Why Western Platforms' China Coverage Isn't Built for AI

If you run data or engineering for a legal AI product, you have probably already had this conversation: "We license LexisNexis. We have vLex. Don't we already have China covered?" It is a reasonable assumption, and it is wrong in a specific, expensive way. The major Western legal research platforms do carry some Chinese legal content—but it was assembled for a human lawyer reading a translated statute, not for a retrieval system grounding answers over millions of primary judgments. For an AI product, those are not the same asset, and the gap between them is where your China feature quietly breaks.

This piece is not a knock on Lexis or vLex. They built exactly what their core customer—a practicing lawyer—needed. The point is narrower and more useful: the format that serves human research is the wrong format for machine retrieval at scale, and understanding why tells you what to source instead.

What the major platforms actually list for China

Start with the public, checkable facts as of mid-2026. They are unambiguous about scale.

Now set those numbers against the source. China Judgements Online, the Supreme People's Court's public repository, had accumulated roughly 143 million effective judgments by December 2023, according to figures cited in SPC press briefings—the largest public body of court decisions in the world. A curated selection of a few thousand translated documents is a fine starting point for a partner skimming Chinese contract law in English. As the grounding layer for an AI product answering questions about Chinese precedent, it covers a fraction of a fraction of one percent of the case law.

That is not an editorial failure. It is a category difference: a reading list versus a corpus.

Three reasons human-research coverage doesn't transfer to AI

The mismatch is not just about how many documents exist. Even where Western platforms hold Chinese content, three structural properties make it unsuitable as an AI data layer.

1. Volume: you cannot ground a model on a reading list

Retrieval-augmented generation only works if the answer is somewhere in the index. A user asking about, say, how PRC courts have treated a particular kind of employment dispute, or the range of damages in a category of IP infringement, needs the system to retrieve from thousands of relevant on-point judgments—not from a hand-picked set of a few landmark translations. Fine-tuning and evaluation are worse off still: a few thousand documents cannot represent the distribution of how Chinese courts actually reason across causes of action, court levels, and years. Curated selections are, by design, the opposite of representative.

2. Format: editorial products resist bulk, structured extraction

Human-facing platforms optimize for the reading experience—headnotes, editorial summaries, cross-links, a search box. What an AI pipeline needs is the inverse: every document delivered in bulk, machine-readable, with the same stable structured fields on every record—case number, court, date, cause of action, parties, outcome. Those fields are what let you dedupe, filter by vintage, build citations a user can verify, and run extraction at scale. Scraping them back out of a portal designed for human browsing is brittle, incomplete, and usually prohibited by the terms anyway, which brings us to the third point.

3. Access: a research subscription is not an AI data license

This is the one that surfaces in legal review rather than engineering. A per-seat research subscription licenses a human to read. It almost never grants the right to ingest content into a model, build a derived index, or redistribute it inside a product. Training, embedding, and retrieval over licensed content are exactly the uses these agreements are written to exclude. So even where the documents technically exist on a platform you already pay for, using them the way an AI product needs to is a different deal—and one those platforms are not structured to sell for mainland Chinese primary law.

Why China specifically falls through the cracks

Western platforms have wired in primary law for many jurisdictions, often by integrating official open sources. In July 2025, for example, Harvey announced integrations with official free legal sources across eight countries—Austria's RIS, Finland's Finlex, and India's Indian Kanoon among them. That model works when a jurisdiction publishes an open, comprehensive, bulk-accessible source you can connect to.

Mainland China does not fit that model, for reasons we have documented in detail. Public availability peaked in 2020 at 19.2 million judgments published in a single year, fell to about 5.11 million in 2023, then partially recovered. In 2024 the system split into a comprehensive internal database—searchable only on the courts' own network—and a small public library of curated reference cases. There is no open, comprehensive, machine-friendly bulk feed for an integration team to simply wire in. The largest case law corpus in the world is also one of the hardest to assemble into AI-ready form, which is precisely why it remains a gap on platforms that cover dozens of easier jurisdictions.

The result: mainland China is the single largest body of case law still largely absent from Western research stacks—not because the documents do not exist, but because turning them into a structured, licensable, AI-ready corpus is specialist work.

What "AI-ready" actually means for a China corpus

If the existing platforms are the wrong shape, what is the right one? When you evaluate a Chinese case law data source for an AI product, hold it to these tests rather than to a human-research checklist:

The verifiability row matters more than it looks. Because public inflow collapsed in 2021–2023—to roughly 78%, 54%, and 27% of the 2020 peak respectively—any corpus is most likely to be thin in exactly that window. An AI-ready source should hand you a coverage breakdown so you can check the depth of that vintage before you build on it, rather than discovering the hole as missing precedent in production.

The practical takeaway for legal AI teams

If China is on your roadmap, the move is not to stretch a research subscription to do a job it was never licensed or formatted for. It is to source primary Chinese case law the way the leading platforms source the jurisdictions they cover well—through a dedicated licensing arrangement, but one built from the start for machine consumption: bulk plus API delivery, structured fields, explicit AI-use rights, and a verifiable coverage report.

That is the category SinoVerdict occupies. We license a structured, machine-readable corpus of more than 130 million Chinese court judgments, with coverage through 2023 and incremental updates on request, delivered as a bulk dataset, a REST API, and an MCP server—with a coverage breakdown you can verify before you commit. For the deal mechanics and evaluation checklist, see our guide to licensing Chinese court judgment data; for how the documents are structured for retrieval and extraction, see our walkthrough of the case law API and document fields.