A Job Interview That Felt Like Something More

According to the LinkedIn post that prompted this article, the candidate applied for a transactional-law position. The next stage was a 28-minute AI interview covering contract negotiation, legal reasoning, AI-assisted contract review, and a more unusual area: data annotation and rule design for training legal AI. He said he answered in detail from years of practice experience. The next day, an automated recruiting system rejected him and provided specific feedback on both his legal reasoning and his AI-training annotation skills.

That sequence led him to ask whether he had participated in a hiring process or provided high-quality professional data for free.

That suspicion should not be turned into a factual accusation. There is no evidence that the employer or platform used his interview answers to train a model. And if the role itself involved training legal AI, testing annotation, rule design, and legal reasoning would be entirely relevant to the job.

Still, the story exposes a broader issue that is becoming harder to ignore.

Where Does Assessment End and Data Production Begin?

Traditional interviews mainly asked a simple question: can this person do the job? Candidates described prior projects, answered technical questions, or completed a short work sample. The purpose was usually easy to understand.

AI-era expert hiring creates a different possibility because the process of answering can itself generate valuable data.

Ask a lawyer how to negotiate three proposed contract changes. Ask a patent practitioner what to check first in a §103 obviousness rejection. Ask an FTO analyst what to do after a product appears to map onto an independent claim. The final answer matters—but often the more valuable material is the reasoning path:

  • What does the expert check first?
  • When is the evidence considered insufficient?
  • Why is an apparently workable option rejected?
  • What additional facts would change the conclusion?
  • How does the expert structure uncertainty and escalation?

Those are exactly the kinds of signals needed to evaluate, benchmark, or improve domain-specific AI systems.

The Scarce Asset Is Not the Rulebook

The internet is not short of statutes, regulations, court opinions, patent documents, contracts, or technical papers. What is harder to obtain is the judgment process of someone who has spent ten or twenty years applying those materials to real problems.

For professional AI, the scarce data may be less about what the law says and more about how an experienced practitioner decides what matters.

A résumé tells people what you have done. What can train a professional AI is often why you did it that way.

Professional Testing Is Still Legitimate

None of this means AI interviews are inherently suspect. Professional roles require professional testing. Jobs that involve training or evaluating AI should obviously test whether candidates can write high-quality answers, design evaluation rules, identify flawed reasoning, and distinguish strong evidence from weak evidence.

The concern begins when a lengthy or realistic professional exercise may serve more than one purpose: evaluating a candidate today and creating a reusable expert-data asset tomorrow.

If interview recordings, transcripts, annotations, reasoning traces, or work samples will be retained for product improvement, model evaluation, or model training, the better practice is to make those uses clear rather than leaving candidates to guess.

Questions Worth Asking Before a Long AI Interview or Expert Test

Professionals do not need to treat every interview as an adversarial process. But for a long, detailed, or project-like assessment, a few basic questions are reasonable:

  • Are my responses used only for this hiring decision, or for other purposes as well?
  • How long will recordings, transcripts, annotations, and evaluation results be retained?
  • Can any of the material be used for model training, product improvement, benchmarking, or knowledge-base development?
  • If the exercise closely resembles real project delivery, why is that level of detail necessary for evaluation?
  • Is there a way to delete the material, withdraw consent, or exercise other data-control rights?

The point is not to demand a legal negotiation before every interview. It is to understand what is being created and how it may be used.

IP Professionals Have an Additional Reason to Care

Patent and IP professionals already know not to paste client trade secrets, unpublished inventions, confidential prosecution strategy, or sensitive case files into an unfamiliar AI environment. The same instinct should extend to their own accumulated professional know-how.

A patent response is a deliverable. An FTO report is a deliverable. A design-around recommendation is a deliverable. But the reasoning that produced those deliverables—the order of checks, the judgment calls, the escalation triggers, the reasons one option was rejected—is also valuable.

Once that reasoning is decomposed into questions, answers, rankings, rubrics, and evaluation criteria, it can become training or evaluation material for increasingly capable professional AI.

Professional Experience Is Becoming a Data Asset

This is not an argument against AI recruiting or expert testing. It is an argument for clearer boundaries and greater awareness of value.

Professionals are already careful with client data because they understand confidentiality and privilege. In the AI era, they may also need to think more deliberately about another category of information: their own reusable decision-making process.

The most valuable data may no longer be just a name, phone number, or résumé. For domain-specific AI, the deeper asset is often the reasoning that explains why an experienced professional chose one path over another.

The question is not only, “What information am I giving this interviewer?” It is also, “Am I producing a one-time hiring response—or a professional data asset that can be reused long after the interview?”