Methodology
How TomLegal works
TomLegal combines retrieval-augmented generation with real-time access to legal databases. Every response is grounded in verifiable sources — not generated from memory or training data alone.
Process
From question to cited answer
Every query passes through a four-stage pipeline designed for legal accuracy and traceability.
Classify
Understanding your legal question
When you submit a query, TomLegal first classifies the legal domain, jurisdiction, and intent. Whether you're asking about a Delaware non-compete clause or a UK employment tribunal procedure, the system identifies the precise area of law before any research begins.
- Jurisdiction detection — federal, state, international
- Practice area classification — contracts, litigation, corporate, employment, IP, regulatory
- Intent analysis — drafting, research, review, comparison
Research
Querying authoritative legal sources
TomLegal searches across multiple legal databases simultaneously, retrieving case law, statutes, regulations, and secondary sources relevant to your query. Each source is fetched in real time — not from a static training snapshot.
- Parallel search across case law, statutes, and regulatory databases
- Real-time retrieval — not cached or stale data
- Relevance ranking tuned for legal citation patterns
Synthesise
Constructing a grounded response
The AI synthesises retrieved sources into a coherent analysis. Every factual claim is anchored to a specific source. The model is instructed to distinguish between settled law, majority positions, and minority or evolving interpretations.
- Source-grounded generation — claims tied to retrieved documents
- Distinguishes binding authority from persuasive authority
- Flags areas of legal uncertainty or jurisdictional variation
Cite
Verifiable references you can check
Every response includes inline citations linking to the original source material. Case names, statute sections, and regulatory references are formatted for professional use. You can follow each citation to verify the underlying authority.
- Inline citations with direct source links
- Standard legal citation formatting
- One-click verification of every reference
Sources
Where the data comes from
TomLegal retrieves from authoritative legal databases in real time. Sources span case law, legislation, regulatory material, and secondary authority across multiple jurisdictions.
Case Law
- Federal court opinions — all circuits, district courts, bankruptcy courts
- State court opinions — appellate and supreme courts across all 50 states
- UK courts — Supreme Court, Court of Appeal, High Court
- Australian courts — High Court, Federal Court, state supreme courts
Statutes & Legislation
- United States Code (USC) — current and historical versions
- Code of Federal Regulations (CFR)
- State statutes and codes — all 50 states
- UK Acts of Parliament and Statutory Instruments
- Australian Commonwealth and state legislation
Regulatory & Administrative
- Federal Register notices and proposed rules
- Agency guidance documents and interpretive letters
- SEC filings, FTC actions, DOJ opinions
- State regulatory agency publications
Secondary Sources
- Legal treatises and practice guides
- Law review articles and academic commentary
- Bar association publications and practice advisories
- Court rules, local rules, and procedural guides
Source coverage varies by jurisdiction and practice area. TomLegal continuously expands its database integrations. If a source is not available for your jurisdiction, the system will disclose this rather than substitute with inapplicable authority.
Accuracy & Trust
Safeguards against hallucination
General-purpose AI models can fabricate citations, misstate holdings, or apply the wrong jurisdiction's law. TomLegal is built with specific mitigations for each of these failure modes.
Multi-source verification
Claims are cross-referenced against multiple independent sources. If a proposition appears in only one source and cannot be corroborated, it is flagged as potentially unreliable.
Citation existence checking
Before a case or statute is cited, the system verifies that the reference exists in the retrieved source material. Fabricated citations — a known risk in general-purpose AI — are caught at this stage.
Jurisdictional scoping
Responses are scoped to the jurisdiction relevant to your query. The system will not silently apply California law to a New York question, or US precedent to a UK matter, without explicit disclosure.
Recency awareness
Legal sources are retrieved in real time, not from a frozen training set. The system identifies when a statute has been amended, a case overruled, or a regulation superseded — and tells you.
Uncertainty disclosure
When the law is unsettled, the system says so. Split circuits, pending legislation, and areas without binding authority are identified rather than papered over with false confidence.
Model-level guardrails
The underlying language model is constrained by a legal-specific system prompt that prohibits speculation, mandates source attribution, and requires explicit caveats where the law is ambiguous or jurisdiction-dependent.
Commitment
AI should earn your trust, not demand it
TomLegal is a research tool — not a substitute for professional legal judgement. Every response is designed to be verifiable. If you find an inaccuracy, we want to know about it.