How it works

How Verosynthea works

Census-grade Australian population data, privacy-safe by design. This page explains what AUSynth is, how it's built, and how the data and the reports differ.

Verosynthea AUSynth is the Australian product in the Verosynthea family. AUSynth is what you use; Verosynthea is the company behind it.

15,343
Suburbs
27.5M
Synthetic persons
47
Census variables

What is synthetic population data?

AUSynth generates privacy-safe individual records that preserve the statistical relationships of the real ABS Census 2021. No row corresponds to a real Australian. The records are fully shareable, modellable, and joinable to your own data, because there is no confidential microdata inside them to protect.

That solves a practical gap. Census microdata is the gold standard for understanding the population, but it is expensive to access, slow to deliver, and bound by confidentiality rules that make routine data work, like training a model or sharing a demo, difficult. Synthetic records carry the patterns without the constraints.

Methodology

The synthetic population is built on ABS Census 2021 aggregate tables, with Bayesian reconstruction (Gibbs sampling across Census conditional distributions) to recover individual-level records that preserve cross-tabular relationships. Figures are adjusted to 2025-26 using published population, wage, and price indices. Generated records pass a suite of pairwise validation rules that remove impossible combinations.

The pre-built statistical reports work differently. They are computed directly from the ABS Census conditionals (adjusted to 2025-26), not from the synthetic data. That separation is deliberate: real Census numbers for the ready-made analyses, synthetic individuals when you need records to build with. Each report states its own data basis.

Quality metrics are published openly. See Quality benchmarks for SRMSE scores per dataset, known limitations, and how small suburbs are handled.

Validation

AUSynth is validated against the real ABS Census 2021 across all 15,343 suburbs. The trust is in the numbers, not a name on a page.

Median SRMSE
0.05
person-level, synthetic vs real Census
Silhouette
0.125
profile clustering, 8 segments
Coverage
47 × 27.5M
variables × synthetic individuals

Full validation metrics, per-dataset fidelity, and known limitations are in the quality benchmarks, and the full method is documented in the methodology.

Verosynthea and AUSynth

Verosynthea is the company. AUSynth is the Australian product, specific to ABS Census data. The same methodology is designed to extend to other countries over time. You buy and use AUSynth; it is made by Verosynthea.

Privacy and licensing

  • No AUSynth record represents a real individual. The synthetic data contains no confidential microdata.
  • Customer query data is not sold or shared. Aggregated usage statistics never identify any user.
  • You pay for what you download. Credits never expire, and outputs are yours to use commercially.

FAQ

Is synthetic data really privacy-safe?

The records are generated from Census aggregates, not from any individual's data, so there is no real person to re-identify. Each row is a statistically plausible individual, not a real one.

How current is the data?

It is built on ABS Census 2021 and adjusted to 2025-26 using published population, wage, and price indices. The data version is shown in the footer and in every report.

Can I use this for commercial purposes?

Yes. Outputs you download or generate are yours to use commercially, including in products, dashboards, and client work.

How is this different from the real Census?

The reports are computed directly from real Census conditionals, so their numbers track the published Census. The synthetic dataset is a separate, privacy-safe reconstruction of individual records for when you need row-level data to model or join.

Can I cite this in academic papers?

Yes. Cite as “Verosynthea AUSynth v1.0 (2026), verosynthea.com”, and see the citation guidelines.

What is the methodology in one sentence?

Bayesian reconstruction (Gibbs sampling) over ABS Census 2021 conditional distributions, validated against pairwise consistency rules and adjusted to 2025-26.

Try it

New accounts get 5 free credits each week. No card required.