Part 1 of 2: Thoughts on Information Management with GenAI in Practice
By James Jordan
Early in my legal support career, I worked as a paralegal at McCutchen, Doyle, Brown & Enersen, a great firm that’s unfortunately no longer around. I think I was originally hired as a fill-in utility junior paralegal, moving case to case, but eventually I landed on one matter with a team of 6-7 paralegals supporting our deposition preparation and motion practice.
One of our primary tasks was data entry using the InMagic DB Text platform. We used it as a case integration tool, ingesting deposition transcripts and relevant chunks from the larger document productions. Part of my job was entering objective data, but I also reviewed prior entries for typos against an index. DB Text could handle proximity and stemmed searches, and an autofill that I’d compare to the Excel dropdown feature. Still, all pretty advanced for the time. It didn’t correct your typos or consider context, so what you entered was recorded exactly as entered until someone corrected it. Tedious QC work, but very necessary for information retrieval.
Now we don’t have to pay attention to entry details the way DB Text forced us to. You can use one or a few words to perform document-level searches across your home or work computer. And now, we have semantic understanding. Your typos are being corrected, whole sentences are being completed, and context is being added automatically.
Think about that. We have miscommunications every day with people we know and love, but now a robot is hitting us with “Well, actually, this is what you really meant.”
Let me be clear, I’m not anti-GenAI. I have strong concerns about IP, AI infrastructure, and environmental impact, and from a creative standpoint, keep it away from anything I’d consider artistic expression. But for back-end operations and research, I see real value.
We have so many tools that index content. Retrieval, even without AI enhancements, is so easy we don’t have to think about it. Just save to the location you’re familiar with now and search with a few words later. Users know how to pull up a document, but saving and organizing it for someone else’s quick access is another story.
I’ve been in rooms recently listening to discussions about fake cases and incorrect interpretations of authority making their way into briefs. There’s a lot of finger-pointing at “bad AI” or the lazy pro per litigant, but teams are submitting bad briefs as well. Near-final QC is being neglected along with other foundational habits upstream because these submissions are getting past multiple sets of eyes.
One of the most commonly adopted GenAI tools for day-to-day use is the relatively low-volume document review and drafting tool. Many of those tools rely on retrieval-augmented generation (RAG) to ground their responses in selected source material. That could be a civil complaint, a business contract, or a folder of related drafts with legal authority from who knows where. RAG relies on text content, file titles, and other metadata for retrieval and context.
When litigation teams prepare motion papers, it’s not uncommon to have one to four draft streams circulating with different variations. One on plain paper, another two on pleading paper, including one reformatted from a similar motion, with multiple attorneys tackling different sections while the client provides comments at the same time. Toss in prior legal research, LexisNexis downloads pulled from opposing counsel’s briefs, and now AI-assisted research, and it’s easy to see things getting confusing.
Even without AI layered on top, you’ll need some organization and a really good traffic cop: “that’s the old motion,” “this is the client-comment version,” “that research was rejected,” etc.
If your reference files aren’t organized or named correctly, you increase the risk that a RAG tool, which may not understand the hierarchy that’s obvious to you, will retrieve the wrong or incomplete material while still producing a polished-sounding answer or phony citations.
Toss in a few of these bad boys, and it can get dicey:
Using the AI RAG tool may require better prompting, and that part of the equation always evolves. Source confirmation and file management (i.e., QA/QC) are the constants.
The combination of simple drag-and-drop filing and easy document-level search has made us a little too comfortable. We know we can probably find the document later, so there’s less pressure to understand where it belongs or why it lives there. But quick retrieval doesn’t build context. Knowing where a document sits, what it relates to, and how it fits contextually does. Judgment comes from situational context.
DMS tagging can help with context. We can add document characteristics and metadata manually, and depending on your operational budget, AI can help with that too. Over time, richer tagging may reduce how much we rely on highly complex folder structures because a document can be organized and retrieved in multiple ways.
But folders still matter for us humans because we have to understand workflow, security, and shared context. That means that even with GenAI, we still need to stay grounded in the fundamentals. If AI starts doing more of the tagging, someone still has to know enough to catch it when it gets something wrong.
Get back to the basics: Start with clear file naming. One of the most important principles in litigation is clear ownership, both substantive ownership and process ownership. The partner owns the legal strategy and final work product, while the senior associate typically owns the overall process. That means managing the drafting workflow, establishing standards, maintaining consistency, and making sure support roles, expectations, and deadlines are clear.
I’m starting to see support staff relinquish case file management responsibilities to associates, and honestly, it’s kind of confusing. Every team member should know the naming convention well enough to spot a poorly named file and correct it on their own, but your non-billable staff members should own this as a standing task.
Your practice group will decide the format. What matters is buying into consistency once they do. Before anyone opens the file, it should include date-related information, the author, and what it is.
Three elements make that work:
If the document is correspondence, include the sender, the recipient(s), and the purpose (see the re: line).
Here are examples showing the range of options, not a single right answer. Teams differ on hyphens versus underscores, abbreviations versus full names, spaces versus none, plaintiff/defendant versus proper nouns, and whether to prefix a client-matter number. Settle on one convention for your practice group and hold to it.
If your team is consistent about the objective information in a file name, the file list should be easy to scan and understand. Abbreviations are fine if they make sense for the team, but be consistent. Consistency travels. If you ever have to move data between platforms or bulk-share files with a client, expert witness, AI tool, or another outside party, those habits make the process much easier and reduce retrieval errors.
Version control means everyone can tell which copy of a document is the right one to work from, without guessing.
Two approaches tend to work well. Either is fine, as long as everyone on the team is using the same one.
If iManage or a similar DMS is available at your firm, a good chunk of this already happens automatically: every save gets its own version, and the file locks while someone’s working in it. Where it gets tricky is the moment a document leaves the DMS, gets emailed out, edited on someone’s desktop, and comes back in. That copy is off the system’s radar the whole time it’s out, so it’s worth applying the same naming discipline to it until it’s checked back in as a new version.
When live co-editing isn’t an option, turning on Track Changes has to be mandatory, no exceptions. Other tools can come into play for comparing versions, Litera Compare being one, but Track Changes is native to Word. Crank it up!
One thing worth watching for is a document getting labeled “Final” a little before it actually is. I’m guilty of using “near final” myself to signal we’re close, but a true final should be ready for service. Once something is genuinely final, it’s worth moving into the matter’s correct subfolder, correspondence, Docket/Pleadings, or discovery, rather than leaving it in whatever folder it was drafted in.
If matter work is happening somewhere with live co-editing, OneDrive, Google Docs, that kind of setup, it tends to work best as a client-managed space, rather than a shortcut around the naming and filing habits above.
I’ll get into folder structure in another article, but the short version is the same: standardize it for the firm, then for the team. If you work in a specific practice area, keep the hierarchy consistent there too. Folder structure provides context.
Thinking about filing and naming conventions is a snoozefest. I get it. But there’s always a new tool, and whatever is cutting-edge today may not be around in a year. Strong foundational habits carry over no matter what shows up next.