No. DNA testing companies do not share your results with each other. Each company genotypes your sample in its own lab, against its own chip, and stores the raw data in its own database. Results move between companies only when you explicitly export and upload your raw file, or when a company's terms and privacy settings permit a specific transfer. Even then, the receiving company reinterprets the same raw data through a different pipeline, so matching percentages and health flags often diverge.
That answer sounds simple, but the mechanism behind it explains nearly every surprise consumers encounter. Two siblings test at two brands and get different "ethnicity" breakdowns. A health report at one company contradicts a report at another. A distant cousin appears on one platform and not another. None of this is a malfunction. It is the predictable output of separate reference panels, separate algorithms, and separate consent contracts.
The genotyping pipeline and why it isolates data
Consumer DNA testing is not whole-genome sequencing. Most companies use a genotyping microarray, a glass or silicon chip spotted with short synthetic DNA probes. Each probe is designed to bind a specific spot in the genome where humans commonly carry a single-letter variation, called a single nucleotide polymorphism (SNP). A typical consumer chip reads somewhere between 600,000 and 1,000,000 SNPs, a sliver of the roughly 3 billion base pairs in your genome.
The chip design is proprietary. Companies choose which SNPs to include based on their own research priorities, the populations they serve, and the traits or conditions they sell reports for. A chip optimized for ancestry in European populations may carry sparse coverage of variants informative in West African or Indigenous American lineages. That design choice is baked in before your saliva ever reaches the lab.
Once your sample is processed, the output is a raw data file: a long list of SNP identifiers paired with your genotype at each position. That file is the only portable artifact. The interpretation layer, meaning the reference panels, the phasing algorithms, the trait models, and the report generation, stays with the company. Federal law reinforces the separation. The Genetic Information Nondiscrimination Act of 2008 prohibits health insurers and employers from using genetic information in coverage and employment decisions, but it does not create any data-sharing obligation between companies (U.S. Equal Employment Opportunity Commission). Nothing in the regulatory framework requires one lab to recognize another lab's interpretation.
A plain-language way to hold this: the raw file is your data, but the meaning attached to it is the company's product.
Quantified context: how much results actually differ
Direct head-to-head studies are limited, which is itself a finding. The most-cited independent comparison asked the same individuals to test with multiple direct-to-consumer ancestry services and found that continental-level estimates, such as the broad percentage of European versus African versus East Asian ancestry, tended to agree within a few percentage points. Finer-grained estimates, such as the specific country or subregional breakdown, diverged substantially, sometimes by double digits for the same person.
Three mechanisms drive the gap. First, reference panels differ. A company's "Irish" cluster is defined by the samples it collected, and those samples are not the same set another company uses. Second, algorithms differ. Some assign ancestry in short chromosomal segments; others use different smoothing and confidence thresholds. Third, reporting granularity differs. One company may report "British and Irish" as a single category while another splits it into four.
Health and trait reports diverge even more because they rest on interpretive choices rather than shared measurement. Two companies can read the identical genotype at a variant and attach different risk language, different population baselines, or different confidence labels. The genotype agrees; the story does not.
Price is not a proxy for sharing either. Consumer ancestry kits commonly sell in the USD 40 to USD 100 range, while health-focused products run higher. Paying more buys a different interpretation, not access to another company's database.
Where results do cross company lines
Results cross boundaries through a small number of well-defined channels, and each has a distinct consent structure.
Raw data upload: you download your file from company A and upload it to company B, which reinterprets it. Many companies accept uploads for ancestry; fewer accept them for health reports, and some charge an unlock fee. Matching opt-in networks: some platforms let you compare against a partner database only if you toggle a specific setting. The default is usually off. Research consents: you may authorize use of de-identified data in research partnerships. This is governed by your consent form, not by inter-company reciprocity. Third-party interpretation tools: independent services accept raw files and generate reports. These are not the original labs and carry their own accuracy caveats. Legal compulsion: a valid court order or warrant can compel a company to produce data it holds. Company transparency reports document how often this occurs.
The counterintuitive detail: uploading your raw file to a second company can produce a report that looks authoritative but rests on genotypes the second company never verified. If the first lab's chip had low coverage at a variant the second company reports on, the second company may be interpreting a gap, not a measurement. Professional genetic counselors flag this as a real source of false reassurance and false alarm.
Edge cases that break the sharing assumption
Identical twins who test at different companies can receive different ancestry percentages from the same shared genome. The genotypes are effectively identical; the reference panels and algorithms are not. This is the cleanest demonstration that the output is interpretive.
Adoptees and people with unexpected parentage face a sharper version of the same problem. A close-family match at one company may not appear at another because the matching database is smaller, because the relative tested elsewhere, or because one of you did not opt into matching. Absence of a match is not evidence of absence of a relationship.
Sibling pairs illustrate the inheritance mechanism. Full siblings share roughly 50 percent of their segregating DNA on average, but the actual figure varies because of recombination. Each sibling inherits a different shuffle of each parent's chromosomes. A trait or ancestry segment present in one sibling can be absent in the other, and each company's chip samples a different subset of positions. Two siblings can therefore get genuinely different, correct results.
People with mixed or underrepresented ancestry hit the panel-coverage limit hardest. When your ancestry falls outside a company's reference samples, the algorithm has fewer anchors and leans on broader regional categories. The percentage may still be reported with the same confident-looking interface, which overstates precision.
Misconceptions about shared databases
The most common misconception is that all consumer genetics companies draw on one master database. They do not. Each maintains a separate reference panel built from its own consented customers and curated public samples. There is no industry-wide shared genotype repository for consumer testing.
A second misconception treats "ethnicity estimate" as a measurement rather than a model output. The estimate is a statistical inference, and it is only as good as the reference panel behind it. Updating the panel changes the result for the same raw data, which is why companies periodically revise past reports.
A third misconception is that deleting your account deletes your data everywhere. It deletes it where you asked. Data you uploaded elsewhere, or that a research partner received under your consent, follows that party's retention rules. Read the deletion clause before you assume erasure is global.
The professional caveat worth internalizing: a consumer report is a screening-grade inference, not a clinical test. Confirmatory clinical genetic testing uses different methods and is ordered through a clinician. Treat any consumer health flag as a prompt for a conversation, not a diagnosis.
How the major model types compare
| Model type | What it shares | Typical consent default | Accuracy caveat |
|---|---|---|---|
| Single-company closed platform | Nothing outside the company | Matching usually opt-in | Limited by its own reference panel |
| Raw-data upload service | Accepts files from other labs | Upload is an explicit act | Reinterprets unverified genotypes |
| Partner matching network | Matches across linked databases | Specific toggle, often off | Coverage depends on partner size |
| Research consortium | De-identified data to researchers | Governed by consent form | Retention follows research rules |
| Clinical genetics lab | Report to ordering clinician | Clinician-mediated | Higher analytical validation |
A plain-language reading of the table: the only reliable way results "share" is if you carry them or a contract carries them.
Timeline: what happens to your sample and data
| Stage | Timeframe | What occurs |
|---|---|---|
| Kit registration | Day 0 | Account created; consent terms accepted |
| Sample receipt | Days 3 to 14 | Lab logs and barcodes the sample |
| Genotyping | Days 5 to 21 | Chip reads the SNP panel |
| Analysis and reporting | Days 10 to 30 | Reference panel and algorithms generate reports |
| Raw file availability | Around report release | Download usually becomes available |
| Matching ongoing | Continuous | New matches appear as others test |
| Panel revision | Periodic, company-set | Prior estimates may be recalculated |
Timelines vary by lab volume and are not a guarantee. The structural point is that your data persists after the report, and matching is a live process, not a one-time calculation.
Practical actions before you share or upload
Start by downloading your raw data file and storing it yourself. That file is your portable asset, and having it means you are never locked into one company's interpretation.
Decide your matching posture deliberately. Most platforms default matching to off or limited; turn it on only if you want relative discovery, and understand that close matches can surface unexpected family information.
Before uploading to a second company, check three things: whether the receiving service accepts raw uploads for the report you want, whether it states its accuracy limits for uploaded versus lab-processed samples, and what its deletion policy says. A threshold worth applying: if a service will not state how it handles uploaded files, do not upload.
Schedule a re-check of your reports after any company announces a reference panel update, because your percentages may change without your genome changing.
If a health-related flag appears, do not act on it alone. Bring the specific variant and report to a clinician or genetic counselor, who can order confirmatory testing. This is the boundary between consumer screening and clinical care.
Limits of what can be concluded
The evidence base for cross-company concordance is thinner than the marketing implies. Independent peer-reviewed comparisons exist but are few, use small samples, and age quickly as panels change. Companies rarely publish head-to-head validation against competitors, so the honest answer is that broad ancestry tends to converge while fine-grained ancestry and health interpretations often do not.
Several things cannot be concluded from a consumer report. You cannot infer a clinical diagnosis. You cannot conclude that a missing relative match means no relationship. You cannot assume one company's risk estimate transfers to another's scale. And you cannot assume that because two companies read the same raw file, they will report the same finding.
The mechanism is stable even as products change: separate chips, separate panels, separate algorithms, separate consent contracts. Results move only when you move them or a contract moves them, and even then they arrive as a new interpretation of an old file. Treat the raw data as yours to steward and the interpretation as provisional, and the apparent contradictions between companies stop being mysterious.