Publish Your Candidate-AI Stance Before the Next Interview Starts

Workflow area: Hiring & Performance · Tier: Core · Hands-on time: ~30 minutes

Ask five of your interviewers what happens if a candidate uses AI in the screen, and you'll get five answers — which means your loop currently has five policies, all unwritten. Candidates are in an impossible position: the tools are their daily workflow, some companies encourage them, some silently fail anyone suspected of it, and almost nobody says which they are. The unfairness runs both directions — honest candidates disadvantage themselves by asking, and interviewers freelance detection judgments they're unqualified to make. A one-page written stance fixes both, and pb-18's round design gives you the vocabulary for it.

When to use — and when not to

• Use it before any hiring cycle with AI-ambiguous roundswhich is every cycle now; the policy should ship alongside the redesigned loop from pb-18.
• Use it when an interviewer flags "I think they were using something"a suspicion with no policy behind it produces either an unfair rejection or an unexamined hire; the policy decides in advance which evidence matters and who judges it.
• Use it as employer-brand signala published, confident AI stance tells candidates you've thought about the actual job; silence tells them you'll improvise their fate.
• Skip a detection-first postureproctoring software, gaze tracking and accusation workflows measure stealth and punish nervous honest people; if your instinct is surveillance, redesign the rounds (pb-18) until covert AI stops helping.
• Skip a policy you won't enforce evenlya stance applied to campus hires but waived for the staff-engineer referral is worse than no policy; it's documented inconsistency.

The recipe

1. 1.

   Decide the stance per round, from the round's purpose. Pull the loop design from pb-18 and fill a simple matrix: for each round — AI allowed (which tools), AI required (it's the skill being tested), or AI-free (and why). The "why" column is mandatory; it becomes the candidate-facing language.

2. 2.

   Draft the policy with AI, covering the three hard cases.

   Copy

   Draft a candidate-facing AI-use policy for our interview loop.
   Round matrix: [PASTE: round → allowed/required/AI-free → why]
   
   The policy must state, in plain language a stressed candidate
   skims in 2 minutes:
   1. PER-ROUND RULES — what's allowed in each round, including
      whether our platform provides the AI tool or they bring
      their own, and what "allowed" covers (autocomplete? chat?
      agents?).
   2. DISCLOSURE — when they must say they're using AI, and our
      promise: disclosed use within the rules never counts
      against them.
   3. COVERT USE — what happens if AI is used where prohibited:
      who reviews the evidence, the candidate's chance to
      respond, and the outcome range. No automated rejection.
   4. ACCESSIBILITY — assistive-technology use is not "AI use"
      and is handled by accommodation, not this policy.
   Tone: confident, not paranoid. We hire people who work with
   AI; the rules exist so every candidate is judged the same way.
   

3. 3.

   Pressure-test the draft against real scenarios. Same conversation: "Test this policy against: (a) candidate uses AI autocomplete out of habit in an AI-free round and self-reports immediately; (b) interviewer suspects covert chat use but has no evidence beyond pace; (c) candidate asks mid-round whether they may use AI. Does the policy give the interviewer a script for each?" Good output exposes the gaps:

   Copy

   Scenario (b) FAILS under the current draft: the policy says
   covert use "may end the process" but gives the interviewer no
   standard of evidence. Suggested fix: interviewers never
   adjudicate live — they note observations in the scorecard;
   the hiring manager reviews, and pace alone is never evidence.
   

   That fix — observations in the scorecard, adjudication by the hiring manager, named evidence standards — is the heart of a fair covert-use process. Adopt it.

4. 4.

   Configure the platform to make the rules physical. Wherever tooling can enforce the matrix, let it: enable the interview platform's AI mode for AI-allowed rounds (pinned model, logged transcript), disable copy-paste or external tools only where the round is declared AI-free, and make the platform's behavior match the published policy exactly. A policy the platform contradicts trains candidates to ignore the document.

5. 5.

   Publish, brief, and version it. The policy goes in the candidate-prep email before the first technical round — not buried in a portal. Brief every interviewer on their script for the three scenarios, and version the document: model capabilities and your loop both change, so the policy carries a date and an owner, reviewed each cycle alongside pb-18's single-prompt audit.

Make it yours

• Small team or startup: a half-page in the interview-invite email is enough — per-round rules, the disclosure promise, and "if something's unclear, ask; asking is never penalized." Your advantage is that all three interviewers can actually agree in one meeting.
• Large org or GCC: route the covert-use section through HR and legal — outcome processes touching candidate rejection need their sign-off, and some jurisdictions regulate automated hiring decisions. Standardize one policy across the site; per-team variants guarantee inconsistent candidate treatment and a discrimination-adjacent audit finding.
• Hiring for client-facing or regulated seats: if client contracts restrict AI use on their work, say so in the policy ("this role's day-to-day AI use is governed by client agreements") — candidates deserve to know the job's real constraints before they optimize for your AI-allowed round.

Quality check

• Every round's rule has a stated reasonrules with reasons get followed; bare prohibitions get tested.
• The disclosure promise is realaudit a cycle's scorecards: if disclosed, in-policy AI use shows up as a negative comment anywhere, interviewers haven't internalized the policy and candidates will learn to hide.
• No interviewer adjudicates covert use aloneevidence goes to the hiring manager, the candidate gets a response opportunity, and pace or nervousness is never evidence.
• Platform settings match the published documentclick through each round's configuration yourself; drift here is silent policy repeal.
• The policy reached candidates before the roundcheck the prep-email template; a policy candidates first hear about when accused is not a policy, it's a trap.