A research memo can read beautifully and still fail the one test that matters when a partner picks it up: they cannot see which source stands behind which sentence.
It usually starts with a question that keeps shifting as the facts sharpen. An associate searches, finds sources, discards some, revisits others, and verifies a few. Notes pile up across browser tabs, a research platform, saved PDFs, a working document, and a couple of emails. A clean first draft appears. Then the responsible lawyer reads it and has to ask the things the draft cannot answer on its own: where did this conclusion come from, was the governing-law assumption ever confirmed, and which of these points is still a hunch wearing the clothes of a holding.
If you run a practice group or a knowledge function, you have watched that gap swallow hours. This guide is about closing it. Ubisar does not do legal analysis, and the conclusion that reaches a client stays with your lawyers. What a workflow can do is keep the question, the sources, and the state of the review in one place, so a good memo is also one that anyone can inspect.
Pin the question before anyone opens a database
Most weak memos are not weak because the research was thin. They are weak because the question drifted. The partner asked one thing on Monday, the facts shifted on Wednesday, and by Friday the associate is answering a slightly different question than the one that will actually be relied on.
So the first thing worth writing down is not a source. It is the question, in the exact form it needs to be answered, together with the facts and assumptions it rests on. Which jurisdiction governs. Which facts are given and which are still being confirmed. What the client actually intends to do with the answer. When those are written before the searching starts, every source that gets pulled has something to attach to, and the reviewer can later check the reasoning against the question that was really asked rather than the one the draft quietly slid toward.
This is also the cheapest place to catch the most expensive mistake. A memo built on an unconfirmed governing-law assumption can be immaculately researched and still wrong. Naming the assumption up front, and flagging it as unconfirmed, keeps it visible until a lawyer resolves it.
Why the trail behind a finished memo goes cold
The finished memo is polished. The trail behind it is not. The question lived in an email. The sources lived in a research platform and a stack of PDFs. The working notes lived in a document, and the one caveat that actually mattered, the "wait, we should double-check this before we rely on it," lived in a chat message that never made it into the file. The draft looks done, but how it traveled from question to conclusion is invisible.
That is what makes review slow and, frankly, nervous. The reviewer has to reconstruct what was searched, which authority supports each point, which assumptions are still open, and where the draft is leaning further than the sources allow. Every memo quietly becomes an audit, and the audit lands on the busiest lawyer in the room. The associate loses time too, because "make the changes" comes back without a clear map of which sentences were even in question.
How research actually moves from question to memo
Before you can improve the workflow, it helps to say plainly how it runs today. In most teams the shape looks something like this, and the trouble is almost always in the handoffs, not the steps.
- A partner or senior lawyer frames the question, often verbally or in a short email, with the facts as currently understood.
- An associate restates the question, identifies the governing law, and starts searching approved research platforms and the firm's prior work.
- Sources are read, saved, quoted, and set aside. Some are approved authorities, some are secondary, some turn out to be off point or no longer good law.
- The associate drafts an issue outline, then prose, pulling quotes and citations from the saved sources.
- Citations get checked, or half-checked, against the primary sources they point to.
- The responsible lawyer reviews the draft, asks where a point came from, requests sharper analysis, and confirms the conclusion.
- The memo is finalized and, sometimes, filed somewhere a future associate will never find it.
Each step is sensible. The fragile part is what travels between them. The question that reached step four is not always the question from step one. The source that backed a sentence in step three is not always the source a reviewer can find in step six. When the memo and the sources separate this early, the review step at the end has to rebuild everything the drafting step already knew.
Where the work loses its citation trail
The breakpoints are predictable. They show up in slightly different forms from memo to memo, but the underlying pattern repeats.
1. The question is never fixed in writing
If the question only ever lived in conversation, nobody can later prove the memo answered it. The associate researches a moving target, and the reviewer cannot tell whether a thin section reflects thin authority or a question that quietly changed.
2. Sources are collected without status
A saved PDF and an approved authority look identical in a folder. So do a case that is still good law and one that has been distinguished into irrelevance. Without a status attached to each source, the draft can cite something the firm would not actually stand behind, and no one notices until it matters.
3. Citations point at the memo, not at the sentence
When a source is attached to the memo as a whole rather than to the specific point it supports, tracing any single sentence back to its authority becomes a hunt. The reviewer reopens documents, searches for the passage, and rebuilds the link the associate already had.
4. Quotes get separated from their pincites
A passage copied into a notes document, then copied again into the draft, tends to lose the page or section reference on the way. The quote survives, its exact location does not, and confirming it means finding the source and the passage from scratch.
5. The draft runs ahead of verification
Under deadline, the drafting starts before the sources behind it are confirmed. The memo becomes a holding pen for half-checked points, and the team spends its remaining time editing around uncertainty instead of resolving it.
6. AI output is treated as a finding
A summary that reads cleanly is not a verified authority. A case an AI tool names is not confirmed to exist, to be current, or to say what the tool claims. When those outputs flow into a draft without being checked against the primary source, the memo inherits a confidence it has not earned.
A worked example on one question
To make this concrete, here is an invented scenario. It is illustrative only, the facts are made up, and it deliberately does not state what any law actually says, because that is the associate's and partner's job, not this guide's.
Say an associate is asked whether a non-compete covenant survives an asset sale in two states, and the client wants to know before signing. The question sounds tidy until the facts sharpen: it turns out the covenant sits inside an employment agreement in one state and inside a separate sale document in the other, and the governing-law clause is being renegotiated as the deal moves. Now the associate is really answering two related questions under two bodies of law, one of which depends on a clause that is not final.
In a scattered workflow, none of that texture reaches the reviewer. The draft says the covenant "likely survives," the partner asks why, and the reconstruction begins. In a workflow that holds the pieces together, the same question shows up with the governing-law assumption flagged as unconfirmed for the second state, the two authorities that back the first state attached with pincites, and the open point, whether the renegotiated clause changes the answer, marked as unresolved and assigned back for follow-up. The associate has not answered the question yet. The partner can see exactly how far the answer has actually gotten, which is the entire point.
Start with a research packet, not a finished memo
The most useful thing to build first is not a smarter memo template. It is the working file that sits behind the memo. Call it a research packet. It holds the question, the fact assumptions, the source list with excerpts and status, the citation checks, the draft sections tied to their sources, and the open questions. It is deliberately not a black box that emits an answer.
A packet earns its keep when it makes a few habits automatic. Attach each source to the point it supports, not to the memo as a whole. Keep extracted facts, research notes, and draft analysis in separate places so nobody confuses a quoted passage with a conclusion. Write the memo prose only once the source packet behind it can actually be reviewed. None of this is heavy. It is mostly a decision to stop letting the trail evaporate between steps.
What the packet should hold
- The question as it stands now, with the facts and assumptions it depends on.
- Each source, its status of approved, pending a lawyer's check, or excluded, and the excerpt with its pincite.
- Draft sections, each tied to the authorities that back them, with a marker where a point has no source yet.
- Open questions and unresolved authority, flagged rather than buried.
- Who owns the first pass, the citation check, the legal review, and the final sign-off.
What the packet shows on a single issue
The value of the packet is that a reviewer can look at one issue and see, in a single view, how solid it is and what it still needs. No hunting, no reconstructing. The table below is what one issue looks like when it is written this way.
| Issue | Source status | How the draft may use it | Reviewer note |
|---|---|---|---|
| Whether the covenant survives the asset sale in state one | Two approved authorities attached with pincites | Written as analysis with citations in place | Confirm both authorities are still current before sign-off |
| The same question in state two | Governing-law clause still being renegotiated; assumption flagged unconfirmed | Background only, not the conclusion | Do not draft the conclusion until the clause is final |
| Effect of the renegotiated clause on the answer | No approved authority yet, marked unresolved | Keep out of any client-facing draft | Assign follow-up research before the memo is reviewed |
Keep what was found apart from what gets decided
The habit that saves the most review time is keeping what the sources say separate from what the lawyer decides. Extracted facts, quoted passages, and the associate's analysis should never blur into the reviewer's conclusions. When they are kept apart, a partner can accept the research and still change the conclusion, or accept the conclusion and know precisely which authorities it rests on.
This is also where the boundary lives. The workflow organizes and surfaces material. It does not weigh authority, resolve an open question, or write the client-facing conclusion. Legal analysis, privilege, and strategy stay with the lawyers, and the packet simply records who signed off on each part. That separation is what lets a reviewer trust the parts they did not personally do.
Citation hygiene is what actually protects the firm
Every point that reaches a memo should be traceable to an approved source, and every authority an AI tool suggests should be verified before a lawyer relies on it. This is the part that keeps a fluent summary from becoming a fabricated citation. The discipline is simple to state: the tool produces a candidate, a lawyer confirms it against the primary source, and only then does it earn a place in the draft.
| What the AI produced | What it is until verified | How a lawyer verifies it | Who owns the result |
|---|---|---|---|
| A summary of a statute or regulation | A starting read, not a finding | Read the cited section in the primary source and confirm it is current | Responsible lawyer |
| A case cited for a proposition | An unconfirmed authority | Pull the case, confirm it exists, is still good law, and says what was claimed | Responsible lawyer |
| A quoted passage with a pincite | A candidate quote | Check the wording and the pincite against the original | Responsible lawyer |
| A point in the draft with no citation | A gap, flagged | Find an approved authority or cut the point | Responsible lawyer |
If a passage cannot be traced to an approved source, it does not belong in the draft, no matter how confidently it reads. That single rule handles most of what people mean when they worry about AI inventing authority.
Where AI helps inside the research, and where it must not
AI is genuinely useful on the assembly work around the research. It can summarize a long source into a first read, pull the passage you need with its citation, organize scattered notes into an issue outline, compare where two sources disagree, and flag a sentence in the draft that has no authority behind it yet. Used this way, it removes the copying and searching that eats an associate's afternoon.
It must not invent authority, smooth over uncertainty, or stand in for legal analysis. A summary is a starting point, never a finding. Every output it produces carries its source links back to the reviewer, and every authority it suggests goes through the verification step above before anyone relies on it. The moment AI is allowed to be the last reviewer rather than the first drafter, the workflow has quietly become the risk it was meant to remove.
The sources and systems the workflow has to hold together
A research workflow usually touches more places than people expect: the research platforms the firm already pays for, the document management system, client documents, prior memos on the same issue, any citation-checking tools, and the editor where drafting happens. You do not wire all of them together at once, and you should not try.
The first design choice is narrower and more important than the integrations: which sources the firm approves for a given type of research. That decision is what gives the packet a foundation. Once the approved sources are settled, the tooling has something reliable to attach to, and the question of "can we cite this" has an answer that does not depend on which associate is asking.
Where the responsible lawyer's judgment stays
A research memo is full of judgment, and a workflow protects that judgment rather than replacing it. A tool can show that two authorities point in different directions. It cannot decide which one governs the client's facts, how much weight a distinguishable case deserves, or whether the answer is confident enough to act on. Those calls belong to the lawyer, and the packet's job is to make sure they are the calls being spent time on.
So the human owns the parts that actually carry risk: deciding which sources are reliable and which are merely convenient, reading ambiguous authority in light of the real facts, separating a genuine open question from a settled one, choosing what the client needs to hear and what stays in the file, and approving the final conclusion. The workflow keeps those decisions from being buried under copying, searching, and version control, which is where they usually get lost today.
Where this usually goes wrong
A few mistakes turn a promising packet back into the mess it was meant to fix.
Letting the draft outrun the sources
If prose gets written before the sources behind it are confirmed, the packet becomes a tidy record of an answer nobody can verify. Draft from confirmed sources, and mark the unresolved points as open rather than writing around them.
Blurring what was found into what was decided
When extracted facts and analysis share a lane, the reviewer has to untangle them all over again. Keep the quotes, the notes, and the conclusions visibly apart.
Approving sources loosely
If anything the associate can find counts as citable, a memo will eventually cite something the firm would not defend. Decide the approved sources for each type of research and hold to them.
Treating an AI summary as the finding
A clean summary invites the team to skip the verification it still needs. Route every AI-assisted point through a source check with a named reviewer before it influences the conclusion.
Finishing the memo and losing it
A verified memo that vanishes into an unsearchable folder helps no one the next time the same question arises. File it where the next associate can find the question, the sources, and the sign-off together.
What tells you it's working
The clearest signal is felt before it is measured: reviewers stop dreading the memo, because the trail is already in front of them. A few numbers confirm the feeling. The time a reviewer spends reconstructing where a draft came from should fall toward zero. Memos should come back less often because a citation was missing or a source could not be traced. And open questions of authority should be flagged before review rather than discovered during it. If those three are moving, the workflow is doing its job, whatever the tooling underneath looks like.
How Ubisar would build this with you
In week one we would take one repeatable memo type and follow it from the legal question to a reviewed draft: the fact assumptions, the approved sources, the excerpts and pincites, the citation checks, the issue outline, and the open questions. The first thing you would have is a research packet that holds all of it together, with the lawyer's sign-off as a real step rather than an afterthought.
In weeks two and three we would connect the minimum set of research platform, document management system, client documents, prior memos, and citation tools that packet actually needs, and put AI behind the assembly work: summarizing sources, organizing outlines, pulling quotes, flagging gaps. Your lawyers keep the analysis and the client-facing conclusions the whole way through. By week four you should be able to run a recent memo through the workflow and see exactly where review gets faster or safer.
| Stage | Focus | What exists by the end |
|---|---|---|
| Week 1 | Make one memo type traceable | A research packet holding the question, approved sources with status, excerpts and pincites, draft sections tied to authorities, open questions, and named owners for each review step |
| Weeks 2 to 3 | Connect the sources and add AI to the assembly | The minimum research platform, document system, prior-memo, and citation-tool links, with AI summarizing, outlining, and flagging gaps under a verification step |
| Week 4 | Prove it on a real memo | One recent memo run end to end, with a clear read on where review got faster and where a citation risk was caught earlier |
When the research feeds a negotiation, the packet hands off cleanly into contract review triage rather than being retyped from scratch.
We work one workflow at a time, month to month, starting from $4,000/month, and you can stop whenever it stops earning its place. If your research-to-memo work is where the time and the worry sit, tell us about it or read how the monthly implementation service works, and we will scope the first month around it.