When HearQA Fits: AI Copilot Risk by Platform

Before deciding whether HearQA fits a scenario, it helps to know what the other side is actually watching. Modern remote-exam and interview platforms aren't one thing — they're a stack of monitoring signals, most of which focus on the computer you're taking the exam or interview on. Understanding each signal is the difference between "this is fine" and "this is a career-ending risk."

Webcam monitoring is continuous video of your face during the session. Platforms like Proctorio, Respondus Monitor, Honorlock, AWS OnVUE, and Kryterion use face detection (to confirm it's still you), gaze tracking (to flag when your eyes repeatedly drift off-screen), and multi-face detection (to catch someone helping in the room). The AI does the first pass; a human reviewer watches flagged sessions after the exam.

Audio monitoring is continuous microphone capture. It flags whispering, lip movement, a second voice, or noticeable background sounds. Most certification proctoring platforms enable it by default. Some academic ones only enable it for high-stakes finals.

Screen monitoring typically runs through a lock-down browser (Respondus LockDown Browser, Proctorio's browser extension, AWS OnVUE's secure browser). These block other apps from opening, prevent tab switches, disable copy/paste, and record screen output. Some can detect external monitors.

Pre-exam checks — ID verification, a 360° room scan with the webcam, a clear-desk check — happen before the timer starts. Once the exam begins, only what's inside the webcam frame is actively monitored. That's the key insight: the room-scan constraint ends when the exam starts.

Behavioral flags are AI-detected anomalies: long pauses before answering, eye-movement patterns, unusual response speeds. A single flag isn't disqualifying; a cluster of them triggers human review. The review happens after you've finished — which means consequences can land days or weeks later.

The product is designed for three kinds of moments: scenarios with no one watching your camera or screen, live conversations where a second device off-camera is unremarkable, and preparation.

Unproctored online courses (Coursera, edX, LinkedIn Learning, HubSpot Academy, Salesforce Trailhead, most Microsoft Learn paths, internal company training) have no webcam, no screen lock-down, and often no timer. The "honor code" is the only barrier, and most of these platforms now explicitly allow or even integrate AI assistance. HearQA is just a better study-and-reference tool here.

Video calls on Zoom, Meet, and Teams (the standard business-meeting flow, not a locked-down interview platform) have one or a few human observers watching a small webcam tile. They aren't running AI analysis on your eyes. A phone tucked below the camera frame is about as detectable as a notepad on your desk — possible to spot if they're looking for it, unremarkable otherwise.

Take-home tests, Canvas/Moodle/Blackboard quizzes without lock-down, internal company compliance training, and practice/mock exams (Whizlabs, Tutorials Dojo, ExamTopics) have no active monitoring. Any device works, any workflow works.

Practice mode itself — the scenario in HearQA's session picker — is the responsible path for preparing for anything high-stakes. Unlimited rehearsals, full AI coaching, upload all your study materials, no one watching because there's no exam. If you're considering HearQA for an exam we tell you to avoid, we mean Practice instead.

Certification exams delivered through AWS OnVUE, Pearson VUE's home edition, PSI, Kryterion, or Certiport use the full monitoring stack — webcam, screen lock-down, pre-exam 360° room scan, AI gaze tracking, and human review of flagged sessions. These aren't gentle systems. A phone becomes detectable as soon as your eyes repeatedly drift the same direction. Even if the phone itself is outside the webcam frame, the gaze pattern is visible.

Consequences are serious: score cancellation, retake fees, multi-year bans from the certification body (AWS, PMI, CompTIA, ISC², CFA Institute have all documented bans), and — in regulated fields like finance, healthcare, or law — professional disciplinary action. A 90-minute AWS exam is $150; the actual cost of getting caught is the years of career impact.

Proctorio, Respondus Monitor, Honorlock, and ProctorU protect university finals and online-degree programs. Consequences here include course failure, academic probation, and expulsion. University integrity offices generally believe their proctors over student appeals. Standardized tests like GRE at Home, TOEFL iBT Home Edition, and IELTS Online use the same technology with even stricter policies — and disqualification often means a multi-year hold on retakes that affects grad-school and visa timelines.

For any of these, our recommendation is unambiguous: switch to the Practice scenario before the exam. Upload your study materials there, run mock sessions, ask the AI every hard question you can think of. The preparation is the help. The live exam isn't the moment to try to hide HearQA from a system specifically designed to catch that.

Standard video interviews — a recruiter, hiring manager, or panel on Zoom, Meet, or Teams — are the lowest-risk HearQA use case after unproctored courses. The interviewer watches a small video tile of your face. They aren't running gaze-tracking AI. A phone propped below the camera, angled so you can glance at it without moving your eyes far from the screen, is genuinely unremarkable.

Behavioral rounds, initial screens, system-design discussions, and most take-home-style technical discussions work well. The challenge is when the interviewer switches modes mid-call: asks you to share your screen for a live coding challenge, opens a whiteboard, or moves you into a proctored coding platform like CodeSignal or HackerRank with proctoring enabled.

CodeSignal's Interview/Screen products and HackerRank's proctored assessment mode use the same monitoring stack as certification exams: webcam, screen, sometimes mic, AI gaze tracking, sometimes code-paste detection. HireVue and similar AI-analyzed one-way video platforms run sentiment and behavioral AI on your recorded responses. Detection risk in all of these is real.

If your interview round is purely conversational — yes, HearQA fits. If you're routed to a proctored coding platform, switch to Practice beforehand instead. For mixed-format interviews (conversation plus a separate proctored coding test), HearQA helps with the conversational portion only.

Business calls — sales, client meetings, negotiations, consultations — are another HearQA-friendly context. The camera is usually on and the audio is captured (for your own transcription, via the tab-audio feature or a second mic), but nothing on the other side is trying to detect AI help. The only wrinkle is when a client asks to share your screen mid-call for a demo or walkthrough. A phone kept off-camera sidesteps that cleanly.

Presentations — internal talks, thesis defenses, conference panels, pitches — aren't stealth situations. You're running the room. The concern here is accidental: a projector or video call connected to your computer can display whatever's on your screen, including HearQA's window. A phone at your side during audience Q&A keeps AI answers to your eyes only.

Practice is the catch-all recommendation for anything high-stakes. Upload your study materials, job description, past exam papers, textbook chapters, interview notes, company battlecards, whatever you're preparing for. Run sessions in the same scenario type (Interview, Exam, Certification, Conversation, Presentation) you're preparing for; the AI adjusts its coaching style. There's no timer, no limit, and no one watching because there's no exam.

The most valuable pattern we see: users take 3–5 practice runs in the same scenario before the real one. By the third run, the AI has learned what trips them up, they've heard the same question three times in three different ways, and the real session becomes a confidence exercise instead of a first-time stumble. This is what HearQA is built for.

Practice has six sub-types. Mock Interview drills behavioral and technical questions matched to a job description. Exam Questions generates multiple-choice / short-answer / essay items from study materials. Coding Challenge poses LeetCode-style problems and grades correctness, complexity, code quality, and edge cases. Presentation Q&A simulates a skeptical audience reading your slides. Sales Roleplay has the AI play a client raising real objections about price, timeline, or competition. Free Study is open-ended Q&A with no scoring — best for last-minute review.

After each practice session the score and per-answer evaluation appears on the Practice progress page (`/portal/practice/`). Trend across sessions, weakest dimensions, and a streak counter — useful for catching whether you're actually improving or just feeling more comfortable. The page only shows up after your first practice session, so the dashboard stays focused for users who only do live sessions.

After the 2024–2025 wave of candidates using Cluely, Interview Coder, and similar tools, recruiters are paying attention. The single tell that comes up first in every recruiter-training piece (Fabric, Aptahire, Sensei Copilot, Intruder.io) is eye-gaze drift — the candidate's eyes drifting in the same direction repeatedly while they answer. Reading from a phone produces a recognizable horizontal sweep plus a downward tilt; reading off a second monitor produces a long off-axis hold. Both are visible to a careful human watcher in a normal-resolution webcam tile, no AI needed.

The mitigation that works: position the phone laid flat just below or behind your laptop screen, never lifted or held up. The eye movement to read it then looks like reading the laptop, not glancing sideways. Most webcams sit on top of the laptop screen — the camera can't see what's on the desk in front of it, only what's above the desk. A phone laid horizontally below the camera plane is invisible to the camera AND your eyes barely move to read it.

The second-biggest tell is unnatural fluency. Verbatim reading sounds generated — it lacks the small disfluencies, pauses, and "ums" that real speech has, and the cadence stays robotically even regardless of question complexity. The fix is the most universal interview advice anyway: read the AI's answer, understand the structure (key claim + evidence + transition), then say it in your own words with your own voice. The AI is a thought-partner, not a teleprompter.

A specific concern: glasses reflections. Researchers at U-Michigan showed that 720p webcams can leak on-screen text via lens reflections at ~75% accuracy for ~10mm-tall fonts, and 4K webcams reach 94% accuracy on common websites. In practice no interviewer is running forensic reconstruction — what they actually notice is light glow shifting on your lenses as the phone updates. Mitigations: dim the phone screen to ~30% brightness, use anti-reflective coated lenses if you have the option, and angle the phone slightly downward so light reflects up off your glasses (away from the camera) rather than toward it. If you don't wear glasses you can skip this entirely.

Background noise and audio leak: Zoom, Meet, and Teams all run aggressive noise-suppression algorithms that filter non-voice background sounds well. Human speech, however, passes through. If HearQA reads answers aloud (TTS) at audible volume, the call audio will pick it up. The simplest rule: keep the phone's media volume off and read the answer silently. If you must use audio, use a single earbud at the lowest volume that's legible to you.

A small but high-frequency catastrophe: phone notifications. A WhatsApp banner ding mid-interview is an obvious "second device active" tell. Enable Focus / Do Not Disturb on the phone before the call. HearQA's pre-flight checklist surfaces this reminder for Interview and Conversation templates so it doesn't get forgotten.

The 2024–2025 explosion of AI interview-assistance tools forced a structural reaction. Per Fabric's "State of Cheating in Interviews 2026" report, ~35% of candidates showed AI-cheating signs in late 2025 — more than double the H1 2025 rate — and over half of candidates use AI on coding challenges. Gartner: 72.4% of recruiting leaders are now running more in-person interviews specifically as an AI-cheat countermeasure (Computerworld).

Concrete moves: Google piloted in-person SWE rounds at Bay Area, Seattle, NYC, Poland, and Bangalore through 2025 and is rolling out a "2 virtual + 3–4 in-person" loop for all software-engineering roles. McKinsey brought back in-person final rounds for consulting hires. Cisco, Anthropic, OpenAI, and several large banks (Goldman, JPM tech) made similar moves. For finance superdays in NYC and London the on-site coding test on a company laptop is increasingly the norm again.

Detection tooling launched in 2025 that you should be aware of. Truely (built by Columbia students, July 2025) is an open-source tool that asks candidates to install a background app for real-time detection — visible and opt-in, but spreading at smaller orgs. Fabric claims an 85% Cluely detection rate via timing-flatline + gaze-pattern analysis and is being adopted as a post-interview review tool. HireVue's 2025 product literature explicitly markets gaze + voice + timing as fraud signals. HackerRank shipped Proctor Mode in July 2025 claiming 85–93% precision on ChatGPT-assisted code submissions. Codility expanded its behavioral-events detection with snapshot capture and typing-rhythm analysis.

What this means for HearQA users: live-assistance during the interview is getting harder; preparation is getting more valuable. If your target company is FAANG, a top consultancy, or top-bracket finance, assume the interview will be at least partly in-person, and shift your prep weight from "live copilot" to "Practice mode rehearsals." This is the structural argument for Practice as the long-term core use case.

Specific platforms to steer away from to Practice rather than fight: HireVue async (Unilever, Hilton, JPM-style screening — gaze + voice + timing analysis), Karat managed (Citi, Roblox, Indeed, Wayfair — a trained Interview Engineer is watching live, this is the hardest human-watcher in the industry), and any round where the interviewer asks you to share your full screen for a coding challenge (rising at Google, Meta, parts of Goldman tech). The phone-stealth thesis breaks the moment your whole monitor is shared.

Most session failures aren't product bugs — they're setup bugs the user notices three minutes into a call. HearQA surfaces a template-aware pre-flight checklist before any Interview, Conversation, or Presentation session so the things that matter get done deliberately. The checks take 60 seconds and are skippable per template once you've gone through them.

Mic permission: requested up-front so the actual session start doesn't hit a permission prompt mid-stride. If you previously denied mic access in the browser, this is where you find out and fix it before it matters.

Phone Do Not Disturb / Focus mode: an acknowledgment-only check (we can't toggle the OS DND from a webpage). The reminder is essential because a notification ding mid-call is one of the most obvious "second device active" tells. iOS Focus, Android Do Not Disturb, both work — pick the option that suppresses banners AND sounds.

Browser: for the Conversation template, Chrome desktop is the only browser that supports tab-audio capture (the call's other side without using the room mic). If you're on Firefox or Safari, the check tells you so and HearQA falls back to your microphone. Audio quality is similar — keep the laptop near the speakers.

External display (Presentation only): when window.screen.isExtended reports a second display is plugged in, the check warns you to set Extend, not Mirror. With Mirror, the projector shows whatever's on your laptop screen — including HearQA's window. macOS: System Settings → Displays. Windows: Win+P → Extend.

Glasses reminder (Interview only): a one-line acknowledgment that if you wear glasses, dim the phone to ~30% and angle it slightly down to reduce lens reflections. If you don't wear glasses, just acknowledge and move on.

Two in-session controls exist for the moments when the situation suddenly changes — the interviewer asks you to share your screen, a colleague walks behind you, or a camera angle shifts. Both are unmissable buttons in the live-session toolbar; neither requires a hotkey to find.

"Hide answers" (Ctrl+Shift+H on Windows/Linux, Cmd+Shift+H on Mac) instantly collapses the answer pane to a 4px sliver. A small banner replaces it with a "tap or press the hotkey to restore" message. The session itself keeps running on the backend — no data is lost, the pane just isn't visible. Press the hotkey again or click the button to restore. Useful for the unexpected over-the-shoulder moment.

"About to share my screen" is an explicit toggle for the planned moment when the interviewer says "can you share your screen?" Click it before you click Share in Zoom/Meet/Teams; the answer pane disappears the same way the panic button hides it, but the toggle stays visibly ON so you know to turn it off when the screen-share ends. This is a clearer signal than the panic hotkey because the user's INTENT to be hidden is recoverable — they turn it off, the pane comes back.

Both controls work the same way under the hood: the main content div gets `display: none`, while the toolbar with the toggle stays visible at the top. The session's mic, AI calls, and message persistence all keep running; only the visual answer pane is hidden. Re-showing happens instantly with no re-fetch needed.

What these don't protect against: an interviewer who asks for full-screen share for an extended technical round. If your whole monitor is shared for 30 minutes, no toggle saves you — the right answer is to know in advance (research the company's interview format) and steer to Practice if the round will be a fullscreen-share coding session.