AI in Interviews: Scheduling, Structured Questions and Interview Intelligence

By Kevin Kim · Updated Oct 11, 2026

AI in interviews is the use of software to handle the logistics and record-keeping around a hiring conversation (finding times, building a consistent question set, transcribing the call, drafting a summary against a scorecard) while the interviewer and the hiring team keep the judgment about who moves forward. The useful versions make interviews more consistent and less admin-heavy. The risky versions try to read emotion, personality or honesty from a face or a voice, and those are where the law is tightest.

This guide is for talent acquisition leads, HR managers and hiring managers deciding what to automate around interviews, what to leave alone, and what to tell candidates. It sits in the AI in recruiting section of the AI for HR collection. It is general information, not legal advice, so involve counsel for the places where you hire.

Key Facts

• Structured interviews have a mean validity of .42 for predicting job performance in the updated meta-analytic matrix by Sackett and colleagues, down from .48 in an earlier matrix, with an 80% credibility interval of .24 to .66 (Berry, Lievens, Zhang and Sackett, updated meta-analytic matrix).
• The EU AI Act's ban on inferring emotions in the workplace has applied since 2 February 2025 (AI Act, Article 5), and the Commission's rules for high-risk employment systems apply from 2 December 2027 (European Commission).
• Under Illinois' AI Video Interview Act, in force since 1 January 2020, an applicant can ask for a video interview to be deleted and the employer has 30 days to delete all copies, including those held by third parties (Baker Sterchi).
• The FBI warned on 23 January 2025 that North Korean IT workers have used AI and face-swapping technology in video job interviews to conceal their identities (FBI IC3).

What AI in Interviews Actually Does

Four jobs account for nearly everything sold under this label.

Scheduling. The tool reads interviewers' calendars, offers the candidate a set of slots, books the room or video link and sends reminders. It's the lowest-risk use. It touches no judgment about a person, only logistics.

Question and scorecard support. A model drafts interview questions from the job's written requirements, tags each question to a competency and builds a scorecard with defined rating anchors. A person approves the set before anyone is interviewed.

Transcription and note-taking. The tool records the conversation (with consent), produces a searchable transcript and drafts a summary organized by scorecard competency. Interviewers can stop typing and listen. This is the Meeting Intelligence pattern in the ACE Framework, where audio is ingested, analyzed and turned into notes and tasks.

Interview intelligence. Some platforms go a step further and analyze transcripts across many interviews: how long interviewers talk, whether every candidate got the same questions, how ratings differ between interviewers. Used on your own team's behavior, it's a coaching and consistency tool. Used to rank candidates automatically, it becomes a screening tool and picks up the legal duties of one.

In ACE terms, the first three jobs mostly Ingest (calendar and audio data), Analyze (transcribe, tag) and Generate (summaries, questions), with Execute limited to booking and reminders. Scoring a candidate on top is a different kind of step, and a different level of risk.

Why Structure Matters More Than the Tool

The strongest argument for AI around interviews isn't speed. It's consistency. A structured interview asks every candidate the same job-related questions and scores answers against defined anchors, and the research behind it is the reason to invest in the process before the software.

The Sackett and colleagues update puts the average validity of structured interviews at .42, the highest of the selection methods in the matrix, though the credibility interval is wide (.24 to .66) because structured interviews can be designed well or poorly (Berry et al.). The same paper notes the earlier matrix had put the figure at .48, because older range-restriction corrections were inappropriate. The practical reading: structure helps, but the benefit depends on how well the questions and rating anchors are built.

That's where AI earns its place. It can draft a question bank per role in minutes, keep interviewers on the script, and make notes consistent so the debrief compares like with like. It can't fix a vague scorecard. If your team can't say what a "4" looks like on a competency, a model will invent a definition, and different interviewers will get different ones.

Real Examples by Function

Sales hiring. A sales director hiring account executives builds a scorecard for four competencies: discovery, handling objections, forecasting honesty and coachability. The tool drafts two behavioral questions for each, the director edits them, and every candidate gets the same set. After each call, the summary lists evidence under each competency, and the interviewer rates it before seeing anyone else's score.

Engineering hiring. A technical interview is paired with a recorded work-sample discussion. The transcript lets a second reviewer assess the reasoning without sitting in the call, which cuts the "I liked them" effect. Interviewers still decide.

Customer support and operations. High-volume roles often use scheduling automation alone. Candidates pick a slot from a link, and the coordinator stops spending hours on back-and-forth. This is usually the best first step for teams with many interviews and few recruiters.

Executive and senior roles. Here the value is in the note-taking, not in scoring. Panel members compare written evidence across a long process with many interviewers.

Internal moves. Interviews for promotions or transfers use the same question banks, which makes internal and external candidates easier to compare. For skills-led approaches, see skills-based hiring.

Consent, Recording and the Law

Recording a candidate is a legal event before it's a product feature. Three layers apply, and the strictest one that fits your candidate usually wins.

Recording consent

In the US, federal law makes it lawful to intercept a communication when you're a party to it or when one party has consented, unless it's done to commit a crime or tort (18 U.S.C. § 2511). States can be stricter. Florida's statute, for example, allows interception when all parties to the communication have given prior consent (Fla. Stat. § 934.03). California's Penal Code § 632 prohibits recording a confidential communication without the consent of all parties (Cal. Penal Code § 632).

The practical consequence: when the candidate and interviewers sit in different states, you can't assume the most permissive rule applies. Ask for consent at scheduling, state it again at the start of the call, and record the consent. Don't record candidates who say no, and offer an alternative such as live note-taking.

Illinois: the AI Video Interview Act

Illinois regulates the case where employers ask applicants to record video interviews and use AI to analyze them, for positions based in Illinois. Before asking for the video, the employer must tell the applicant that AI may be used to analyze it and assess fitness for the position, explain how the AI works and what general characteristics it evaluates, and get consent. An employer may not use AI to evaluate applicants who haven't consented (820 ILCS 42/5). The Act took effect on 1 January 2020. Videos may be shared only with people whose expertise is needed to evaluate the applicant, and an applicant who asks has the video deleted within 30 days, including copies held by third parties (Baker Sterchi). Commentators note the Act doesn't define "artificial intelligence" or say how detailed the explanation must be, so write a plain-language notice and keep it.

New York City: bias audits

If the tool scores, ranks or screens candidates, New York City's Local Law 144 applies to employers using it for NYC roles. The tool must have had a bias audit within one year of use, the audit information must be public, and candidates must receive notice, which the city's revised guidance puts at 10 business days before use. Enforcement began on 5 July 2023 (NYC Department of Consumer and Worker Protection). A transcript-and-summary tool that doesn't score anyone sits differently from one that rates answers, so ask counsel where yours falls.

EU: emotion recognition is banned, employment AI is high-risk

Article 5 of the EU AI Act prohibits AI systems that infer a person's emotions in the workplace and educational institutions, except for medical or safety reasons. That prohibition has applied since 2 February 2025 (AI Act, Article 5). Analysis of the Commission's guidelines says hiring processes fall within the workplace context, so an interview tool that reads emotion from faces or voices is in scope (Future of Privacy Forum). The Commission also classes AI tools for hiring and managing workers as high-risk, with those rules applying from 2 December 2027 (European Commission). If you hire in the EU, don't buy a feature whose pitch is "detects confidence" or "reads sentiment from expressions."

The vendor precedent

The clearest real-world case is HireVue, which said in a press release dated 12 January 2021 that it had removed visual analysis from its new assessments, because language analysis had grown more predictive and the visual component no longer added significant value (HireVue). The lesson for buyers is that facial analysis was dropped by the company that sold it, so a tool still offering it deserves hard questions.

Comparing the Approaches

How to Implement It, Step by Step

1. Write the scorecard first. Define the competencies, the questions and what each rating means, in plain language. Tools should mirror this, not replace it.
2. Start with logistics. Automate scheduling before anything touching candidate content. It builds trust in the tool at no legal cost.
3. Add structured questions. Let the model draft the bank, then have the hiring manager approve and lock it for the role.
4. Decide your recording policy. Say when you record, who sees it, how long you keep it and how a candidate declines. Default to all-party consent everywhere, which also satisfies the stricter states.
5. Write the candidate notice. Explain in plain words what's recorded, what the AI does with it, what it doesn't do, and who reviews. Send it with the invitation.
6. Keep summaries as drafts. The interviewer reads the transcript or checks the summary against it, and signs the ratings. This is human-in-the-loop in practice, and it has to be a real step.
7. Check for bias. Compare pass rates and ratings across groups, and across interviewers. See bias in AI for where it comes from.
8. Set retention and deletion. Build a way to delete a candidate's recording and every copy, including the vendor's.
9. Pick tools last. When you're ready, the best video interview software comparison covers the options.

Handling AI-Assisted Candidate Cheating

Candidates now have AI too. Real-time tools can listen to a question and feed a candidate an answer, and in the worst cases AI is used to fake who the candidate is.

The documented risk is identity fraud rather than clever answers. The FBI's Internet Crime Complaint Center warned on 28 June 2022 about complaints involving deepfake audio and video in remote-work interviews, noting signs such as lip movement that doesn't match the audio and coughing or sneezing that doesn't match what's on camera (FBI IC3). On 23 January 2025 it said North Korean IT workers have used AI and face-swapping in video interviews, and advised verifying identity during interviewing, onboarding and employment, checking HR systems for applicants with matching resume content or contact details, and doing as much of hiring and onboarding in person as possible (FBI IC3).

Most "how many candidates cheat" statistics come from vendors who sell detection tools, so treat them as directional and don't build policy on a single percentage.

What works in practice:

• Say what's allowed. Tell candidates whether AI assistance is permitted in each round. If you're testing how someone works with AI, say so. If you're testing their own reasoning, say that.
• Ask follow-ups. Structured questions with probes ("what was your specific role?", "what would you do differently?") are hard to feed in real time and easy to tell from rehearsed text.
• Verify identity at a defined point. Check a government ID against the applicant before an offer, or at an in-person final round for sensitive roles.
• Use work samples. Live exercises where the candidate explains their thinking test the skill you're hiring for.
• Don't treat a flag as proof. A detection tool's signal is a reason to ask more questions, not a reason to reject. Look for a person's explanation before deciding.

Common Pitfalls

• Buying emotion or personality scoring. It's banned in EU workplaces, and a tool that infers traits from a face or voice is hard to justify to candidates anywhere.
• Recording without a plan for consent. A candidate in an all-party consent state, interviewed by someone in a one-party state, creates exposure for you.
• Letting the summary stand in for the interview. Summaries drop nuance and sometimes state things the candidate never said. Reviewers should be able to open the transcript.
• Skipping the scorecard. AI on top of an unstructured interview just produces a tidy record of an inconsistent process.
• Keeping recordings forever. Retention without a purpose is a liability, and deletion requests need a working path.
• Hiding the AI from candidates. Even where no statute requires notice, surprise erodes trust. The related AI candidate experience guide covers how to tell candidates.

What to Measure

Track both efficiency and fairness. Time from application to scheduled interview, interviewer hours spent on notes and the share of interviews following the approved question set show whether the process is working. Rating spread between interviewers, pass rates by group and the share of summaries corrected by interviewers show whether it's working fairly. Candidate feedback tells you whether the experience got better or just faster.

Where to Go Next

If you do one thing this quarter, write a structured scorecard for your highest-volume role and automate its scheduling. Then add transcription with a consent step, and review it at the next hiring cycle. For the rest of the hiring funnel, see AI resume screening, and for the function-wide view, AI in HR.