Examples
Worked examples
- Is an instance
Generating a synthetic version of a sensitive administrative dataset (e.g. patient records) using a differentially-private generative model, so researchers can develop and test analysis code without accessing real identifiable records
- Is an instance
Simulating data from a known statistical distribution to stress-test a new analysis method's behaviour under conditions a real dataset may not exhibit
Counter-examples
Looks similar, but isn't
- Not an instance
A real dataset with identifying fields removed or masked is de-identified data, not synthetic data -- the underlying individual-level records are still real
- Not an instance
An AI-generated image inserted into a paper's results section as if it were a real experimental image is a synthetic-image integrity problem, not a legitimate synthetic-data use
Editorial commentary
Synthetic data has a genuine, well-established methodological role, distinct from the integrity concerns that arise when synthetic content is passed off as a real observation. Its core legitimate use is privacy-preserving research: generating a dataset that reproduces the statistical structure (distributions, correlations, missingness patterns) of a real, sensitive dataset closely enough to be useful for method development, teaching, or preliminary analysis, without exposing any real individual’s record. Differentially-private generative models are the most rigorous current approach, since they carry a formal, quantifiable privacy guarantee rather than just an informal claim that no record ‘looks like’ a real one.
Where it can go wrong
Two failure modes matter for a research office. First, poorly-generated synthetic data can still leak information about real training records through memorisation, particularly with smaller source datasets or overfit generative models — this is a variant of the same mechanism covered in data leakage (training), and is why a formal privacy guarantee (not just visual dissimilarity from the source) is the standard to look for before treating synthetic data as safely shareable. Second, synthetic data presented, without disclosure, as though it were real observational data is a fabrication issue — the synthetic-vs-real status of any dataset used in published research needs to be stated plainly.
How this differs from related AI-band terms
- vs. synthetic image: the same real/fabricated distinction applies, but the risk profile is different in practice — synthetic tabular/structured data has a mainstream, disclosed methodological use case (privacy-preserving research), whereas an AI-generated image standing in for a real experimental image in a publication is overwhelmingly an integrity concern rather than a legitimate research method.
- vs. training data provenance: provenance concerns tracking where an AI model’s own training data came from; synthetic data is itself a category of data (which may or may not be used as training data for another model) rather than a record-keeping practice.
Also known as
Generated data · Simulated data
Machine-readable encodings
Use in your systems
<role vocab="credit"
vocab-identifier="https://casrai.org/dictionary/"
vocab-term="Synthetic data"
vocab-term-identifier="https://casrai.org/dictionary/term/synthetic-data" />{
"@context": "https://schema.org",
"@type": "DefinedTerm",
"@id": "https://casrai.org/dictionary/term/synthetic-data",
"name": "Synthetic data",
"identifier": "https://casrai.org/dictionary/term/synthetic-data",
"description": "Data generated artificially -- via simulation, statistical modelling, or a generative AI system -- to mimic the statistical properties of real data without any of its records corresponding to a real individual observation, used for privacy-preserving research, method testing, training-set augmentation, or teaching.",
"inDefinedTermSet": "https://casrai.org/dictionary/domain/genai-disclosure#set",
"url": "https://casrai.org/dictionary/term/synthetic-data",
"sameAs": [
"Generated data",
"Simulated data"
],
"license": "https://creativecommons.org/licenses/by/4.0/",
"publisher": {
"@id": "https://casrai.org/#organization"
},
"author": {
"@id": "https://casrai.org/#editorial-team"
},
"datePublished": "2026-05-21T01:55:05",
"dateModified": "2026-08-22T14:51:54",
"inLanguage": "en-GB",
"isAccessibleForFree": true
}
