AI interview policy template for candidates

A ready-to-send policy that tells candidates exactly where AI is allowed in your interview process and where it isn’t. It’s Variant B from the Templates book of Unscripted, published here in full. Fill the brackets, paste it into your invite email.

Jump to the template Download the free policy pack

In a nutshell

Send it with the first interview invitation, and link it from the job post. Candidates should know the rules before they prepare, not when they join the call.
It splits your loop into two lanes. AI-allowed stages, where you score the judgment around the tool. Independent stages, where you see how someone reasons on their own.
It removes the grey zone. A written policy turns “did they cheat?” into “did they follow the rules we gave them?” The second question has an answer.
Keep the FAQ. It answers the questions honest candidates ask, and it makes accommodation requests a normal thing to send.
Make every stage name match your invites, and tell every interviewer which lane their stage is in.

Why do candidates need a written AI policy?

Because candidates and employers don’t agree on what counts as cheating. In a Greenhouse survey of 2,200 job seekers, only 19% of US respondents considered AI use in a live interview cheating, and 27% had never seen an employer AI policy (source, vendor data).

So if you never state the rules, you’re scoring candidates against rules they didn’t know. Honest candidates who hold back look weaker than candidates who didn’t, and your scores stop being comparable.

Stating the rules also seems to change behaviour. In one assessment vendor’s analysis of 2,000+ participants, an upfront honesty agreement cut unauthorised help from 28% to 13% (Talogy, vendor data). That’s one study, but the cost of trying it is one paragraph in an email.

Why two lanes, not a ban?

Because a ban you can’t enforce rewards the people who ignore it. Any stage you don’t supervise live is effectively AI-allowed, whether you say so or not.

The two-lane policy mirrors the job. Nobody writes production code without tools, but everyone has to explain their own decisions in a code review without a script. So you allow AI where the work allows it and score judgment there, and you keep short, conversational stages where you hear the candidate reason on their own.

Template 01 · Variant B · Two-lane policy

How we think about AI in our interviews

At [Company], engineers use AI tools every day, and some of our interview stages reflect that. Other stages are designed to show us how you reason on your own, the way you would in a design discussion or a code review. Each stage below is labelled so you always know which lane you're in.

Stage	Lane	What that means
[Stage 1: Recruiter screen, 20 min]	Conversation	No technical tools needed
[Stage 2: Technical screen, 45 min]	Independent	Please work without AI assistants. Documentation for syntax is fine.
[Stage 3: Paid work sample, [X] hours]	AI-allowed	Use any tools you like. Tell us what you used.
[Stage 4: Work sample review, 60 min]	Independent	We'll discuss your submission. Please talk us through it without AI help.
[Stage 5: Final round, [format]]	[Lane]	[One-line description]

In AI-allowed stages: use any assistant, completion, search, or reference. Include a short note on what you used and for what. We score your decisions, not whether a tool was involved.

In independent stages: please close AI assistants, including overlay tools, voice assistants, and chat windows. Standard editor features (syntax highlighting, basic autocomplete) and official language documentation for syntax lookups are fine. If you're unsure, ask the interviewer at the start. We'd much rather you ask.

In every stage: only you take part. No other person may help you or speak for you.

Why we do it this way: the independent stages are short and conversational. We're not testing memorisation; we want to hear how you think through a problem, including where you get stuck. Getting stuck and reasoning your way out is a good outcome in our scoring.

Questions about this policy? Reply to [contact email] any time before your interview.

Attach to the policy · Candidate FAQ

Can I use autocomplete?

Yes. Standard editor features such as syntax highlighting, bracket matching, and the basic autocomplete your editor ships with are fine in every stage. In [independent / live] stages, please turn off AI-powered completion (tools that suggest whole lines or functions). If you're not sure whether your setup counts, tell the interviewer at the start and they'll help.

Can I Google syntax?

Yes. Looking up syntax or an API signature in official documentation is fine, and normal. Please mention it out loud ("I'm checking the argument order for [function]") so your interviewer can follow along. What we ask you not to do in [independent / live] stages is search for the solution to the problem itself, or paste the question into an AI tool.

What if I need an accommodation?

Please ask. Common requests include extra time, a written copy of questions, a camera-off portion, captioning, breaks, or a different format for a stage. Email [contact email] with what would help. You don't need to share a diagnosis or a reason, and the request goes to [recruiting coordinator / People team], not the interview panel's scoring. If an accommodation involves a tool that would normally be outside the policy (for example, speech-to-text or a screen reader), we'll confirm that it's approved in writing so nobody is surprised.

What happens if I accidentally break the policy?

Tell us. If a tool popped up or you used something out of habit, mention it. We'll note it and carry on. Disclosure is never held against you; concealment is a different matter.

Will you record the interview?

[We record live rounds for interviewer calibration only, with your consent at the start. Recordings are deleted after [X] days.] / [We don't record interviews.]

Why do you have rules about this at all?

Because we want every candidate assessed on the same footing. Clear rules protect the candidates who follow them.

How do I use it?

Name the stages exactly as they appear in your invites. If the email says “Technical conversation” and the policy says “Technical screen”, someone will be confused. One name, everywhere.
Don’t name specific cheating tools in candidate-facing text. It reads as an accusation and dates quickly. “Overlay or interview helper tools” covers the category.
Decide the documentation rule before you send. “Official docs for syntax” is a sensible default for independent stages. If you’d rather allow full search, say so, and make sure every interviewer applies the same rule.
Route accommodation requests to someone outside the panel. Interviewers need to know what format to run, not why.
Version it. Put “[Policy version: YYYY-MM]” at the bottom. If you change it mid-hire, candidates already in the loop keep the version they were sent.
Brief your interviewers. A candidate who follows the policy should never be marked down by an interviewer who didn’t read it. State the lane at the top of every interviewer brief.

Which variant fits your team?

Variant B is my default for most teams. The kit includes two others, each with the same FAQ:

Variant	Best for	What you score	Where
A. AI allowed everywhere, with disclosure	Teams where AI tools are part of daily work	Judgment: what they asked, kept and rejected	In the kit
B. Two-lane, stage by stage	Most teams	Tool fluency in AI lanes, reasoning in independent lanes	Free, above
C. AI not permitted in live rounds	Regulated work, or a team that wants a baseline first	Unaided reasoning and communication	In the kit

The rest of the loop

A policy tells candidates the rules. The kit gives you the interview that goes with them.

Variants A and C, the matching job-post insert and candidate emails, one-page interviewer briefs with the lane on top, and laddered questions for every stage.

See the full kit Take the 3-minute audit

This template isn’t legal advice. If you record interviews, use AI to evaluate candidates, or collect biometric data, check local rules with counsel first.