Skip to content
Novandri.

AI Legal Research: a Decision Framework

Legal AI can find and summarise authority, not decide what it means. Read a framework for use, a worked check, and where the human must stay.

Rachmadwipa Novandri4 min read

Legal research has two tasks that are often conflated: finding the relevant authority and deciding what it means for a client. AI is good at the first when the sources are bounded and citable, and unreliable at the second because interpretation depends on facts, jurisdiction, and judgement, as AI in Legal Research: Retrieval, Summarization, Citation Checks explains for citation checks, as AI for Contract Review: What to Automate and What to Escalate treats material clauses as a human decision. A framework should keep the two separate.

Direct answer:

Use AI for retrieval, summarisation, and citation checking under a defined scope, with a lawyer responsible for interpretation and a record of the sources used.

Anthropic's use-case guides cover summarisation patterns, and AAPOR's standards apply to any survey used in the research. The boundary: this is educational orientation, not legal advice — the law and the rules of professional conduct are jurisdiction-specific and change, as the copyright and AI Act articles treat current law as requiring a dated check.

A flow: define scope, retrieve and summarise, check citations, then lawyer interprets.
Conceptual: scope, sources, responsibility. A wrong citation is worse than no summary.

The framework

Scope the sources. Define the jurisdiction, the date range, and the types of authority the system may use, which the market-definition article analogises to defining the relevant set, as the data-governance article requires provenance for relied-upon records.

Ground every claim. Each summary point should cite a passage that supports it, which the legal-research article treats as the core control, as the hallucination article warns about invented citations, as the reduce-hallucinations article treats citation checks as a discipline.

Separate finding from interpreting. A tool can locate a statute and summarise it; a person decides how it applies, which the contract-review article treats as the boundary, as the human-in-the-loop article argues for a meaningful review.

Record the work. Keep the queries, the sources retrieved, and the review, which the decision-memo article requires for a recorded judgement, as the data-governance article requires for provenance.

Three consequences follow. The system's value is time saved on search, not authority, as AI in Legal Research: Retrieval, Summarization, Citation Checks explains.

And the scope should be reviewed when the matter changes, as the monitoring article treats as a trigger.

A worked check

The following is hypothetical and illustrative. A lawyer asks a tool to find authority on notice periods for a contract type in a jurisdiction.

Three checks decide whether the output is usable. First, citation existence: does the cited section exist, which the legal-research article requires? Second, support: does the passage actually say what the summary claims, as the reduce-hallucinations article treats as a separate check from existence? Third, completeness: was the controlling authority retrieved at all, which the reranking article treats as a recall problem, as AI in Legal Research: Retrieval, Summarization, Citation Checks warns that a clean summary of the wrong source is worse than none?

Three conclusions follow. The lawyer verifies every citation, as the oversight article requires for a meaningful review. The tool's output is a draft, not an opinion, as AI for Contract Review: What to Automate and What to Escalate argues for human judgement on material clauses. And the queries and sources should be recorded, as the decision-memo article requires.

What the framework cannot do

It cannot give legal advice, which the legal context requires a qualified person, as the AI Act article treats role-specific obligations, as the governance checklist requires an accountable owner. It cannot guarantee completeness, which the reranking article treats as a recall limit, as AI in Legal Research: Retrieval, Summarization, Citation Checks warns about missing authority. And it does not remove professional duties, which the current rules define, as AI in Legal Research: Retrieval, Summarization, Citation Checks requires a dated check.

The counterargument that the tool is faster than manual research is fair and incomplete, which the worked check shows: speed is useful only when the citations hold, as the evaluation-rubric article argues for outcome measures.

A decision test

  1. Define jurisdiction, dates, and source types, as the scope rule advises.
  2. Require a citation for every claim, as AI in Legal Research: Retrieval, Summarization, Citation Checks requires.
  3. Check existence and support separately, as the hallucination article advises.
  4. Keep interpretation with a qualified person, as AI for Contract Review: What to Automate and What to Escalate argues.
  5. Record queries, sources, and review, as the decision-memo article requires.
  6. Check current professional rules with a dated source, as AI in Legal Research: Retrieval, Summarization, Citation Checks advises.
  7. Review the scope when the matter changes, as the monitoring article recommends.

For the task pattern, see AI in Legal Research; for the clause boundary, see AI for Contract Review.

References & Further Reading

  • Anthropic, 2026. Guides to common use cases - Claude Platform Docs — Production use-case guides for ticket routing, customer support agents, content moderation, legal summarization and commerce agents.
  • American Association for Public Opinion Research, 2026. AAPOR — The professional association for public opinion and survey research; its standards and resources underpin the survey-design discussion.