Artificial intelligence can save land professionals countless hours. But what happens when one seemingly small mistake costs a company millions?
“Mom, I’ve been hit by a car!”
Those are words no mother ever wants to hear.
My daughter was crying when she called to tell me she’d been involved in an accident while riding her electric bicycle. She had been traveling on a sidewalk when a car pulled out of an apartment driveway directly in front of her. Unable to stop in time, she collided with the vehicle.
An ambulance was called. She had broken her wrist and two front teeth, suffered abrasions to her face and arms, and required stitches in her chin.
Thankfully, she had been wearing a helmet.
When I received the police report, I was surprised to discover that the investigating officer had attributed fault to my daughter for riding on the wrong side of the roadway.
There was just one problem. She wasn’t riding on the roadway. She was on the sidewalk.
As an attorney, I wanted to investigate the legal significance of that distinction, so I turned to artificial intelligence (AI), a tool I frequently use to assist with legal research.
AI confidently informed me that the state’s Driver’s Handbook instructed motorists to watch for bicyclists on sidewalks when entering or exiting driveways. It even provided the page number where I could find the provision.
Perfect! Exactly what I needed.
Except there was another problem.
The provision no longer existed.
It had appeared in an earlier edition of the handbook but was removed several years ago.
Fortunately, I checked the actual handbook before relying on the information.
At a recent NADOA conference, I heard a speaker make an interesting observation: AI wants to please you. It wants to give you the answer you’re looking for.
There’s actually a term for this tendency—AI sycophancy. AI systems can sometimes reinforce a user’s assumptions instead of challenging them. They can also provide inaccurate or outdated information with remarkable confidence.
Whether AI was trying to give me the answer I wanted or simply relied on outdated information, the result was the same. It gave me a convincing but wrong answer.
And that got me thinking about the implications for the oil and gas land profession.
What Happens When AI Gets the Lease Wrong?
One of the most promising AI applications in our profession is lease data entry.
Imagine uploading an oil and gas lease and having AI automatically extract the parties, legal description, royalty rate, primary term, rental provisions, and other contractual obligations. For companies managing thousands of leases, the time savings can be enormous.
But what happens when AI gets something wrong?
Suppose a company acquires a three-year oil and gas lease requiring annual delay rental payments. The company uploads the lease into an AI-assisted land management system, which incorrectly identifies it as paid up.
No rental obligations are entered. No payment reminders are generated.
Nobody notices the mistake.
A year later, the rental deadline passes.
According to standard lease language, that lease would terminate automatically. The company may not discover the problem until after drilling is completed. Now we have no right to be on the land. We’ll have to renegotiate the lease with the mineral owners.
Imagine the leverage they’ll have for bonus and royalty payments! Or, they could sue us for trespass! That’s a potential multi-million-dollar mistake! All because AI missed the delayed rental payment.
A million-dollar lease could be lost because AI checked the wrong box.
And here’s the frightening part: every other provision in the lease abstract might be perfectly accurate.
One mistake could still cost the company its lease.
Can AI Really Run Title?
I teach a course on mineral title examination, and one exercise I give my students involves tracing mineral ownership through a complicated chain of title.
Students must examine conveyances, interpret reservations and exceptions, and determine who ultimately owns the minerals. It’s not simply a matter of reading documents and calculating fractions. It requires an understanding of property law and the legal effect of each instrument.
Now imagine giving that same assignment to an AI title examination system.
Could it identify every relevant conveyance? Recognize an ambiguous mineral reservation? Distinguish between a mineral interest and a nonparticipating royalty interest? Apply the appropriate state’s law?
Perhaps. AI technology is advancing rapidly.
But would I accept its ownership determination without independently verifying the chain of title?
Absolutely not.
An incorrect mineral ownership determination could result in leasing from the wrong party, paying substantial bonuses to someone who doesn’t own the minerals, or distributing royalties incorrectly.
Depending on the value of the interests involved, that’s potentially another million-dollar mistake.
And just like the lease abstract, an AI-generated title report might look perfectly professional while containing a critical error.
In mineral title examination, getting 99% of the analysis right isn’t necessarily good enough. The remaining 1% could change who owns the minerals.
We’ve Been Verifying Data for Decades
When I worked in Exxon’s Land Department, two highly skilled employees entered lease information into the company’s computer system. As a lease analyst, my job was to independently verify that the information they entered accurately reflected the underlying leases.
In those days, Exxon required its lease analysts to be attorneys, so the verification process involved a separate level of professional review.
In all the years I performed that work, I never found a single mistake made by either employee. They were that good.
But the verification process remained in place, regardless of how accurate their work was.
I see AI potentially serving much the same function today. It may extract lease provisions and enter them into a land management system with remarkable speed and accuracy.
But someone still needs to verify the information against the original lease.
Technology may change who—or what—performs the initial work. It doesn’t eliminate the need for professional oversight.
The Bottom Line
I use AI regularly, and I believe it has tremendous potential to improve efficiency throughout the land profession. I’m certainly not suggesting that we avoid it.
But we need to understand its limitations.
When AI analyzes a lease, someone must verify the critical provisions.
When it examines title, someone must verify the ownership conclusions. And when it cites a legal authority, someone needs to make sure that authority actually exists and says what AI claims it says.
My experience with the Driver’s Handbook reminded me that a confident answer isn’t necessarily correct.
Fortunately, I checked the source.
As the oil and gas industry explores the potential uses of AI in lease analysis, title examination, mineral ownership research, and division order calculations, I hope you’ll remember the importance of independently verifying its conclusions before relying on them. Because in the land profession, a million-dollar mistake can begin with a very convincing answer.