Why you should ask any interview integrity tool for its false-alarm rate

A useful answer gives counts: how many honest sessions were tested, how many were flagged, under which conditions, and when.

This guide explains the two numbers every check has, what a good answer looks like, how to read small samples, and the questions to put to any vendor, including us.

Every check has two numbers

• Catch rate: of the sessions where something was really wrong, how many did the check find?
• False-alarm rate: of the honest sessions, how many did the check flag anyway?

The two pull against each other. Make a check stricter and it catches more, and flags more honest people. Make it looser and it flags fewer honest people, and misses more. A catch rate quoted without a false-alarm rate leaves out half the picture.

Why false alarms cost so much in hiring

• Good candidates are lost. A flag on an honest candidate can end a process that should have gone on.
• Interviewers stop trusting the tool. After a few wrong flags, people ignore the real ones.
• Some people are flagged more than others. Checks that read signs, such as eye movement or pauses, react to nervous, careful and neurodivergent candidates, and to people speaking a second language.
• It is unfair. A result that sounds like an accusation is hard to take back.

What a good answer looks like

1. Counts, not just percentages. "0 of 30" and "0 of 3,000" are both 0%, and they mean very different things.
2. The conditions. Which devices, which apps were open, live calls or recordings, how many different people.
3. A date. Tools and the things they look for change month to month.
4. Who tested it. The vendor's own testing is useful. Say so plainly.
5. What happens in between. Does the check have a neutral result, such as "Can't confirm", or is every session either cleared or flagged?

How to read a small sample

A simple rule helps: if a test shows 0 problems in n tries, the true rate could still be as high as about 3 in n (the "rule of three", at 95% confidence). So 0 false alarms in 30 tries is good news that still allows up to about 1 in 10. 0 in 300 allows up to about 1 in 100. Ask how many tries sit behind any zero, and ask for more tries where it matters to you.

Questions to ask any interview integrity tool

1. What is your false-alarm rate on honest candidates, as a count, and how many sessions is it based on?
2. Were those sessions live calls or recordings? How many different people?
3. Were everyday apps open during the honest sessions, such as chat, music, notes or a second monitor?
4. What does the interviewer see when the tool is unsure?
5. Does a result name what was found, with proof, or is it a score?
6. Does a flag depend on behavior such as eye movement, pauses or typing?
7. When was this tested, and will you share the method?
8. Can a candidate decline, and what does the host see then?

Our own numbers, false alarms included

Here is how Verified Call answers these questions. All of it is our own testing; the method and conditions are on the Research page.

Verified Call checks facts on the candidate's computer and in the call, such as hidden AI tools, camera software and a change of face or voice, rather than reading behavior. When something cannot be confirmed, the host sees "Can't confirm", not an accusation. Results are advice for the host, not a verdict.

For a wider view of the field, see our comparison of cheating detection tools.