The Black Box Candidates Dread: How Small Hiring Teams Can Show Their Work

Call it 'How we hire.' Put it on your careers page, link it in every job ad, and paste the URL into your auto-reply. It should be short enough that a candidate reads it on a phone while waiting for coffee. Plain language, no HR jargon, no 'talent acquisition ecosystem.'

Name each step in order. Something like: application, 20-minute screening call with me, a short paid exercise, a 60-minute interview with the hiring manager, reference check, offer. Give a realistic timeline for each step, not a fantasy one. If you know you take five working days to respond after an application, say five working days.

Then state exactly where automation is involved. If you use an AI tool to rank applications, say so. If you use a scheduling tool like Calendly to book calls, that's automation too, and it's fine to name it. The point is that nothing happens to a candidate's application that they don't know about.

Finally, name where a human reviews. Not 'a human is in the loop' as a vague reassurance, but the actual moment: 'Every application is read by a person before any rejection is sent.' That sentence alone will do more for your reply rate than a redesigned careers page.

Not every transparency statement carries the same weight. Based on what candidates ask about and what regulators require, three disclosures do most of the work.

First, disclose AI use at the application stage, not after rejection. GDPR requires candidates to know when automated decision-making is used, and telling them early builds trust before it's tested. A single line in the job ad and on the application form is enough: 'We use an AI tool to help us match applications to the role. A person reviews every shortlist before anyone is contacted.'

Second, explain the criteria. Candidates don't need your prompt or your model weights. They need to know what you're actually screening for. 'We look for three years of experience with Python, a portfolio link, and evidence you've shipped to production.' That's an explanation. It also forces you to define criteria before the tool ranks anything, which is the step most small teams skip.

Third, give a human review right. Say it in plain words: 'If you want a person to review your application, reply to this email and we'll do it.' Under GDPR and the EU AI Act, candidates have the right to request human review of automated decisions. Offering it proactively costs you almost nothing and signals that you're not hiding behind the tool.

Here's a starting point. Adapt the specifics, keep the structure.

How we hire at [Company]

We're a small team. Here's exactly what happens when you apply, so you know what to expect.

Step 1: Application (day 0). You apply through [Greenhouse / Ashby / Workable]. We use an AI tool to help us sort applications against the role criteria. A person reads every application before any rejection is sent.

Step 2: Shortlist review (within 5 working days). One of us reviews the shortlist and the AI's reasons for each candidate. We don't reject anyone on the AI's recommendation alone.

Step 3: Screening call (20 minutes). A conversation with [name], not a bot. We'll ask about your experience and what you want to do next.

Step 4: Skills exercise (paid, 2 hours). We send a short exercise based on real work. You'll get written feedback within a week either way.

Step 5: Interview (60 minutes). Structured questions, same questions for every candidate, scored against the criteria above.

Step 6: Decision and offer (within 3 working days of the final interview). We'll tell you either way. If we say no, we'll tell you why.

Your rights. You can ask us what data we hold, ask for a human review of any automated decision, or ask us to delete your data. Email [address] and we'll respond within 30 days.

That's the whole thing. It fits on one screen.

The most common failure I see is the silent rejection. A candidate applies, hears nothing, then gets a generic 'we've decided to move forward with other candidates' three weeks later. They assume a machine did it. Sometimes they're right.

Another failure is the unexplainable shortlist. A founder uses a tool, gets a ranked list, and can't say why candidate four beat candidate seven. If a candidate asks, or a regulator asks, there's no answer. Keeping records of the criteria and the human interventions at each step is not bureaucracy, it's the thing that lets you defend a decision.

Bias is the third risk. If your AI is trained on historical hiring data, it can replicate past patterns. Regular audits, checking selection rates by group, and documenting what you changed are the basics. Small teams can do this with a spreadsheet and an hour a quarter.

None of this requires a compliance department. It requires someone to write the page and someone to check that reality matches it.

If you're a founder or a boutique recruiter running a search without a recruiting team, the practical problem is usually time. You don't have an afternoon to write a transparency page, and you don't have a compliance lead to check it.

Moments AI turns a plain-words role brief into a shortlist of candidates with the evidence behind each score. That evidence is the part that matters here: it gives you something concrete to put in your 'how we hire' page and something to point to if a candidate asks why they weren't shortlisted. It also writes first emails from your own mailbox and books the intro call when a candidate says yes, which means the human moments in your process stay human.

It won't write your transparency page for you, and it shouldn't. That page is a statement about how your team works. But it makes the 'where is AI used' section easy to fill in honestly, because you can see what the tool actually did.

A page is only useful if it's true. Review it every quarter, or every time you change a tool. If you switch from one ATS to another, update the name. If you stop using AI screening for a role, say so.

Ask candidates what they thought. A two-question survey after each hiring cycle, one about communication and one about transparency, will tell you more than any vendor deck. Most people will answer. Some will tell you exactly where the process felt like a black box.

And when a candidate asks for a human review, do it. Quickly. The whole point of publishing the page is that you're making a promise. Keeping it is the transparency.