DECK: *Indeed listing requires applicants to run surveillance prompt against their own machine history, instruct the machine the results are private, and then deliver them to a stranger.*
BYLINE: By Silas Vane / Business Correspondent, Slopgate
The hiring process in the United States has, over the past decade, acquired layers of procedural complexity that would have seemed implausible to the personnel directors of an earlier generation. The applicant tracking system replaced the manila folder. The behavioral interview replaced the handshake. The personality assessment, administered by third-party vendors at scale, replaced the gut feeling, or supplemented it, or existed alongside it in a state of mutual irrelevance. Each addition followed a discernible commercial logic: the vendor sold the tool, the employer purchased peace of mind, and the applicant submitted because submission is the price of employment. What has surfaced on the job board Indeed this December represents something structurally new, not because it uses artificial intelligence—half the industry does, with varying degrees of candor—but because it requires the applicant to use his own.
The listing, posted by a firm seeking personnel, directs candidates to a Google Docs page containing a prompt of considerable length. The applicant is instructed to paste this prompt into his personal ChatGPT account. The prompt then directs the machine to search the applicant's full conversation history—every query, every draft, every idle question posed at three in the morning—and to produce from this corpus a scored psychological and competency profile. The machine is told, in the language of the prompt itself, that the results are "not for any external evaluation." The applicant, having received this profile generated under false assurance of privacy, is then instructed to share a public link to the conversation with the recruiter.
The architecture of the arrangement deserves enumeration, because its elegance is of a kind that repays close attention. First, the applicant must act as operator—he initiates the surveillance, against himself, using his own account and his own history. Second, the machine must be deceived—it is told the assessment is private, a therapeutic exercise, so that its outputs will achieve the candor that a machine calibrated against human preference might otherwise soften. Third, the applicant must betray the confidence he himself has just established, delivering the results to precisely the external evaluator whose existence the prompt denies. The applicant is subject, operator, and courier. The employer touches nothing. The employer's hands are, in the legal sense that matters most, clean.
What the prompt asks the machine to score has the contours of a standard competency matrix—communication skills, problem-solving aptitude, emotional intelligence, and technical proficiency—rendered uncanny by its data source. A conventional personality assessment draws on the applicant's responses to controlled stimuli: hypothetical scenarios, Likert scales, and forced-choice pairs. This prompt draws on everything the applicant has ever said to a machine he believed was listening only to him. The difference is not one of degree. A man who answers a questionnaire in an interview is performing for an audience. A man who asks a chatbot to help him draft an apology to his landlord at midnight is not. The prompt converts the latter into the former, retroactively and without consent—or rather, with consent extracted at the moment of application, which is to say, under duress of a particular ambient kind.
The legal framework governing this practice is, at present, nonexistent in any meaningful sense. The Illinois Artificial Intelligence Video Interview Act regulates machine analysis of recorded interviews. New York City's Local Law 144 requires bias audits of automated employment decision tools. Neither contemplates an arrangement in which the tool belongs to the applicant, the data belongs to the applicant, and the analysis is generated by a third party's product at the applicant's own instruction. The employer has not deployed an automated decision tool. The employer has asked the applicant to deploy one against himself. The distinction may prove durable.
No completed assessment was available for review. The Google Docs prompt, which remained publicly accessible as of this writing, runs to several hundred words and includes instructions for formatting the output as a structured report with numerical scores. The Indeed listing did not disclose the salary range, a separate compliance matter in jurisdictions that require it, but one that acquires a certain atmospheric density in context: the applicant is asked to expose the full record of his private machine interactions before learning what the position pays.
The broader market implications are modest in scale and considerable in precedent. If the practice stands—if no regulator objects, no plaintiff sues, no platform amends its terms of service—then the machine conversation history becomes, functionally, a new class of pre-employment record. Not a credit report, which is regulated. Not a background check, which requires disclosure. Not a drug test, which is administered by a certified third party. Something closer to a diary, requested politely, with the understanding that refusal is always an option and employment is always at stake.
The slop, in this instance, is not the output. It is the workflow.