By: Jeff Johnson, Chief Innovation Officer
Attend any eDiscovery technology conference or demo today, and you’ll likely hear some variation of:
“Search terms are obsolete. AI is better.”
It’s an appealing idea. For decades, we’ve dealt with the limits of keyword search. Terms miss relevant documents. They pull in junk. They take weeks or months to negotiate. They’re hard to validate. Even done well, they rarely capture everything that matters, without including a lot of what doesn’t.
Now we have generative AI enabled analysis and the assertion that it can plug in and solve all of the problems inherent in search terms.
Does it? Sometimes. Not yet, in most large cases.
For now, the real question is how AI and search terms work together. I am grateful to our friends Leah Bays of Robbins Geller and Tessa Jacob of Husch Blackwell for joining me to discuss exactly this on a recent webinar.
Why Search Terms Still Matter
For all of their shortcomings, search terms provide something the legal industry values: transparency.
Both sides understand how they work.
Both sides know how to negotiate them.
Both sides can evaluate hit reports, discuss scope, and understand what is being included or excluded.
In the midst of demonstrable imperfection, search terms provide a comfortable framework that courts, litigants, and discovery professionals have been working with for decades.
Additionally, search terms can remain an important part of cost management in truly large (multi-million) document collections.
As a result, many matters still involve search terms, even when the review/production workflow also utilizes AI.
In fact, one of the most common approaches we’re seeing is not search terms versus AI;
It’s search terms plus AI.
Search terms may help exclude the most obvious irrelevant data. AI can then prioritize, categorize, and analyze within the reduced set.
That may not be the revolutionary transformation some predicted, but it is often a practical and defensible one.
AI Is Not an Easy Button
One misconception that continues to surface is the idea that AI can simply be pointed at a document collection and trusted to deliver the right answer.
If you’ve used these tools beyond experimentation and simple use cases, you know that’s not how it works.
Effective AI use in early case assessment and review requires planning, testing/calibration, validation, and expert oversight.
These workflows often involve multiple rounds of prompt refinement, sampling, and quality control before teams are comfortable relying on the results.
In many ways, the legal work and expertise needed shifts rather than disappears.
Review teams spend less time reading every document and more time validating outcomes, refining workflows, and ensuring the technology is behaving as expected.
An AI-enabled workflow should never prioritize efficiency above defensibility. The only path to defensible AI use is the appropriate expert oversight.
Defensibility Is the Standard, Not the Selling Point
Efficiency is easy to demonstrate. Defensibility is what holds up when the process gets challenged.
Every AI-enabled workflow should be measurable and documented.
The metrics by which we measure these workflows are standard search and retrieval measurements like richness, recall, elusion, and precision. These measurements demonstrate whether we are finding and producing the relevant documents we need to, or not.
Statistical sampling and validation review of a relatively small set of documents (typically 1-3K, depending on the situation) allow for statistically sound estimates of these metrics.
Document the process as you go, not after the fact. Account for every document in your starting population, record your prompts, your sampling methodology, your validation results, and the decisions your team made along the way. That documentation is what turns “we used AI” (efficiency) into “here is how we did the work, and here is how we know our result is reasonably correct” (defensibility).
That’s the standard. Do the work correctly, then prove that you did.
Perspective Depends on Your Side of the “V”
One of the most interesting themes from our webinar was how differently plaintiffs and defendants often view the same technology.
For the defense, the lens is usually volume, cost, and speed:
- High/Very high document volumes
- Compressed deadlines
- High review costs
Anything that reduces, prioritizes and accelerates review earns consideration.
On the plaintiff side, the lens is different. The question isn’t how fast you can review (while timelines are still tight!). It’s whether something important got missed in the acceleration.
That distinction drives everything. The same AI workflow that creates efficiency for the producing party can create questions and doubts for the receiving party. The technology is the same. The perspective – is different.
If you’re deploying AI in discovery, design workflows for both objectives from the start.
Disclosure and Cooperation Change the Outcome
Here’s the tension at the center of every AI-enabled production.
The producing party controls the workflow. The receiving party carries the risk of what that workflow might miss. Left unaddressed, that gap often turns into arguments.
The fix isn’t complicated, but it takes effort early. Talk before you produce, not after. Don’t spend weeks or months negotiating search terms, before you disclose that you plan to use AI to further reduce attorney review. Negotiate your search terms with mutual understanding that is the plan.
That conversation should start at meet and confer and continue as needed. The Sedona Conference has been making this case since its 2008 Cooperation Proclamation: cooperation on process reduces cost and disputes without compromising advocacy.
Cooperation is not capitulation. You aren’t giving up strategy. You’re establishing that your process was reasonable and defensible, which is exactly what a court will want to see if another party challenges your production. Generally, the party that discloses a sound process and validation protocol is in a far stronger position than the party defending a black box after the fact.
Looking Ahead
We’ve been here before.
When technology-assisted review first arrived, some predicted it would replace manual review outright, and others refused to trust it at all. Courts eventually accepted it, in the context of a defensible, well-documented process. The tool was never the point. The process and validated results were.
Generative AI is following the same path. It will handle more of the work over time. It won’t remove the accountability that sits with the legal team.
So where does that leave search terms? Could we eventually reach a point where AI replaces search term negotiations altogether?
Maybe. But we’re not there yet.
In the near-term, search terms will remain one potential input among several. They will keep doing what they do well, managing cost in massive collections and giving both sides a transparent, negotiable framework. AI adds refined reduction, prioritization, categorization, and analysis that keywords and hit counts never could.
The teams getting the most out of this aren’t picking one tool. They’re combining methods, using them synergistically, validating results (from a to z), and documenting the work.
That may not be as exciting as relegating search terms to the past, but it is reality.
Let’s Build Defensible Workflows Together
Whether AI, search terms, or both fit your matter depends on the collection, the timeline, and what you have to defend. That’s a workflow question, not a product question.
That’s the work we do at Purpose Legal. We design and execute AI-enabled workflows through our PurposeXi delivery framework — combining search terms, AI, and analytics with oversight and expertise (yours and ours). Documented and measured workflows that hold up to challenge.
If you’re planning how to bring AI into your next matter without giving up defensibility, let’s talk through it before discovery starts, not after.
Click here to schedule time with Jeff Johnson and continue the conversation.
