once matches are sorted into clusters, a hint that puts the "common ancestor" on the wrong side of the family becomes obvious fast, no Y-DNA test required.
Triangulation: three lines, one segment, one ancestor
A single confirmed match, however solid, is still one data point. Triangulation turns a handful of data points into something closer to proof. The method: find three or more people who share the same segment of DNA (same chromosome, same start and end position) and who each independently descend, on paper, from the same documented ancestor. Same physical stretch of DNA, three separate paper trails, one person they all converge on. That's far stronger evidence than any single match or any hint engine's guess, because it's genuinely hard to produce by coincidence.
Setting it up takes more than staring at a match list. A match list only tells you two people share some amount of DNA in total. It doesn't say where on the genome it sits. Triangulation needs a chromosome browser or a shared-segment comparison tool instead, the kind that shows the actual start and end points of an overlapping segment across several matches at once. That's how you tell whether three people match you at the same physical location, instead of matching you separately at three unrelated ones. Group the matches who overlap at the same segment. Confirm each one's documented line back to a common ancestor independently, not by comparing trees with each other, and only then treat that ancestor as confirmed. It's slower than accepting a cluster of matches at face value. It's also the version of the evidence that would actually hold up if someone else went and checked it.
Endogamy breaks the math entirely
All of this assumes a family tree that spreads outward with each generation instead of doubling back on itself. Endogamy (marriage within a small, relatively closed population over many generations) or a more localized run of cousin marriages breaks that assumption. When the same ancestors keep reappearing on multiple lines, descendants inherit DNA from those people more than once, and total shared centimorgans climbs well past what an ordinary chart would predict. Two actual second cousins from an endogamous community can share enough DNA to look, on paper, like first cousins. That's not a flaw in the test. It's an accurate reflection of a smaller, more interconnected gene pool. Anyone with known endogamy in their background (common in Ashkenazi Jewish, Acadian, and many Pacific Island and Indigenous family lines) should treat a relationship estimate as a rough placement, not a calculation.
When DNA reveals what paper records never would
Sometimes a DNA test doesn't just fail to confirm a documented ancestor. It surfaces something no paper record ever could have: a parent who isn't the biological parent a birth certificate names. A discovery like this shows up often enough in DNA testing communities that genetic genealogists have a shorthand for it, and it's one of the sharpest differences between DNA and documents as evidence. A birth certificate records what someone told a clerk at the time. It doesn't, and can't, record a biological fact the people involved may not have known themselves.
This deserves more care than an ordinary brick wall, because the person on the other end of that discovery didn't consent to learning it, at least not in the moment they took a test out of ordinary curiosity about where their family came from. There's no research technique that makes that conversation easy, and there shouldn't be a playbook that treats it as just another puzzle to solve. Slow down. Confirm the finding with a second test or a second line of DNA evidence before treating it as settled. Give the people it affects time and privacy before it goes anywhere near a shared tree, a family group chat, or a video. The DNA is real. How, whether, and when it gets shared from there is a separate decision, and it belongs to the people it's actually about, not to whoever found it first.
Let the paper trail do the confirming
None of this is an argument against using DNA. It's an argument against letting a match list stand in for proof. A shared-cM figure and a hint-engine suggestion are both leads: reasons to go looking for a specific document, not permission to skip looking for one. Confirmation still has to come from something with a date and a place that can hold up on its own: a marriage record, a will naming an heir, a baptismal register. That's the order Deep Research follows when a DNA lead comes into a tree. It checks what the matches and the network suggest, goes looking for the document that would have to exist if the suggestion is true, and says plainly when that document isn't there yet.
A match list will always be more available than a document. That's exactly why it needs one behind it. The centimorgans tell you where to look. Only the record tells you what actually happened.
