LunaPrompts
A useful AI interview gives candidates clear instructions before they begin, a manageable task, a way to get help, and a visible handoff to the hiring team. For HR, candidate experience belongs in the rollout plan alongside interviewer capacity and completion rates.
If someone discovers halfway through a conversation that the interviewer is automated, the process has already created an avoidable surprise. An invitation and a support route are small details with a large role in whether the experience feels understandable.
What recent candidate research tells us
Greenhouse’s 2026 Candidate AI Interview Report surveyed 2,950 job seekers across the US, UK, Ireland, Germany, and Australia. Among US respondents who had experienced AI evaluation, 70% said AI had not been clearly disclosed before their most recent AI interview. The finding applies to that surveyed group; it is not a measurement of every employer or every AI interview. Greenhouse’s report.
For a recruiting team, the practical question is straightforward: could a candidate explain what will happen, what will be evaluated, and who to contact before clicking Start? The checklist below is our suggested way to design for that outcome.
Decide what the AI interview is for
Write one purpose statement. For example: “This first-round conversation collects evidence about how candidates approach the role’s recurring support tasks.” Then decide what belongs in a later human interview.
An early-career campaign might use an AI-led round to collect initial answers before a panel reviews work samples. A specialist role may need a much more detailed human conversation. Do not add an automated round simply because it is easy to schedule; every additional stage asks candidates for more time.
Define the evidence the team needs and the decision the round supports. Avoid an undefined output such as “overall fit” that no reviewer can explain.
Tell candidates six things before they begin
| Explain | A useful answer contains |
|---|---|
| The format | Whether the conversation is AI-led and what interaction to expect |
| The purpose | The role-related skills or tasks being explored |
| The commitment | Expected time, deadline, and time zone |
| The rules | Permitted tools, preparation, and how work is collected |
| The support route | Technical help and a way to discuss an alternative or accommodation |
| The next step | Who reviews the evidence and when the team expects to respond |
Use the wording that matches the actual deployment. Do not promise a human review, alternative route, response date, or deletion schedule that the organization has not arranged.
An invitation you can adapt
The following is an illustrative draft. Replace the placeholders and confirm each statement before using it:
As the next step for [role], we invite you to an AI-led conversation about [job-related topics]. It should take approximately [tested duration]. Please complete it by [date, time, and time zone]. You may use [permitted tools and materials]. We will collect [describe the actual answers, recording, or transcript] for [approved purpose]. [Named team or role] will review [describe the evidence and process]. If you need technical help or would like to discuss an alternative or accommodation, contact [support route]. We expect to update you by [date or stated process].
Put the important information in the invitation, not only behind a policy link. A short practice interaction can also help a candidate understand the format before the assessed questions begin, if your configured workflow supports it.
Design questions around something the candidate can explain
Start with the role. For a campus support-engineering role, an example question might describe a customer report with missing information and ask what the candidate would clarify first. A follow-up could introduce a new error message and ask how the next step changes.
Before launch, have subject-matter experts answer the questions and identify what evidence is actually observable. Remove prompts that depend on unstated company knowledge. Agree how reviewers distinguish an incomplete answer from an answer that demonstrates a misconception.
For a technical assessment followed by an interview, avoid repeating every task. Use the submitted work to identify where a conversation would add evidence. The AI Viva guide explores that handoff.
Make the interrupted session part of the design
A candidate’s connection can fail. A microphone may not work. A question may be unclear. Decide what happens before those situations arrive in a real campaign.
Give support staff a simple way to identify the session, understand the problem, and follow the approved restart or rescheduling process. Tell candidates which route to use. Review technical interruptions separately from performance so that an operational failure does not silently become a poor score.
If a candidate needs an alternative or accommodation, route the request to a person with the authority to arrange it. Avoid asking the automated interview to negotiate an exception it cannot actually provide.
Define the human review handoff
A report is useful only if the next reviewer knows what to do with it. Agree which evidence they receive, what uncertainty looks like, and how to record a correction or disagreement.
For each candidate, the reviewer should be able to distinguish an observed answer, a generated summary, and their own conclusion. Where the report points to a particular response, the team should have an appropriate way to inspect the underlying evidence in the configured workflow.
Do not substitute vocal confidence, accent, or a vague personality impression for job-relevant criteria. Use the same agreed rubric across comparable candidates, and discuss reviewer disagreements before broadening the campaign.
Pilot one role before a campus-wide rollout
Start with internal testers, then a small, approved hiring cohort. Check the invitation, actual completion time, question clarity, support process, and reviewer usefulness. Include a deliberate connection interruption in the internal test.
Choose the conditions for expansion in advance. For example, the team may require that support can resolve the tested failure cases, reviewers can explain their decisions, and candidates receive the information promised in the invitation. Set thresholds appropriate to the workflow rather than copying another employer’s numbers.
LunaPrompts offers AI interview agents for screening and campus or high-volume campaigns, alongside practical assessments and candidate reporting. In a pilot, ask to see your question set, the candidate experience, available support controls, and the reviewer handoff together. Confirm what is available in your configured environment. The campus hiring field guide covers the assessment workflow in more detail.
Measure experience as well as throughput
| Measure | How to define it | What to investigate |
|---|---|---|
| Invitation-to-start rate | Unique candidates starting divided by valid invitations | Clarity, delivery, and relevance |
| Start-to-completion rate | Completed sessions divided by started sessions | Friction, duration, and interruptions |
| Support contact rate | Candidates contacting support divided by invited candidates | Confusing instructions or technical issues |
| Time to human review | Time from completion to the agreed review step | Reviewer capacity and handoff delays |
| Candidate feedback | Responses to a short, voluntary experience question | Recurring problems and unmet expectations |
Use one defined cohort and reporting period. Keep undelivered invitations, withdrawals, and technical failures visible. A high completion rate does not by itself establish a fair process, a useful assessment, or better hires.
Ask candidates one focused question, such as “Was anything about this stage unclear?” Read the answers alongside the operational measures. A low response rate should limit how broadly you interpret the feedback.
Questions HR teams ask
Should candidates know the interviewer is AI?
Tell them clearly before the session and explain how the stage works. Match that explanation to the real process, including what the hiring team reviews and how to get help.
Should the AI interview make the final decision?
Design the round to provide specific evidence within an accountable hiring process. Define the human review and decision responsibilities before rollout; do not leave them implicit in a generated score.
What if candidates also want to use AI?
State the rule for this stage. An AI-fluency exercise may deliberately allow assistance, while a different task may require unaided work. The interview integrity review guide explains how to handle uncertainty fairly.
How do we get started with LunaPrompts?
Bring one role, an outline of the interview loop, and the questions the team needs answered. Use the assessment templates to prepare the discussion, then plan a limited pilot with a clear reviewer handoff.