Why Capifiq
Three unrelated products get called "AI interview tools". Some evaluate the candidate for you. Some host the call. Capifiq keeps the judgement with your interviewer and makes the session itself verifiable, the interview runs here, on a client that can actually see the candidate's device, and every finding is timestamped and sealed. Three axes, compared separately below.
Axis one
"AI interview tool" covers three unrelated products. The first thing to separate is whether the software forms a view about the person.
Automated evaluation
An AI interviewer conducts the session and scores the answers, or an interview-intelligence tool scores the recording afterwards. The product's value is the judgement it produces: a rating, a competency profile, a ranked shortlist. A person may sign off, but the number reached them first.
Capifiq
Your interviewer conducts the interview and reaches the conclusion. Capifiq prepares questions, runs the room, and records what the session can be trusted to show. No fit score, no ranking, no recommendation, and nothing for an automated screen-out to act on, because none is produced.
This is not a claim that AI is absent. Capifiq uses it for question preparation, media-authenticity analysis, captions and report assembly. The line is drawn at candidate evaluation. Where the automation sits, in detail.
Axis two
Integrity is a property of the session, so it follows whoever owns the session. This axis is independent of the first, a product can be human-led and still have no view of the candidate's device.
Built for meetings of every kind, and good at that. It records and transcribes. It has no view of the candidate's desktop, no interview protocol, and technical work happens in a separate editor or screen share, so the record of what the candidate actually did lives outside the meeting.
A detection layer bolted onto a meeting someone else owns. It sees the layer the host platform chose to expose, and nothing beneath it. It also cannot change what is being asked, so the entire off-device category, a phone, an earpiece, a person off camera, stays out of reach.
The interview is the product. Interviewers join from the browser; the candidate joins through the desktop client, which is what makes device-level verification possible at all. Video, the shared code pad, the whiteboard, captions and the integrity timeline are one session and one record.
No software sees a phone that never touches the interview computer. We say so plainly, and address that category through question design rather than pretending to detect it. That is what the Question Engine is for.
Among the products that do address integrity, two approaches are common: infer cheating from the candidate's behaviour, or compare running software against a maintained list of known tools. One produces suspicion. The other only recognises what it already knows. Capabilities vary by vendor and module, and some products combine both.
Capifiq is designed to surface unauthorized assistance even when the exact tool is new, renamed, privately built, or adapted from an open-source project.
These tools look for patterns that may correlate with cheating, but the signal remains indirect.
These scanners inspect known names, signatures, hashes, or window indicators associated with cataloged cheating tools.
Behavioral tools infer. Process scanners recognize known indicators. Capifiq verifies and preserves evidence.
Why it holds up
Coverage isn't a list of product names. New, renamed, private, or open-source-forked tools are in scope: the kind a signature-based scanner hasn't catalogued yet.
Every finding is timestamped, hashed at capture, and sealed into the report: independently verifiable. Not a probability, not a risk band, not a ranking.
The detection approach is defensible, not copyable. A provisional patent application is on file. The result holds up under scrutiny and in a dispute.
Capifiq stands apart because it verifies the interview environment and produces direct, timestamped evidence, coverage that does not depend on the tool already being catalogued, and a record that says what was observed rather than what it suspects.
Analyze gaze, posture, response timing, and other behavioral cues.
Look for names, signatures, hashes, window titles, and known indicators.
Surface integrity-relevant activity and record evidence as it happens.
| What matters | CapifiqEnvironment verification | Behavioral toolsCandidate inference | Process scannersKnown-tool matching |
|---|---|---|---|
| Core detection method | Direct environment signalsVerifies integrity-relevant activity during the session. |
Behavioral inferenceUses gaze, movement, timing, or response patterns. |
Catalog matchingChecks names, signatures, hashes, or known indicators. |
| New or previously unknown AI tool | Designed to detect the activityCoverage is not dependent on the product already being named in a list. |
Cannot identify the toolMay only generate a suspicion score from behavior. |
May miss itThe scanner must first know what indicator to search for. |
| Renamed executable or changed window title | Not dependent on the displayed nameVerification is not reduced to a filename lookup. |
No direct visibilityBehavior analysis does not validate the software identity. |
Name-based checks can failRenaming or modifying known indicators can evade simplistic lists. |
| Custom or open-source GitHub build | Environment-level coverageDoes not require the private or forked build to be cataloged first. |
Still only suspicionCannot prove that a custom assistant was present. |
Often outside the known listA changed build, hash, or process identity may not match. |
| Evidence produced | Timestamped and sealed evidenceEvents are preserved for an authorised reviewer and the final integrity report. |
Suspicion scoreRequires interpretation and may be difficult to act on confidently. |
Process or indicator matchShows a listed match, but not necessarily the complete interview context. |
| Live interviewer visibility | Live integrity dashboardThe interviewer sees events as they happen and decides how to proceed with the conversation. |
Varies by productOften centered on automated scoring rather than direct evidence. |
Scanner alert onlyTypically limited to the indicator the scanner recognizes. |
| Final interview record | Sealed integrity reportCombines the session timeline, findings, and retained evidence. |
Behavior reportDocuments observations and probability, not direct proof. |
Known-match logLimited by the scanner's catalog and the indicators it monitors. |
What the record contains
A comparison of detection approaches is only useful if the output means something specific. Capifiq labels the two kinds apart: on the page, in the interviewer's view, and in the report.
Something observed directly during the session: an active remote-control channel, a virtualised environment indicator, a display-topology change, a covered application window, a non-local input event. Each is recorded with a time, a source and an evidence reference, and sealed so later alteration is detectable.
Model-derived observations about the media itself, whether the feed appears synthetic or internally consistent. These are probabilistic. They are labelled as such, never described as proven, and always routed to a person rather than acted on automatically.
The absence of an event is not evidence that nothing happened. It means nothing covered was observed. Detection is an arms race, and a record that claims more than it saw is worth less in a dispute, not more.
Rights and review
An evidence product is only as good as the process around it. These are workflow rules enforced by the software, not policy statements.
The interviewer who ran a session is barred from reviewing it. A different authorised person reads the record and writes the integrity report. There is no override, including for a solo operator.
The candidate consents before anything is captured, and the client is active only inside the interview window the interviewer opens. It is a standalone application, when the interview is over, the candidate deletes it.
Candidates can request an adjustment before the interview, and assistive technology is filtered on their own computer before transmission: Capifiq never receives it. Recordings delete on a schedule you set, within limits we enforce, or sooner on request.
The full account is on the candidate rights page, written for the candidate rather than the buyer.
A category move
Proctoring watches the room, the webcam, and the candidate's behavior, and the newest tools are built to slip past all three. Capifiq doesn't watch the room and it doesn't lock anything down. It verifies the interview environment, and proves what it finds. Lockdown browsers and behavior scores are looking in the wrong place.
Questions
Get started
Run your next interviews on Capifiq and judge the evidence for yourself.
Free white paper
Fifteen pages on how live AI assistance works, why conventional proctoring misses it, and what actually protects your interviews: written for hiring leaders.
