By Bill Hartzer · Published · How this site is researched
Sequence. The expert reaches each conclusion from the evidence before consulting any system about the matter. No system is asked what conclusion the evidence supports, and no system is asked to support a conclusion specified in advance.
Why the order of the work comes first
The clauses of the Standard are numbered deliberately. Nearly every other obligation can be repaired after a lapse. Sequence cannot. Once a tool has seen a question with the expert's preferred answer built into it, the expert cannot unlearn the reply, and no later check will show how far it shaped the work.
The book traces this to a property it calls agreeableness (the tendency of these systems to produce answers their user finds satisfying, which is not the same thing as accurate). A question that signals the answer it hopes for tends to receive agreement, supported by reasoning assembled afterward. Nothing defective appears on the page, and the report survives a full verification pass, because the defect lies in which questions were asked and which contrary reading was never tested. A retained expert knows which side is paying. Procedure, not good intention, keeps that knowledge out of the opinion.
Exploration is not the forbidden question
The clause is easy to over-read. It rules out two requests only: asking a tool which conclusion the evidence favors, and asking it to back a conclusion already picked. It does not forbid research. Asking how a method works or what the literature says about a technique is ordinary preparation. The answer is a lead to be pulled and read, never a finding.
The working test is the wording of the request actually typed. "What approaches are recognized for estimating lost profits?" is open. "Explain why the before-and-after method is correct for this plaintiff" is leading, and a chat assistant will comply with it cheerfully. A question that carries the facts of the matter and the hoped-for answer has stopped being exploration. A legal theory that surfaces along the way should go to retaining counsel as a question rather than into the report.
Nor does the clause bar asking a model, after the conclusion exists, to build the strongest case against it. The conclusion came first, and the tool is testing it rather than supplying it.
Proving the order in a working file
Sequence is invisible in a finished report and visible in a record. Write each question down before asking it and read it back; if it names a conclusion, rewrite it until it does not. Make a dated entry in the AI-use log on the day a conclusion is reached, stating that no tool was consulted about the merits beforehand. Keep any contrary-case testing in its own named session, dated after that entry. Treat material that arrives pre-sorted or summarized by someone else's software as a framing already applied, and go back to the records.
When a lapse comes to light, the remedy the book sets out is to put the output aside, re-derive the conclusion from the evidence, record what was done and why, and tell counsel. That cure is harder than it sounds, and still far better than a file showing an expert who never noticed. The dated entry is what makes the order provable under Clause 7.
What the record shows
In Ferlito v. Harbor Freight Tools USA, a product-design expert put a question to ChatGPT after finishing his report, and the answer matched the opinion he had already reached. The federal court denied the motion to exclude, relying on the order of events: the query followed the report, and the opinion rested on the expert's experience. The timing came from his own account at the hearing, not from a log.
The Watson Grinding explosion trial (3M expert) shows the reverse order, as press reports describe it. An engineering expert's produced chat history reportedly included prompts asking the system to show that his client was 0% at fault. He was examined on those logs at trial. No tribunal ruled on the AI use, and the entry rests largely on news reporting.
In Plonit v. Clal Insurance (DigitalOwl summary), the tool belonged to a party. An insurer ran a claimant's medical records through an analysis platform and sent the resulting summary to the court-appointed expert; in an interlocutory order, the court told the expert to disregard it and had the records resent unmarked. Directed emphasis can set the sequence for an expert who never typed a prompt. And in Colombia: Cartagena ChatGPT ruling and Sentencia T-323/24, the Constitutional Court found no due process violation where a judge had decided the case before consulting ChatGPT, while finding that transparency and accountability had not been fully met.
The rules it sits beside
The Standard does not restate any court rule and does not depend on one. In United States federal courts, Federal Rule of Evidence 702 asks whether an opinion reflects a reliable application of reliable methods to the facts; an opinion formed after a system was asked what the evidence supports invites that question. Other systems ask the same questions in their own terms, and the matters below include tribunals in several of them.
Matters in the record that engaged Clause 1
5 of the 103 matters in AI in the Courtroom engaged this clause. Each is listed with the sentence from its entry that says how. The mapping describes conduct, not blame, and each entry sets out the facts it rests on.
Expert witnesses and AI
- Watson Grinding explosion trial (3M expert) “The reported prompts engage Clause 1, sequence, and Clause 2, non-delegation.”
- Ferlito v. Harbor Freight Tools USA “This matter illustrates Clause 1, sequence, being met.”
- Plonit v. Clal Insurance (DigitalOwl summary) “Clause 1, sequence, is engaged from an unfamiliar direction.”
Judges, courts and AI
- Colombia: Cartagena ChatGPT ruling and Sentencia T-323/24 “The Court's reasoning follows the order in Clause 1 of the Expert Record Standard, sequence.”
AI inside the courtroom
- Jones v. Delta Air Lines (ChatGPT at a deposition) “Clause 1, sequence, is engaged in its plainest form.”