™Sooner or later, on a file where you used AI, opposing counsel is going to ask about it. Maybe in a deposition, maybe on the stand, maybe in a motion to exclude your report. They'll ask the same way they used to ask about your note-taking habits or which database you pulled an address from: looking for the soft spot that turns a solid finding into a doubtful one.
The investigators who handle this well aren't the ones who avoided AI. They're the ones who can explain exactly what it did and exactly how they checked it, in a calm voice, without a lecture.
Three answers that make it worse
Denying or minimizing it. If AI touched the work, it's probably discoverable in your own file: the prompts, the drafts, the time entries. Shading the answer turns a methodology question into a credibility question, and credibility questions have nothing to do with whether your findings are right.
Explaining the technology. Walking a jury through how a language model works buries your finding under material that opens more questions than it closes. Nobody on the jury needs to understand the tool. They need to trust the fact.
Getting defensive. Treating the question as an attack tells everyone in the room you think there's something to hide, even when there isn't.
The answer that holds up
Keep it short, keep it factual, and put the weight on your verification rather than on the tool. Something close to this:
"I used an AI tool to help organize and summarize a large volume of records. I independently verified every material fact in this report against the primary source before it was included."
That sentence does three things. It admits the tool was used, which you can't credibly deny anyway. It states the standard you held yourself to. And it moves the argument back onto the verified fact, which is where you want it.
The question is never really about the tool. It's about whether you can show your work.
The four questions to expect
"What AI tool did you use, and for what?" Name the tool and name the task. Naming it isn't a concession. It's a fact, and hedging on it looks worse than saying it plainly.
"How do you know it didn't make something up?" This is your opening to talk about verification, not to defend AI. "Because I checked it against the certified record myself" ends the line of questioning faster than any explanation of how reliable the technology is.
"Did you personally verify this specific fact?" If yes, say so and say how. If the honest answer is no, you have a real problem, and it's one you should have caught long before you sat down in that chair. That's what the audit trail below is for.
"Isn't it possible the AI got this wrong?" Yes. And that possibility is exactly why nothing in your report rests on AI output alone. Said plainly, that answer tends to close the subject instead of opening it.
What you should never say
Don't say the AI "found" or "determined" anything of investigative significance. Don't imply it replaced your judgment. It organized, summarized and flagged. You investigated, verified and concluded. Your license is on the report, and your testimony should sound like it.
Where the answer actually comes from: the audit trail
The one-sentence answer only works if it's true and you can prove it. That's the job of an AI audit trail: one row for every AI-assisted step that could change a finding. Skip it for grammar passes and formatting. Use it for anything that could end up in a report, an affidavit or a cross-examination.
Each row records:
- Case number and date
- The tool and what it was asked to do
- What material went in, and whether it was redacted first
- What came back
- Which primary source you checked it against, and how
- Who verified it, and when
- What happened to it: entered in the report, discarded, or kept as working material
A typical entry looks like this. A tool does a first pass on 412 pages of supplemental police reports and flags six discrepancies between witness statements. You read each one against the original Bates-numbered pages. Four hold up. Two were artifacts of the summary, and you throw them out. The four go in the report with page citations. The AI chronology stays in the file, labeled as preliminary working material, and never leaves the office.
Before any deposition or trial on a file where AI was used, pull that log and reread it. You should be able to say, without hesitating, which findings came through an AI-assisted step and which primary source confirmed each one. Vague recollection is where credibility erodes. A specific, documented answer is where it holds.
Courts are already asking
This isn't hypothetical anymore. In Mata v. Avianca, Inc., 678 F. Supp. 3d 443 (S.D.N.Y. 2023), attorneys were sanctioned after filing a brief that cited cases an AI tool had invented. Since then, federal judges have issued dozens of standing orders governing AI in filings. A Bloomberg Law analysis counted 39 federal judges with such orders as of April 2025.
Their requirements vary. One of the first, Judge Brantley Starr's "Mandatory Certification Regarding Generative Artificial Intelligence" in the Northern District of Texas, requires counsel to certify either that no portion of a filing was drafted by generative AI or that any AI-drafted language was checked for accuracy by a human being.
Most of those orders govern attorneys and filings, not investigators directly, and there's no nationwide rule. But the pattern is the same everywhere it shows up: disclose it, verify it, stand behind it. An investigator whose practice already works that way has nothing to adjust when the question reaches him.
Do this before your next file
- Start an AI audit trail log today, even a spreadsheet with the seven columns above.
- Add a short AI disclosure to your engagement letter, so the client heard it from you first.
- Write out your one-sentence answer and say it out loud until it sounds like you, not like a script.
- Never put privileged, sealed or juvenile material into a public AI system. No efficiency is worth your license.
This article is adapted from Chapter 13 of The AI-Powered Private Investigator, which includes the full audit trail template, a client disclosure you can drop into your engagement letter, and a testimony-preparation prompt that builds your answers for a specific case. The Foundation module builds disclosure and testimony prep into every file you open.
Jon Jacobson, MBA, is a licensed Arizona private investigator (license #1799200) and principal of Old Pueblo Investigations in Tucson. He is not an attorney, and this is not legal advice. Check your jurisdiction's rules and talk to counsel before relying on any of this in a proceeding.