How to do Legal Research in Six Steps

Most legal questions begin the same way, with someone staring at a set of facts and trying to find the law that governs them. In one recent year, California enacted close to 1,000 new laws, the federal government added hundreds more, and courts published thousands of fresh opinions on top of that. The body of authority grows faster than any person can track, which is why a repeatable method beats guessing.

Strong legal research is methodical. You define your exact legal question and the facts behind it, start with secondary sources to get your bearings, find the law that actually controls, keep your findings organized as you work, confirm that law is still valid, and check your work for gaps before you rely on it. Each step sets up the next, and the sequence is built to be repeated.

This article walks through that six-step process, and it can be applied to legal questions across practice areas or jurisdictions. It also shows how AI changes the way this work gets done in 2026, where AI genuinely speeds things up, and the verification habits that keep AI-assisted research from going wrong. By the end, you’ll have a method for AI-assisted legal research that you can put to work today.

What is Legal Research and Why Does it Matter?

Legal research is the work of finding the relevant, currently valid law that applies to a specific set of facts. It covers locating the controlling authority, confirming that authority still stands, and assembling it into a record you can rely on. Finding the law and applying it are two separate jobs, and this guide takes on the first. Everything here is about the disciplined search for the right authority, and the work of arguing it comes later.

So why does this matter more now than it used to? The volume of new law keeps climbing, so the odds of missing something relevant climb right along with it. AI raises the stakes further. It can surface authority in seconds, but it can also produce confident text with no real source behind it, which means a sloppy method now carries a faster and more public penalty. A reliable process is your defense against both the volume and the speed.

Legal authority sorts into three tiers. Primary sources are the binding law itself, such as constitutions, statutes, regulations, and judicial opinions. Secondary sources explain and analyze that law, including treatises, law review articles, and practice guides. Tertiary sources, such as indexes and legal dictionaries, mostly point you toward the other two. In day-to-day practice, lawyers spend most of their time in the first two tiers, using secondary sources to find their way to the primary law that controls.

How to do Legal Research in Six Steps

Think of these six steps as one continuous loop. Research is cyclical, so a gap you find at Step 6 often sends you right back to Step 1 to widen the question or rework your search terms. The list looks linear, but you'll move up and down it as the work tells you to.

Step 1: Define the issue and write your statement of facts

Vague questions produce vague research. Before you open a single database, pin down the precise legal issue and the outcome your client wants, whether that’s damages, an injunction, a clean closing, or a defense that holds. A sharp question tells you what law is relevant and, just as usefully, what isn’t. Spend the time here, because every later step inherits the precision or the fuzziness of this one.

Next, write a full statement of facts and lay the events out on a timeline. Capture who did what, when, and to whom, and don't filter as you go. Facts that look trivial now, a date, a job title, a passing email, often turn out to decide which rule applies. You want a complete record you can return to, since a tidy early summary tends to discard the very detail you'll need later.

From that record, build a working list of search terms. Pull the distinctive facts and the legal concepts in play, then add synonyms for each so a database doesn’t hide a key case behind different wording. A simple framework helps here, whether you use TARP or the five Ws to break the facts into searchable parts. Settle your jurisdiction at the same time, since the controlling answer shifts depending on where the question will be heard, and researching the wrong jurisdiction wastes the whole effort.

Step 2: Start with secondary sources

Don’t dive straight into the cases. Start with secondary sources instead, the materials that explain and organize the law before you ever read a judicial opinion. It feels backward to read commentary before you read the law itself, but it saves hours.

A few secondary sources do most of the heavy lifting. Legal encyclopedias and treatises give you the framework for an unfamiliar area, law review and journal articles go deep on narrow or unsettled questions, practice guides show how the law works in the real world, and legal dictionaries pin down terms of art. The reason this approach works is simple. A reputable secondary source has already run a research pass for you, so its footnotes amount to a curated map straight to the primary authority.

Use that map. When a treatise states a rule and cites the statute or the leading case behind it, follow the citation directly to that primary authority. You skip the slow work of building the framework yourself and land on the controlling law that an expert already identified. Mining those citations is the fastest way to move from a vague question to the cases that actually matter, and it spares you a blind search through the databases.

Step 3: Find and read the primary authority

Primary authority is the binding law itself, and it follows an order. Constitutions sit at the top, then statutes and codes, then the regulations agencies issue under them, and then the case law that interprets all three. Enacted law generally controls first, so you read the statute or regulation before you reach for the cases. Knowing this order keeps you from building an argument on a case when a statute already answers the question.

Annotated codes and regulations are your bridge from the enacted text to the cases. When you find the controlling statute, read its annotations and notes of decisions, which collect the opinions that have applied and interpreted that exact provision. That built-in list saves you a separate search and points you toward the cases courts in your jurisdiction already treat as relevant. Start there before you run a single case query.

When you reach the cases, read past the headnotes to the holding itself. Headnotes are an editor’s shorthand and a useful filter, but the holding is the law, and only the opinion itself tells you how the reasoning maps onto your facts. Once you find one strong, on-point case, use the “one good case” method. Run that case through a citator to surface everything that cites it, and you’ll quickly assemble the cluster of related authority around your issue.

You can run all of this on subscription platforms like Westlaw and LexisNexis, or on free options like CourtListener, Justia, GovInfo, and official .gov sites, depending on your budget and the depth you need. However you search, you have two modes. Terms and connectors give you precise control with Boolean operators, while natural language searching reads more like a plain question and casts a wider net. When one mode stalls and you keep hitting the same dead ends, switch modes. Rephrasing the same query a dozen times rarely breaks the logjam; changing how you search usually does.

Step 4: Organize your findings and take notes

Research you can't retrace is research you'll redo. The most common practical failure has nothing to do with bad searching. It's good searching that nobody wrote down, followed by a frantic hunt two days later for the case you know you read. Build your tracking system early, while the sources are still a trickle. The five minutes it takes to set up will save you hours.

Capture the full citation for every source the moment you use it. That means the case or code cite, the author and title for secondary sources, and the relevant dates, all recorded while the source is still in front of you. Skip this and you’ll spend real time later tracking down a citation twice, usually right under deadline pressure.

How you organize the authorities is your call. Some researchers use the folder tools built into Westlaw or Lexis to file sources by issue. Others keep a simple spreadsheet that tracks each issue, the search terms that found it, and which authority supports which point in the argument. Either approach works, so pick the one you’ll actually maintain when the matter gets busy.

This trail does more than keep you tidy. In an AI-assisted workflow, it becomes your verification record, the documented proof of what you checked and where, and the later sections of this guide lean on that record heavily. Build the habit now, while the stakes are just your own time.

Step 5: Confirm you have good law

A citation that looks perfect is worthless if the law behind it no longer stands. Relying on a case that’s been overruled, or a statute that’s since been amended, can be disastrous for the client and for the lawyer who missed it. So before any authority enters your work, confirm it’s still good law. This step isn’t optional, and it isn’t a formality.

Confirming good law means running each authority through a citator. A citator traces the later history of a case and tells you whether courts have reversed, distinguished, limited, or superseded it. Shepard’s and KeyCite are the established tools for this, and they flag the status of a case at a glance. Make this pass on every case you intend to cite, without exception.

A citator result still needs your judgment, though. A red flag doesn’t automatically kill a case, because the negative treatment might concern a point of law that has nothing to do with your issue. Read what the flag actually addresses before you discard a useful authority or lean on a wounded one. The signal tells you where to look, but you decide what it means for your facts.

Statutes and regulations need the same scrutiny. Confirm that the provision you're citing reflects the version currently in force, since amendments and repeals can quietly outdate a code section you pulled weeks ago. Bluebook citation form matters too, but that's a formatting question you can settle later. This step is about substance. What matters is whether the law is current and still binding; the shape of the cite can wait.

Step 6: Analyze, synthesize, and identify gaps

Step 5 is mechanical, a check on whether the law is good. This step is judgment. Here you step back and ask the harder question: whether your research actually answers the question you started with. Confirming that a case is valid tells you nothing about whether you’ve found enough of the right cases.

Run a gap check on every argument. Break each point into its required components and test whether you've supported all of them. If a legal test has five factors, your research has to reach all five, even the three nobody wants to chase down. Miss one and you've left a hole an opponent or a judge will find before you do.

When you find a gap, you go back. That's the loop in action, returning to Step 1 or Step 2 to amend your search terms, add a synonym you missed, or test a different jurisdiction. Research rarely runs in a straight line, and the willingness to circle back is what separates thorough work from work that merely looks finished. Treat each gap as a prompt to keep researching. Papering over a weak spot only hides it until it costs you.

You also need to know when to stop. The clearest signal is repetition, and when fresh searches keep surfacing the same authorities you’ve already logged, you’ve probably covered the field. Writing the memo or the brief is the next task, and it’s a different one, so resist the urge to start drafting before you’re sure the foundation holds.

How AI Changed the Legal Research Workflow

The six-step method hasn’t changed. What’s changed is how long each step takes. Work that used to fill an afternoon, like reading through secondary sources to get oriented, running Boolean searches and refining them, and pulling holdings from a stack of printed cases, now happens in a fraction of the time.

Look at where that time gets saved, step by step. In the orientation phase, Steps 1 and 2, AI surfaces the relevant legal framework and a set of candidate secondary sources in seconds. That shortens the path from a vague question to a usable issue statement and a working list of search terms, the part of the process that used to mean hours of reading just to know where to start.

In the finding phase, Step 3, grounded legal AI locates controlling statutes and on-point cases, summarizes their holdings, and runs the kind of one good case expansion you'd otherwise do by hand. What took methodical browsing through citators and digests now returns a first cluster of authority in moments. That leaves you to do the reading and the judging, while the system handles the hunting.

In the synthesis phase, Step 6, AI drafts a first version of how the authorities fit together and flags gaps in an argument, such as whether you’ve addressed every factor of a multifactor test. You’re no longer staring at a blank page. You’re editing a structured starting point, which is a far faster place to begin the analysis only you can finish.

Add these up and the payoff lands in terms your organization already cares about. You cover more matters, you turn work around faster, and you redirect a lawyer's hours away from mechanical retrieval and toward the analysis that only a lawyer can do. This is about capacity and reach. You take on more and go deeper, and the headcount stays exactly where it is.

One distinction decides whether any of this is safe to use. A general-purpose chatbot answers from its training data, and it can produce fluent, confident text with no real source behind it. AI built for legal work behaves differently. A platform like Harvey grounds each answer in actual source authority and shows you its citations and its reasoning, so you can trace every claim back to the law it came from.

A Practical Protocol for AI-Assisted Legal Research

AI doesn’t replace the six-step method. It compresses it, running each step faster, and it adds one layer you can’t skip. That layer is verification, and it’s the price of working at AI speed. Here’s a protocol you can utilize.

1. Use AI to orient and surface candidate authorities, and treat everything it returns as a rough draft you still have to prove out.

2. Verify every citation against a primary database before it enters any work product, with no exceptions for the citations that look obviously correct.

3. Confirm each authority is still good law by running it through a citator, exactly as you would in Step 5.

4. Document the verification, recording which databases you checked and on which dates, building directly on the tracking system from Step 4.

The record you build isn’t busywork. It’s the evidence that you did the job right, and courts have started treating it exactly that way.

None of this is hard to understand, and that’s the trap. Most lawyers already know they’re supposed to verify. What breaks down is doing it consistently at 11 p.m. with a filing due in the morning, when the AI output reads cleanly and the pull to just trust it is strongest. The protocol works only if you treat it as a rule you don’t break, which costs you time and discipline exactly when both are in short supply. Name that cost honestly, build the habit before the pressure hits, and the protocol holds when you need it most.

What Good Legal Research Looks Like in Practice

When you put it all together, the picture is simple. You define the issue and the facts, orient with secondary sources, find and read the primary authority, and organize everything as you go. Then you confirm the law is good and test whether your research truly answers the question, adding a verification layer whenever AI is in the mix. The whole thing runs as a loop, since a gap in the last step sends you straight back to the first. Good research comes from the judgment and discipline you bring to every pass, far more than from the database you happen to open.

Grounded, citation-linked AI research is moving from novelty to standard infrastructure, the way email and electronic filing once did. Speed stops being a differentiator the moment every team has it. What sets the strongest research apart is verifiability, the ability to prove that every answer traces back to real, current law. The lawyers who build that habit now stay ahead of the ones who adopt it only when they have no choice.

This is the case for Harvey, and it's a plain one. The method doesn't change, but a platform built for legal work runs each step faster and returns results already tied to their source. Harvey's Knowledge grounds every query in trusted data, connecting your work to legal databases, curated public sources, and the institutional knowledge your team already holds. It reaches more than 500 legal data sources worldwide, lets you ask LexisNexis about US case law, statutes, and regulations without leaving Harvey, and ties every answer to the citation behind it. The point isn't fewer lawyers, it's more capacity and more confidence, the room to take on harder work and the proof to stand behind it. Request a demo and watch grounded research run on a question from your own practice.