Why do two DNA services give different health results from the same file?
Upload the same raw file to two interpretation services and you will get two different reports. Sometimes they contradict each other outright — one flags something the other does not mention at all.
Your DNA did not change between uploads. What changed is everything layered on top of it, and that layer is most of the product.
Why do two DNA services give different health results?
Because the file is only the input. Each service chooses which markers to report, which studies to trust, how strong the evidence must be before something is included, and how to phrase the result. Those four choices are made independently by each company, and they compound.
Think of the raw file as a photograph and each report as a description of it. The photograph is fixed. What gets described, in what order, with what emphasis, is entirely up to the describer.
This is why comparing two reports rarely settles anything. They are not competing measurements of the same quantity; they are different editorial products built from one shared dataset.
Do all DNA services use the same underlying research?
Largely, yes — and that surprises people. The foundational sources are open: ClinVar for variant significance, gnomAD for population frequency, the GWAS Catalog and PGS Catalog for associations and scores, and public pharmacogenomics resources for medicines. Almost everyone draws on the same pool.
Some services also license commercial databases, though product pages rarely name them, so a claim of proprietary data often refers to internal curation rather than exclusive access to anything.
The practical consequence is that a service is not usually differentiated by what it knows. It is differentiated by what it chooses to say, and by how carefully it says it.
Why does one service report something another leaves out?
Different evidence thresholds. A marker with one supporting study and no replication will appear in a report built for breadth and be excluded from one built for confidence. Neither is lying; they have set the bar in different places, and where that bar sits is usually undisclosed.
Report counts follow directly from this. A service advertising 500 reports has not found more in your file than one advertising 50 — it has set a lower threshold for inclusion, or it is counting differently, or both.
That is worth remembering when comparing products by feature count. A larger number in the marketing means a lower bar somewhere, and the bar is the thing you actually care about.
Which report should you believe when they disagree?
The one that shows its work. A result you can check — the marker named, the source cited, the strength of evidence stated, and the coverage of your own file disclosed — can be verified. A result presented as a conclusion with no visible basis cannot be, however confident it sounds.
Confidence of tone is not evidence of quality, and in this market it often runs the other way. The reports that read most decisively are frequently the ones that set the lowest evidence bar.
If two reports disagree on something that matters to you, that disagreement is itself information: it means the underlying evidence is not settled. The right response is a clinician, not a third report. A line in a raw DNA file is not a diagnosis.
Does the same problem affect ancestry results?
Yes, for a related but distinct reason. Ancestry estimates compare your markers against a company’s own reference panel of people with known origins. Different panels, different region definitions and different algorithms produce different percentages from identical data — and panels get updated, which is why your results can shift on their own.
We have written about the ancestry case separately, because the mechanism there is specific enough to deserve its own explanation — see what the major services’ files actually contain.
The pattern is the same throughout consumer genetics, though. The file is stable; everything built on top of it is a choice, and choices vary.
How we turn published research into a report → · What the major services’ files contain →