Dataset guides
How to read a listing, choose the right dataset and use it within its license. Written around the LAW.co datasets, but the same reading applies to every provider.
Which dataset for which job
| If you need… | Use | Watch out for |
|---|---|---|
| Opinion text to ground answers | U.S. Court Opinions | Not official text; uneven coverage; duplicates |
| Quick triage of what a case is about | LAW.co AI Case Summaries | AI-generated; never cite as authority |
| Which cases cite which | LAW.co Case Citation Graph | Unresolved citations; per-case caps upstream |
| Estate-planning thresholds by state | Statute Reference Values | Small curated table; contact-only |
| Statutes or regulations | Not offered | See the statutes and regulations pages |
Four things to check before you build
Access status: “Available” is the only status you can build against today. “Not yet connected” and “awaiting rights confirmation” mean the method is planned but blocked, and the listing says why.
Redistribution: public, paid, restricted or unknown. This decides whether you can show records to your own users and whether bulk export is possible.
Record count and dates: when a listing says “Not yet verified”, treat coverage as unknown rather than complete.
Derived fields: anything marked derived is AI or editorial output with its own license. Keep it visually and logically separate from source text in your product.
What each provenance field tells you
Every record carries a provenance block. These are the fields worth checking before you rely on a record.
| Field | Read it as |
|---|---|
| provider, sourceId | Who served the record and their own id for it — what you cite and re-fetch by. |
| upstreamSource | Where the provider got it, such as CAP or CourtListener. Null means the provider does not say. |
| canonicalUrl, upstreamUrl | Where to verify the record against a public original. |
| retrievedAt, providerRetrievedAt | How current your copy is, and how current the provider's copy was. |
| licenseId, accessLevel | Which terms apply to this record and how it may be accessed. |
| isDerived, generator | Whether this is source or AI/editorial output, and what produced it. |
A short fitness check before you build
- 1
Define the question
Write down what your product needs to answer and which jurisdictions, courts and dates matter.
- 2
Read the listing
Check coverage, date range, quality notes and each access channel's status. Unverified values mean unknown, not complete.
- 3
Test known items
Search the preview for cases you already know should be there. Gaps you find are more informative than any headline count.
- 4
Check the license
Confirm the redistribution value and restrictions fit your use, especially display to end users, storage, bulk use and model training.
- 5
Plan for limits
Design around documented limits, such as sampled search, missing citation resolution or duplicate records, rather than discovering them in production.
Using legal data responsibly
- Attribute the provider and upstream source
- Label AI summaries as AI-generated
- Don't bulk-copy where the license forbids it
- Don't train models without written permission
- Link to the canonical source
- Tell users the data is not legal advice
Keeping AI enrichment apart from source text
LAW.co's records show why layers matter. The opinion text is the court's own language, gathered from public sources, and can be quoted and cited. The summary, holding, key issues and topics next to it are generated by LAW.co's models, published only after LAW.co's own review gate, and owned by LAW.co. They are useful for triage and ranking, but they are not authority, and they are licensed differently.
In your product, render the two in visibly different places, label the enrichment as AI-generated, and never splice a summary sentence into a quotation. In a retrieval pipeline, index the source text for answers and use enrichment only to rank or filter. In storage, keep the enrichment's own provenance — its generator and license id — next to it, so you can remove or refresh it independently of the source.
Questions buyers ask
Is the LAW.co opinion text the official version?
No. It is opinion text gathered from public sources such as CAP and CourtListener. For filings, verify against the official reporter or the court.
Does Data.co offer statutes or regulations?
No. LAW.co holds no statute corpus and no regulations. Its only statute data is a small curated table of estate-planning reference values, listed as contact-only.
How do I know whether a record is demo data?
Every record has a demo flag, and responses set meta.demo and the X-Dataco-Demo header when any record is synthetic. Live LAW.co records return meta.demo false; the previews on this site use demo records.
Can I show AI summaries to my users?
Where your plan is entitled to them and the license permits display, yes, labeled as AI-generated and kept apart from the opinion text.
Which dataset should I start with?
Usually U.S. Court Opinions, because its text and facts are public-domain material and it is the standard tier. Add summaries or the citation graph once you know they help your users.
Questions about a dataset?
Ask us about coverage, licensing or fitness for your use case.