Does HireVue catch ChatGPT? How its scoring and integrity checks work

What HireVue can and cannot see, in one list

This post is about detection and integrity. If you want the interview format itself — how many questions, prep time, retakes — that is covered in our guide to the HireVue one-way video interview. Here is what HireVue's own pages say, sorted by whether it is documented:

• Documented, employer-configurable: browser-focus tracking (tab or window switches), disabling paste, periodic webcam snapshots, ID verification, "identity continuity", location signals and a pre-interview conduct statement. All listed on HireVue's cheating and fraud page.
• Documented for coding challenges: a code-similarity score based on the MOSS algorithm and an editor playback that marks pastes, time away from the challenge and idle periods.
• Documented as not used: facial expressions, body language, your background and tone of voice. HireVue's 2025 AI Explainability Statement says the AI uses only what you say.
• Not documented anywhere we could find: a detector for AI-generated spoken answers, eye or gaze tracking, or reading the contents of other apps or tabs on your computer.

"Not documented" is not the same as "does not exist". HireVue ships features faster than its help pages are rewritten, and employers can layer their own processes on top. But it is the honest boundary of what can be said from public sources.

How HireVue's AI scoring actually works

A lot of the fear around HireVue comes from a picture of a camera reading your face. That is not how the current system is described. According to the 2025 AI Explainability Statement, an AI-scored video answer goes through three stages:

1. Speech to text. Your recording is transcribed by a third-party speech-to-text system from Rev.ai.
2. Understanding the text. The transcript is passed through HireVue's own language model, called CUSTARD, which is built on RoBERTa and fine-tuned on interview data — one model per competency, 19 in total.
3. Scoring. The output is scored against a behaviourally anchored rating scale for that competency, with five levels from "novice" to "expert". The models were trained to mimic the ratings of trained human evaluators.

Three details in the same document matter for the "can it be gamed" question. First, HireVue says the language model works on meaning rather than vocabulary, which it argues makes it harder to game the interview by stuffing in particular words or phrases. Second, if your answer is too short, unclear or inaudible for the system to score, it is not scored by AI at all; it is flagged for human review. Third, candidates are shown an AI consent statement and can opt out of AI evaluation, in which case recruiters rate the same answers by hand against the same scales.

Note what this pipeline is built to do: it estimates how well an answer demonstrates a competency. It is a scoring system, not a lie detector or an authorship checker, and nothing in the statement describes it judging whether the words were your own.

Facial analysis and eye movement: what was dropped and what never was

HireVue did once use visual analysis. In a post dated 12 January 2021, the company said its research found visual analysis had far less correlation with job performance than other elements, that adding non-verbal data to language data gave no significant extra predictive power, and that it would "not use any visual analysis in our pre-hire algorithms going forward." SHRM reported that the facial-analysis component had been discontinued in March 2020 and publicly disclosed in January 2021.

The 2025 statement repeats this in plain words: the AI does not use any video analysis or other audio characteristics, so it does not assess facial expressions, body language, background and surroundings, or tone of voice.

On eye movement specifically: none of the HireVue pages we checked — the explainability statement, the integrity page, the candidate help centre or the employer help centre — mention eye or gaze tracking. HireVue's own candidate tips go the other way and tell you not to worry about making eye contact with the camera. The same page even says you can keep notes next to you during the interview.

The one place your face is still captured for integrity purposes is webcam snapshots (next section). Those are still images for a person to glance at, not input to the scoring model.

The integrity features an employer can switch on

None of these are on by default for every interview. HireVue describes them as controls "configured by role risk", and an employer chooses which to use. From the cheating and fraud page and an October 2025 HireVue post on mitigating cheating:

• Browser-focus tracking. The platform monitors whether you switch tabs or windows during the interview. HireVue's own framing is that this is a signal that responses may deserve a closer look, not proof of anything.
• Disabling paste. Copy-and-paste can be limited, which matters most for essay and coding questions.
• Candidate image capture. Periodic webcam snapshots to confirm it is the same person throughout. HireVue's candidate help article says that if you see a small video of yourself in the bottom-left corner, the employer has enabled this; snapshots are taken at random intervals, mainly during non-video questions such as written answers, coding challenges and multiple choice, to check that nobody else is helping you.
• Identity checks. ID verification and "identity continuity" are listed as features, alongside location signals.
• A conduct statement. A checkbox before the interview, customisable per job; the example HireVue gives is "I agree to answer all questions in my own words."
• Screen-sharing questions. Separately, an employer can add a question type that requires you to share your screen while answering. The integrity page also mentions explaining "screen" checks to candidates before the interview, without detailing them.

The common thread: these features observe the browser session and the camera. HireVue does not describe software that inspects other applications on your computer.

Does HireVue detect ChatGPT-written answers?

Not in anything HireVue has published. Its 2025 post on cheating names "AI-generated answers" as one of the more concerning forms of cheating, but describes no mechanism for detecting them. The tools it does describe — focus tracking, snapshots, a conduct statement — target the circumstances around an answer, not the text of it. The same post quotes a HireVue scientist saying the company has "not seen any score drift" since large language models arrived, based on data from 2022 onwards; HireVue has not published the figures behind that.

You will find blog posts claiming specific similarity thresholds or hidden AI-text classifiers in HireVue. We could not trace any of those claims to a HireVue source, so we do not repeat them.

What the scoring model does do is compare your answer against behavioural anchors written for a specific competency. A polished but generic answer — the kind a chatbot produces when it knows nothing about your actual work — has little concrete behaviour to anchor to. That is not detection; it is just how the rubric is built. And a human reviewer watching the video sees what no classifier needs to: whether you sound like you are remembering something or reading it.

HireVue coding challenges: similarity scores and keystroke playback

Coding questions (HireVue calls the product CodeVue) are where the integrity tooling is most detailed, because code leaves a clearer trail than speech.

• Code Similarity. Your solution is compared with other candidates' solutions to the same challenge in the same language, across HireVue's database of submissions. It is based on MOSS (Measure of Software Similarity), the plagiarism-detection algorithm long used in university computer science courses. Results come back as Extremely Similar, Expected Range or Extremely Unique, and need more than ten submissions per challenge and language before they appear. Supported languages are C, C#, C++, Java, JavaScript, Objective-C, PHP and Python. (HireVue help: Code Similarity)
• Code Editor Playback. Reviewers can replay how your solution was written. The timeline marks pastes (with the number of characters pasted), navigating away from the challenge, running tests and periods of inactivity. (HireVue help: Plagiarism & CodeVue Interviews)
• Prevention settings. Paste can be disabled inside the editor, challenges can be randomised per candidate, and employers are encouraged to write custom challenges that are less likely to have solutions online.
• Follow-up video questions. HireVue's recommended way to check authorship is a recorded question such as "Explain your solution and why you chose that approach."

HireVue is unusually candid about the limits. Its help pages say similarity is not a definitive indicator of plagiarism, that simple challenges have only a few valid solutions, and that a textbook-looking answer can simply mean the candidate knows the textbook. An "Extremely Unique" result is also flagged for review, not just an "Extremely Similar" one. The documented signals are about how code arrived in the editor and how it compares with other submissions; there is no documented check for whether code was written by an AI model specifically.

For more on how coding platforms handle this, see HackerRank and CodeSignal AI-assisted interviews.

What the employer actually sees

HireVue is careful to position itself as a tool, not a decision-maker. The explainability statement says its platform does not make recruitment decisions, and that the final decision always remains with the employer. The integrity page says HireVue does not automatically reject candidates for suspected cheating and that "a single behavior should not automatically be treated as proof of fraud."

Put together from those pages, a reviewer at the employer may see:

• Your recorded answers and, for AI-scored assessments, competency results. HireVue's recommended setup shows result tiers such as Top, Middle and Bottom and turns raw numeric scores off, but the employer configures this.
• Any integrity signals the employer enabled: focus-loss events, webcam snapshots, identity checks, brought together with performance in one report alongside "the underlying evidence".
• For coding: the final code, test results, the similarity category, the playback timeline and your follow-up video.

So the realistic risk is rarely an automated "AI detected" verdict. It is a recruiter seeing three focus-loss events on a question where you then gave a fluent, oddly generic answer, and deciding to look closer or move on. How strict that reading is depends entirely on the employer.

An honest note, and where Interview Copilot fits

Detection aside, reading answers off a second source tends to show. Eyes track sideways at a steady pace, pauses land in odd places, and the phrasing stops sounding like you — which is exactly what the human watching the recording is there to notice. Employers set their own rules for each interview, many spell them out in a conduct statement you tick before starting, and you are the one responsible for following them.

It is also worth being straight about our own product. Interview Copilot is built for live, two-way interviews: it listens to the interviewer's voice through system audio and suggests answers in real time. A one-way HireVue recording has no interviewer speaking, so that core loop has nothing to listen to. For HireVue, its practical use is preparation — rehearsing behavioural questions aloud, checking your stories against the competencies you expect, and getting used to answering in about ninety seconds. See how the real-time interview assistant works for live rounds, and the AI coding interview assistant for practising the kind of problems a HireVue coding challenge sets, so that you can explain your solution when the follow-up question comes.

There is a free tier, and Pro is $25 for 30 days with no subscription.

Some employers name video assessments in their written AI rules. See which companies allow AI in interviews.

Sources

Checked against these pages on 31 October 2026. Which features apply to your interview depends on how the employer configured it.

• HireVue — Cheating and fraud (integrity features)
• HireVue — Mitigating cheating and enhancing candidate experience in the era of AI (October 2025)
• HireVue — 2025 AI Explainability Statement (PDF)
• HireVue — New audit results and decision on visual analysis (January 2021)
• SHRM — HireVue discontinues facial analysis screening
• HireVue candidate support — Webcam monitoring during an interview
• HireVue help — CodeVue OnDemand: Code Similarity
• HireVue help — Plagiarism & CodeVue Interviews
• HireVue help — Overview of Interview Question Types
• HireVue — Candidate interview tips

Related articles