"Research EU data protection" isn't a research question. "What are the lawful bases for processing under Article 6 GDPR, and how have EU courts read legitimate interest for large-scale profiling?" is. The gap between those two sentences is most of the work, and it's the part that decides whether the answer is any good.

The rest is discipline that lawyers already have and most AI research doesn't. Know which law applies before you look anything up. Start with commentary and end with the statute. Never let a persuasive case pass for a binding one. And check the citation against the source rather than against a summary of it.

Settle the jurisdiction first

A rule from the wrong jurisdiction still reads like an answer, which is what makes it dangerous. So nothing gets searched until that question is closed.

  • Name it, and your companion gets on with it. If the facts point somewhere else, it says so rather than going quiet.
  • Say nothing, and it asks before searching.
  • Genuinely unsure, and it sets out the plausible regimes and the factors that decide between them: where the parties sit, where the contract was formed, where the harm happened. That is a starting point for you or a qualified reviewer to confirm, not a ruling on which law governs.
  • Ask something general with no facts attached, and it says plainly that the answer is illustrative and shouldn't be relied on for any specific position.

What it won't do is blend two systems into one tidy answer.

Want to try it?

Ask your Strawberry companion: “Research this legal question in my jurisdiction. Start with secondary sources, then primary law, separate binding from persuasive authority, and flag anything you could not verify.”

Skill

Research a legal question

Research this legal question in my jurisdiction.

Commentary first, then the law

There's a reason this is taught that way. Treatises, practice guides, and firm analysis tell you what the landscape looks like and which authorities actually control it. Go straight to case law instead and you'll find something that says what you hoped, and miss the thing that governs.

Strawberry can work through the databases and subscriptions you already pay for, then follow a citation out of the commentary and into the primary source. That second step is the one people skip, and it's where the difference shows.

Everything comes back labeled with three things: which jurisdiction it's from, whether it's primary or secondary, and whether it binds you or merely persuades. A persuasive authority from another system is often worth having on a novel question. It just has to arrive marked as one rather than sitting in the list looking like law.

Knowing when to stop

Legal research has no natural end. There's always another case, another commentary, another jurisdiction that might be interesting. Knowing when you're done is as much a skill as knowing where to start.

The signal is convergence. Independent sources start pointing at the same authorities, new searches return what you've already read, and the controlling statute and leading cases are identified. More work would add depth without changing the answer.

Where authorities genuinely conflict, you get told that rather than getting one of them picked for you.

What arrives, and what you check

You get the answer first and the authority it rests on, with each proposition sitting next to the source that supports it. Then the uncertainty: what's contested, what's thin, what couldn't be verified.

It also tells you what was searched and what wasn't. A named gap is useful. A confident answer hiding one is the thing that gets people in trouble.

Then you verify, because that part is yours. Every source has to exist, say what it's cited for, carry a correct pinpoint, still be good law, and belong to the jurisdiction on its label. Anything that fails is pulled or flagged before it reaches you, so what you're doing is checking rather than salvaging.

Once your jurisdictions, sources, citation system, and preferred output are settled, they can be saved so the next question starts there.

How to compare laws across jurisdictions with AIWhen one question turns into five because you operate in five places. How to research a company with AIFor when the question turns on the counterparty rather than the law.
Official Strawberry skill
Copy skill file Download skill file
name
research-a-legal-question
description
Research a legal question in a named jurisdiction using secondary sources then primary law, and return a source-linked answer that separates binding from persuasive authority and flags what could not be verified. Use for statutory questions, case research, regulatory scope, and background before drafting or advising.

Research a Legal Question

Produce the first draft and the source assembly for a legal question. A qualified human verifies before anything is relied on, and this skill should make that verification fast rather than assume it away.

1. Fix the jurisdiction before anything else

Legal research is jurisdiction-specific, and a rule from the wrong jurisdiction is worse than no answer. Establish which law applies before searching:

  • The user names a jurisdiction: proceed, but flag it if the facts suggest another country's law may actually apply.
  • The user does not name one: ask before researching.
  • The user does not know: identify the plausible regimes and the factors that would decide between them, such as where the parties are, where the contract was formed, and where the harm occurred. Research those regimes as far as is useful, but present the result as a preliminary view rather than a determination. The user or a qualified legal reviewer confirms the governing jurisdiction before anything here is relied on.
  • The question is general and the facts are too thin to determine applicable law: say so explicitly in the output, generalise with significant caveats, and recommend the user identify the jurisdiction before relying on anything.

Never blend rules from different jurisdictions into one answer.

2. State the question precisely, then plan

"Research EU data protection" is not a research question. "What are the lawful bases for processing under Article 6 GDPR, and how have EU courts interpreted legitimate interest for large-scale profiling?" is. Restate the question, identify the legal issues inside it, and agree the depth and the output the user needs before running searches.

3. Protect the matter before searching

Search queries leave the session. Never put client or party names, deal terms, financial figures, or privileged facts into a web search. Use anonymized or hypothetical formulations instead: search "limitation of liability clause SaaS agreement", not the parties and their terms.

If the user pastes material containing identifiable confidential information without having confirmed their confidentiality posture, stop and flag it before processing — even mid-workflow. Do not carry confidential material into memory, custom skills, or anything else that persists into later sessions unless the user explicitly asks for that.

4. Secondary sources first, then primary law

Start with commentary — treatises, practice guides, law-firm analysis, encyclopaedias — to understand the landscape and find the controlling authorities. Then go to the primary law itself: statutes, regulations, case law.

Strawberry can work through the databases and subscriptions the user is already entitled to use and follow citations into the underlying sources rather than stopping at search summaries.

For every authority, record three things: the jurisdiction, whether it is primary or secondary, and whether it is binding or persuasive for this user's jurisdiction. Label persuasive authority as persuasive in the output. Do not present a case from one jurisdiction as though it controls in another.

5. Stop when the research converges

Legal research has no natural endpoint. Stop when independent sources point to the same primary authorities, new searches return what you have already found, the controlling statute and leading cases are identified, and further work would add depth without changing the answer. Say so when authorities genuinely conflict rather than picking one.

6. Verify before returning anything

This is not optional, and it is the step that makes the rest usable. For every citation:

  • Confirm the source exists, in an authoritative database or the original.
  • Confirm it supports the proposition it is cited for, by reading the passage rather than trusting a summary.
  • Check pinpoints. Treat page references as unverified until checked; article and section references are structural and more reliable, but still verify them.
  • Check the law is current — statutes amended, cases overruled, distinguished, or questioned.
  • Check the jurisdictional attribution is right.

Anything that cannot be verified is removed or explicitly flagged as unverified. Never leave an unchecked citation sitting quietly in the output.

7. Return something a lawyer can check

Lead with the answer and the authority it rests on. Keep each proposition next to the source that supports it, separate binding from persuasive, and state plainly what remains uncertain, contested, or unverified.

Say what was searched and what was not. A named gap is useful; a confident answer with a hidden gap is dangerous.

Use strawberry/legal/compare-across-jurisdictions when the question spans several jurisdictions, and strawberry/legal/brief-a-regulation when the result should be a briefing on a specific instrument rather than an answer to a question.

After the work proves useful, preserve the accepted jurisdictions, source hierarchy, citation system, and output format as a custom skill.