AI Hiring Disclosure: Notice, Consent, and Accommodation Templates

AI hiring disclosure is the set of messages you send candidates so they know a tool is involved, what it does, and how to reach a person. Most teams treat it as legal copy pasted into a job posting footer, which is why candidates get surprised mid-process and recruiters get questions nobody owns. Treat it as product operations instead: three notice layers, one decision rule, and a log that proves what you sent.

This is an operational starting point, not legal advice. Have counsel adapt the wording for your tool, your roles, and every jurisdiction you hire in.

Start with an inventory, not a template

You can't write an accurate notice until you know what the tool actually does. Open your hiring flow and list every point where software touches a candidate's application or responses, then write down what comes out the other side.

Three categories matter, because the duties attached to them differ:

1. The tool only assists. It transcribes an interview, formats notes, or schedules. No ranking, no score.
2. The tool scores or ranks. It produces a number, a shortlist order, or a fit rating that a recruiter reads before deciding.
3. The tool makes or meaningfully drives the decision. It auto-advances, auto-rejects, or produces output a recruiter follows without independent review.

Category 3 is where people get into trouble, and it's usually accidental. A recruiter who reviews 200 AI resume screening scores in nine minutes and clicks the top 20 is not doing meaningful review, whatever the internal policy says.

Once the inventory is done, map candidate and job location. Duties depend on where the role is based and where the candidate is, so this is the point to bring counsel in rather than after the notice is live.

The minimum viable disclosure

Whatever the jurisdiction, a notice that doesn't answer these six questions will generate candidate emails:

• Which stage uses a tool, named plainly (resume review, phone screen, interview scoring)
• What the tool does and what it produces (a transcript, a score, a ranked list)
• What candidate data or responses it uses
• Whether a person reviews the output, and at what point they decide
• What choice the candidate has, and what happens if they decline
• How to ask for an accommodation, an alternative process, or a human, with a real contact and a timeframe

Here's the map filled in for a common setup. Keep your own version in the same file as your hiring workflow, not in a policy binder.

What AI hiring laws require, and what they leave to you

AI hiring laws are a jurisdiction-by-jurisdiction patchwork. There is no single national consent rule, and disclosure, consent, bias audits, data retention, human review, and accommodation are separate duties. A notice template satisfies one of them.

Two examples show how different the obligations look.

New York City's Local Law 144 covers use of an automated employment decision tool that substantially assists or replaces discretionary hiring or promotion decisions. Covered employers must ensure an annual independent bias audit and publish required information. The candidate notice has to state that an AEDT will be used, the job qualifications and characteristics it assesses, and how to request an accommodation, and it must go out at least 10 business days before use through permitted methods. Applicability turns on use in the city, so read the official FAQ's location examples rather than assuming the rule follows your headquarters.

Illinois takes a narrower slice. Under the Artificial Intelligence Video Interview Act, for positions based in Illinois, an employer that asks applicants to submit video interviews analyzed by AI must tell the applicant AI may analyze the video, explain how the AI works and the general characteristics it evaluates, and get consent before the interview. The Act also limits who the video can be shared with and requires deletion within 30 days of a candidate's request, including instructing other recipients to delete their copies. It applies to AI analysis of applicant-submitted video. Don't stretch it into a universal rule for every AI interview.

Then there's accommodation, which applies far more broadly than either statute. The EEOC's guidance on software, algorithms, and AI in employment selection is direct: employers should have a working process for reasonable accommodation requests and should not use tools that screen out people who could do the job with an accommodation. Practically, that means a monitored inbox and a named alternative, not a line saying requests are welcome.

Build the disclosure in three layers

One notice can't do all the jobs at once. Split what the notice says from when it's delivered, and the candidate experience stops feeling like a trap.

Layer 1: job-posting preview

Two or three sentences in the posting, above the apply button where possible. The goal is no surprises, not full detail.

How we screen for this role: after you apply, we use software to check applications against the requirements listed above, and we invite shortlisted candidates to a short recorded interview that our AI tool transcribes and summarizes. A recruiter reviews every candidate's responses before any decision to advance or decline. Need an accommodation or a different format? Email [email protected] and we'll reply within two business days.

Layer 2: full pre-use notice

This one is stage-specific: the full notice for a given AI tool has to reach the candidate before that tool is used on them. An earlier notice about your resume parser does not cover the interview scoring step later in the flow. Send it in the same message as the invite to that step, with enough lead time to satisfy whatever notice period applies to the role's location. Ten business days is the NYC benchmark; build your calendar around the longest window you're subject to.

Subject: Your interview for [Role] and how we review it

Hi [Name],

Thanks for applying to [Role] at [Company]. Before you start, here is exactly how this step works.

What you'll do: answer [number] questions in a recorded interview using the link below. It takes about [number] minutes and you can complete it any time before [date].

What the tool does: our screening tool, [Vendor and product name], records your answers, produces a transcript, and generates a summary scored against these criteria: [list the actual criteria, for example relevant experience with X, availability for Y shift, licensing]. It does not evaluate your appearance, accent, tone of voice, or personality.

What it uses: your recorded answers to the listed questions, plus the information you entered in your application. Nothing else.

Who decides: [Recruiter name or team] reads your transcript and the summary and makes the advance-or-decline decision. The tool does not reject candidates on its own.

Your choices: if you would rather do this step live with a recruiter, or you need an adjustment for a disability or medical condition, reply to this email or write to [monitored address]. We will respond within [number] business days and there is no penalty for asking.

Data: we keep your recording and transcript for [retention period]. To request deletion, email [address].

Questions about any of this go to [named person, real address].

Two details operators skip. Name the vendor and the actual configuration you run, not the vendor's marketing description of everything the product can do. And list the real criteria, because "assesses fit for the role" is technically true and useless.

Layer 3: consent or acknowledgment

Where consent is required, a checkbox only works if the plain-language explanation sits next to it. And one checkbox does not satisfy every law. Some jurisdictions require specific consent before AI analysis of video, some require notice with no consent step, some require published audit results you'll never capture in a form field. Keep the checkbox scoped to what it actually covers.

[ ] I have read the notice above describing how [Company] uses [Vendor and product] in this interview, what it analyzes, and who makes the hiring decision. I agree to complete this recorded interview and to [Company] using AI to transcribe and summarize my answers for the purpose of evaluating my application for this role.

You do not have to agree. If you prefer a live interview with a recruiter, or need an accommodation, choose "Request an alternative" below and we will contact you within [number] business days.

[ Request an alternative ]

Notice version [X.Y], shown [date].

Layer 4 is a reply, not a document

The accommodation route only counts if someone answers it. Give the recruiter who owns the inbox a canned reply so response quality doesn't depend on their mood.

Hi [Name],

Thanks for letting us know. We've cancelled your recorded interview invitation, so you don't need to complete it.

Instead, [Recruiter name] will run the same screening questions with you live by phone or video, whichever you prefer. Here are three times that work on our side: [times, with timezone]. Reply with one, or send times that suit you better.

The questions and the criteria we assess are identical to the notice you received. Your request has no effect on how we evaluate your application. If there is anything else that would make this step easier, tell me and we'll sort it out.

The pass, clarify, stop rule

Run this against your live flow, not your documentation, before you send another invite. Ask a colleague outside recruiting to read the notice and describe what happens to their application.

• PASS: the candidate knows which tool is used, what it does, when it acts, and how to reach a person, and the workflow matches that description.
• CLARIFY: the wording is technically true but vague ("we may use technology to assist our review"), the criteria aren't named, or the notice and the live workflow disagree on timing. Fix the wording or fix the workflow, then re-check.
• STOP: consent is required for that role's location but you aren't collecting it, the accommodation address is unmonitored or bounces, or candidates are being rejected automatically while the notice claims meaningful human review. Pause the tool at that stage until it's resolved.

The third STOP condition is the one that shows up in real audits. Someone enables an auto-reject threshold to clear a backlog, nobody updates the notice, and now your candidate communication is inaccurate on the exact point regulators care about. Any change to scoring thresholds, question sets, or the vendor's model should trigger a notice review. If you're standing up scoring at all, decide the human decision point first and write it down; the mechanics of automated interview evaluation and where a person still signs off are worth settling before the notice goes out.

Log enough to prove it

Consent stored without the notice version is close to worthless, because you can't show what the candidate agreed to. Four fields per candidate cover most of it: delivery timestamp of the notice, the notice version they accepted, the tool and configuration in use at that time, and how any request was handled and when.

The mistakes that generate complaints are mundane:

• A consent box with no plain-language explanation above it
• An accommodation address that routes to a shared inbox nobody owns
• Vendor documentation describing features you have switched off, or don't have
• Remote roles where the notice reflects your office location instead of the candidate's
• A notice written once and never revisited after three configuration changes

There's an operational argument for good disclosure alongside the legal one. An AI step a candidate didn't see coming can create surprise or hesitation mid-process, and a clear notice with a visible human decision point removes some of that uncertainty; what candidates actually think about transparency in these flows is a topic in itself, covered in whether candidates like AI interviews. If you run AI candidate screening at volume, wiring the notice into the invite sequence rather than sending it as a separate message makes it harder for the notice and the tool to drift apart. Where the human decision point sits in an AI interviewer workflow is a design choice you make once and then have to describe accurately every time.

Key Takeaways

• AI hiring disclosure starts with an inventory: name every stage where a tool touches an application, what it produces, and whether it assists, scores, or effectively decides.
• Disclosure, consent, bias audits, retention, human review, and accommodation are separate duties. NYC's AEDT rules, the Illinois video interview law, and EEOC accommodation guidance each demand something different, so get jurisdiction-specific review before you finalize wording.
• Use three layers: a short job-posting preview, a full pre-use notice naming the vendor and the real criteria, and a scoped consent or acknowledgment step where required. Each tool's full notice has to land before that tool runs, not just before the process starts.
• The accommodation route has to be a monitored address with a stated response window and a named alternative process, or the rest of the notice doesn't hold up.
• Apply the pass, clarify, stop rule against the live workflow. Automated rejection alongside a claim of meaningful human review is a stop condition, not a wording problem.
• Log the notice delivery time, accepted version, tool configuration, and request handling, and review all of it whenever the tool or the workflow changes.