GedMend › Guides

What to look for in a GEDCOM merge tool

Six questions to take to any tool that offers to combine two family trees, including the merge feature already built into the program you use.

Last updated 1 September 2026

This page deliberately names no products. Tools change faster than any page can track, and the honest thing to hand someone is the list of questions rather than a snapshot of answers that will be wrong in six months. Take these to whichever candidates you are weighing, including the one this site sells, and to the merge feature already built into the program you use.

1. 1. Does it modify your original files?

   The single most important question, and the one most easily answered. A merge that happens inside your master tree has put the thing you cannot replace into the blast radius. A merge that reads two files and writes a third has not.

   This also determines what a mistake costs. If the output is a new file, a merge you dislike costs you the time you spent on it and nothing else — delete it and start again. If the merge wrote into your master, you are restoring a backup, assuming you took one, and assuming you notice in time to know which one.

   Ask: after a merge, are both input files byte-for-byte as they were? Is the result always a new file?

2. 2. Do you confirm each match, or does it match in bulk?

   Every tool of this kind produces a list of candidate pairs. The difference that matters is what happens next. A candidate list is a reading order. It is not a decision.

   Bulk matching — "merge everything above 85%" — fails in the direction you cannot recover from. Two different men with the same name, town and decade will score very high indeed, because they genuinely look alike. Fusing them corrupts two lines simultaneously and leaves no trace that it happened.

   Ask: can it merge anything I have not personally looked at? Can I leave a pair undecided and have that mean "keep both"?

3. 3. Does it show you why two records matched?

   A score with no reasoning behind it cannot be argued with, which means it cannot be checked. You want to see the parts: how much came from the names, how much from the dates, how much from shared family, and what tipped it.

   Related, and worth asking directly: is it deterministic? If the same two files can produce different results on different days, you have no way to verify anything, no way to reproduce a problem, and no way to hand a colleague the same view you are looking at. Rule-based matching gives the same answer every time. Probabilistic and language-model matching does not, and it will not tell you when it has changed its mind.

   Ask: can I open any proposed pair and see the arithmetic? Will the same two files give the same result next month?

4. 4. What does it do with source citations?

   The most commonly lost thing in a merge, and the least commonly asked about. An unsourced tree is a rumour. Three separate behaviours are worth checking, because tools differ on all three:

   • Are citations from both records kept when two people are merged — or only those attached to the value you chose to keep?
   • Do citations stay attached to the specific fact they support, or are they loosened into a general note about the person? A citation that no longer says which fact it supports still counts, but it no longer proves anything in particular — what happens to your citations when you merge goes through this in detail.
   • What counts as a duplicate citation? Two references to the same register at different folios are not duplicates, and a tool that collapses them is discarding evidence.
   Ask: if I choose File A's birth year, does File B's evidence for its birth year survive?

5. 5. Does it handle legacy text encodings?

   Genealogy files carry text in several different schemes for historical reasons, and one of them — the encoding the original standard chose — stores an accent as a separate mark placed before the letter it belongs to. Software that assumes otherwise mangles exactly the names you were most careful about.

   A file can also declare one encoding in its header and contain another. Inherited files do this routinely. The behaviour you want is a tool that decodes what is actually there rather than believing the label, and that tells you when the two disagreed.

   Ask: does it warn me before merging if the two files use different encodings? What happens to a character it does not recognise — is it flagged, or quietly replaced?

6. 6. Does it show its work before writing, and after?

   Before: a count you can verify. People in A, plus people in B, minus the pairs you confirmed, equals people out. It is simple arithmetic and it catches an entire category of failure. A tool that will not show you that number before it writes is asking you to take the result on faith.

   After: a written report of every match confirmed, every field decision, and which of them were yours rather than a default. You will want it long after you have forgotten the session.

   Ask: can I see the expected counts before anything is written? Do I get a record of my decisions I can still read in ten years?

Three questions that are not about merging at all

Where does the file go? Some tools do this work in a browser, some on a server, some entirely on your own machine. All three can be legitimate, and the useful question is not the technology but what you are told: is it stated plainly, and can you check it? A tool that works with your network disconnected has answered the question in a way that needs no trust.

What happens when it is wrong? Every tool of this kind is wrong sometimes. What you want to know is which way it fails. A tool that leaves two records separate when it was unsure has failed in a direction you can fix in ten minutes. A tool that fused them has not.

Is it honest about what it does not do? This is a surprisingly good proxy for everything else. A page that lists limitations — the formats it will not write, the merges it will not attempt, the duplicates it may miss — was written by someone who has thought about the failure cases. A page of superlatives was not.

A shorter version

If you only ask two questions, ask these. Does it touch my original files? And can it merge anything I have not looked at? A tool that answers no to both is one you can afford to be wrong about.