Compare
How to choose an interview integrity tool.
Most of these products sound identical in a demo. They are not. One question separates almost all of them, and it has nothing to do with how the technology works.
Does it output a number about a person, or a record of what happened? That single question decides what you can show a candidate who disputes the result, who actually makes the decision, and which compliance obligations you take on.
The difference
Two very different things to buy.
A number about a person
The tool watches the candidate and estimates how likely it is that they cheated. It reports a score, a confidence level, a risk band, or an accuracy percentage.
- You are told a conclusion, not shown what happened
- The number reaches your team before they decide
- You cannot explain to a candidate how it was reached
- It is an automated assessment of a person, which regulators treat differently
A record of what happened
The tool records the observable facts of the session, timestamps them, and hands them to a person. No estimate of guilt is produced.
- You can show a candidate the same record you saw
- Nothing is concluded before a person reads it
- The reasoning is the record, so it can be explained
- No automated assessment of the person is generated
Why it matters
Three reasons a number is the harder thing to own.
It misreads people who differ
Tools that infer cheating from gaze, eye movement, posture, response timing or typing rhythm are reading behaviour. Behaviour varies for reasons that have nothing to do with honesty.
A 2019 complaint to the US Federal Trade Commission argued that eye tracking in hiring could disproportionately affect candidates with disabilities, naming visually impaired applicants whose eyes do not move the way a sighted applicant's do. The same concern reaches autistic candidates, people with facial paralysis, and people with motor or speech conditions.
After an independent audit, HireVue stopped using facial analysis in its assessments in 2021. A tool sold to you today on gaze tracking is built on the approach the largest vendor in the category stepped away from.
A score decides, even when a person clicks
Vendors answer this with "a human is always in the loop." Ask when the human enters it.
If an automated score reaches the decision-maker before they decide, the score shaped the decision and the click only recorded it. A human reviewing a number is not the same as a human reading a record. The first inherits the machine's conclusion, the second forms their own.
So the presence of a review step is not the question. The question is what the reviewer is looking at when they start.
A score brings obligations with it
Tools that score, rank or screen candidates are increasingly regulated as automated employment decision tools, and the obligations land on you as the employer, not only on the vendor.
- New York City Local Law 144. An independent bias audit within the past year, a public summary of it, and advance notice to candidates.
- Illinois AI Video Interview Act. Notice, an explanation of how the technology works, and candidate consent.
- EU AI Act. AI used in recruitment and selection is classed as high risk, with risk assessment, bias testing, documentation and human oversight obligations phasing in.
Other jurisdictions are adding comparable rules. Employment, disability, privacy and biometric obligations depend on the jurisdiction and on how a product is used, so treat this as a prompt for your counsel rather than as advice.
The market
Four kinds of product, often shown side by side.
Buyers are frequently comparing products that do different jobs. Sorting them first makes the shortlist much shorter.
1. Add-ons that watch an interview held somewhere else
A bot that joins your video call, an agent the candidate installs, or a browser link that runs checks during the session. They are quick to adopt, and they inherit one limit: they can only see their own layer, because the interview is happening in a tool they do not own.
2. Tools that infer cheating from behaviour
Gaze, posture, pauses, typing rhythm and voice patterns, resolved into a suspicion score. This is the approach with the disability and explainability problems described above.
3. Tools that match software against a list
A scanner checking running processes against a maintained list of known cheating tools. It works until a tool is renamed, forked or written privately, at which point the list is the limit.
4. Screening and assessment platforms
Products whose main job is conducting or scoring interviews at volume, where integrity is one feature among many. Worth evaluating on their own merits, but they are answering a different question from an integrity tool.
Due diligence
Three questions to ask any vendor, including us.
Every vendor can answer these. The answers tell you more than a feature matrix will.
- When it flags someone, what exactly can I show them? A score and a confidence level, or a timestamped record of what occurred?
- Who makes the decision? If a number reaches my team first, the tool has already decided. What does my reviewer see when they start?
- If a candidate's lawyer asks how we reached this conclusion, what do I hand over? If the answer is a model output, ask who is going to explain it.
Capifiq's answers. You can show the candidate the same timestamped record your team saw. No score, ranking or recommendation is produced, so there is nothing for a reviewer to defer to. What you hand over is the sealed record of the session, and a person decides what it means.
Compare properly
See the record before you decide.
The fastest way to tell these products apart is to look at what each one hands you at the end of an interview. We are happy to show you ours next to anything else on your shortlist.