Five Questions Every HR Leader Should Ask Before Trusting Their AI Video Interview Tools
By Tatiana Teppoeva, PhD. AI Hiring Risk Consultant, Human Signal Advisory™
AI video interview tools promised HR leaders faster screening, more objective evaluation, and better hiring outcomes. Many organizations adopted them in good faith. But as regulations tighten and candidate complaints increase, a more urgent question is emerging: can you actually explain and defend what your tools are doing?
To attract the strongest candidates, stay compliant with emerging AI hiring regulations, and actually rely on your AI video interview tools and process with confidence, you need to be able to answer these five questions.
1. Why did this candidate receive this score?
AI video interview tools score behavioural signals: speech pace, vocal consistency, facial expression patterns, keyword alignment, response structure. They produce confident outputs. What many organizations still struggle to obtain is a clear explanation of which signals drove the score, how those signals were weighted, or whether the scoring model was validated for your specific roles and candidate population.
When a candidate asks why they were rejected, “the algorithm said no” is not a defensible answer. Emerging regulations such as Colorado’s AI Act increasingly push organizations toward being able to explain and document AI-assisted hiring decisions in job-relevant terms. If your team cannot do that today, the gap is already present.
2. What is your tool actually measuring and does it predict job performance?
AI video interview tools measure what they can quantify: vocal pace, keyword frequency, eye contact with the camera, facial expression consistency. What most tools cannot reliably measure: judgment under pressure, leadership capability, genuine expertise in context, or how a candidate actually performs once hired.
The critical question is whether given scores predict success in your specific roles, with your specific candidate population. Most vendors cannot answer that question for your organization specifically. You need an independent evaluation for that.
3. Are candidates whose communication patterns differ from the baseline being systematically filtered out?
AI video screening tools are trained on existing data. That data reflects existing patterns, which means candidates whose communication style differs from the algorithmic baseline may score lower not because they lack capability, but because the tool was never designed to recognize their signals as competent.
This includes neurodivergent professionals, non-native speakers, and candidates with disabilities. It also includes highly experienced candidates who communicate with deliberate pacing, direct answers, and minimal performance.
If your organization cannot identify which candidate profiles your tool systematically scores lower, you cannot objectively assess your legal exposure under Title VII, the ADA, or state AI hiring regulations.
4. Is your human review layer doing real work or rubber-stamping AI scores?
Many organizations have some form of human review in their AI screening process. Far fewer have a structured interpretation process.
A genuine human review layer means a qualified reviewer actively evaluates what the AI scored and asks whether that score reflects job-relevant capability.
Did a lower communication score reflect lack of knowledge or nervousness?
Did unusual speech patterns indicate poor communication skills or a non-native speaker?
Did the tool react to background noise or camera quality rather than candidate behavior?
Recent research from the University of Washington found that people often follow AI recommendations even when those recommendations contain measurable bias. If your reviewers are approving AI outputs without a structured framework for questioning them, you have only the appearance of human oversight.
5. What documentation do you have to explain your decisions?
When a candidate challenges a screening outcome under federal discrimination laws or emerging state AI regulations, your organization needs to produce more than the AI score. You need documentation showing what was measured, evidence that those measures are relevant to the role, records of meaningful human review, and a clear explanation of why the final decision was made.
Most organizations can reproduce the output of the tools but not the reasoning behind it. That is the real gap where legal exposure accumulates.
What to do if any of these answers are uncertain
These five questions reveal three categories of risk: signal gaps in what your tools measure, compliance gaps in what you can document and defend, and infrastructure gaps in how your tools are configured and your human review process is designed.
Independent evaluation can help identify gaps that may not be visible through vendor-provided validation studies alone.
If you want to check whether your process has any of the 12 structural gaps that turn high-volume hiring into high-risk outcomes, you can download a free diagnostic guide at tatianateppoeva.com.
If any of these questions land close to home and you would like to talk through your specific situation, I offer a free 30-minute strategy call. I will ask you some questions and share what I am seeing: what your AI video interview tools are likely scoring and what they are probably missing, where your screening process may be creating legal exposure, and whether your tool setup and human review process are actually serving your hiring goals.
Tatiana Teppoeva, PhD is an AI Hiring Risk Consultant and founder of Human Signal Advisory™, helping HR leaders identify where their AI video screening tools are filtering out strong candidates, creating legal exposure, and operating beyond what the organization can explain or defend. She spent 17 years building predictive AI systems at Microsoft and Boeing, holds a U.S. patent related to predictive AI systems, and a Harvard MS in Data Science. She also holds four certifications in nonverbal communication, the human signal expertise that makes her AI hiring audits very different from a purely technical evaluation.
Learn more at https://tatianateppoeva.com