TL;DR: Eye tracking lets a headset know where the wearer is looking, many times a second. In a business XR app it is good for five things: measuring something clinically, seeing what trainees actually paid attention to, letting people select things by looking at them, rendering sharper images more cheaply (foveated rendering), and making avatars feel present in shared sessions. It is not available on every headset, some platforms deliberately hide raw gaze from apps, it needs calibration that some users will fail, and gaze data is among the most sensitive data a headset collects. Decide which of the five jobs you need before you pick hardware, prove the signal on the real device first, and plan the privacy answer before anyone asks.
Eye tracking shows up in almost every XR pitch deck now, usually as a bullet point between "hand tracking" and "analytics." That framing hides the real decision. Eye tracking is not one feature. It is a sensor, and what it is worth to you depends entirely on what you do with the stream it produces.
We have built an XR product where eye tracking is the whole point. Nystag is VR eye-tracking diagnostics on the Vive Focus 3, used for precise medical assessments and clinical evaluations. That project shaped most of what follows: when the gaze signal is the deliverable, everything else in the app is secondary to getting it right.
What eye tracking actually gives you
Cameras and infrared illuminators inside the headset watch each eye. Software turns those images into a gaze direction for each eye, often a combined gaze ray, and on many systems extra values such as pupil size and whether the eye is open. Your app receives that as a stream, frame by frame.
Everything useful is built on top of that stream. The raw data does not tell you that someone "noticed the hazard" or "lost focus." Your app has to decide what a fixation is, how long counts as reading something, and which object in the scene the gaze ray hit. Those interpretation rules are where most of the engineering, and most of the judgment, lives.
Five jobs eye tracking does in business XR
1. Clinical and research measurement
This is the most demanding use. The eyes are the measurement: how they move, how they respond to a stimulus, how stable they are. A headset is attractive here because it gives a controlled, repeatable environment with the sensor already positioned in front of the eyes.
It is also where tolerance for noise is lowest. A dropped frame or a drifting calibration is not a cosmetic glitch, it is a wrong result. We go deeper on what that bar means in clinical VR diagnostics.
2. Attention data in training
In VR training for high-risk teams, what someone looked at is often as important as what they did. Did the trainee check the gauge before opening the valve? Did they scan the room before entering? Gaze lets you answer those questions instead of guessing from hand movements.
The value is real, but only if you define the events in advance. "Looked at the exit sign for at least half a second before the alarm" is measurable. "Was situationally aware" is not. We cover how to design scores instructors will trust in measuring XR training outcomes, and how to show gaze to a trainer live in VR training instructor tools.
3. Looking as input
Gaze can be a pointer: look at a button, then pinch or press to confirm. Done well, it is fast and needs no controller. Done badly, everything the user glances at reacts, which is exhausting. Eyes move to look, not to click, so gaze almost always needs a second confirming action. The design rules for this sit with the rest of XR interaction design.
Gaze input can also open an app to people with limited hand mobility, which is worth considering alongside the other options in XR accessibility. It needs fallbacks, though, because some users cannot be tracked reliably at all.
4. Foveated rendering
Your eye only sees sharp detail in a small central area. Eye-tracked foveated rendering draws full detail where the user is looking and less elsewhere, freeing up processing power. On a standalone headset where every millisecond of frame time matters, that headroom can go into richer scenes or steadier frame rates, which also helps comfort (see VR motion sickness).
This is mostly an engine and platform feature rather than something you build from scratch. Your job is to confirm the target device and engine support it, and test that it holds up in your actual scenes.
5. Presence in shared sessions
In multi-user XR, avatars whose eyes follow where the real person is looking feel noticeably more like people. If your app is about meetings, collaboration or coaching, this can matter. If it is a solo training scenario, it does not. We cover shared sessions in mixed reality collaboration, where our MR Camera project lets multiple users place and interact with 3D models in the same space.
Hardware and platform: check this first
Eye tracking narrows your hardware choices quickly, and the details change between device generations, so treat any list as something to verify at the start of a project.
- Built in, add-on, or absent. Some headsets ship with eye tracking, some support an add-on module, and many popular consumer headsets have none. The Vive Focus 3, which Nystag runs on, supports eye tracking through an add-on tracker. If your whole fleet needs it, that changes the per-seat cost and the device shortlist in choosing an XR headset.
- What the platform lets your app see. Some platforms use eye tracking for the system interface but do not hand the raw gaze direction to apps, as a privacy choice. Apple Vision Pro is the well-known example: looking at things drives the interface, but apps are not told where the user is looking. That rules it out for gaze analytics while still allowing gaze-driven interaction. More in our Apple Vision Pro app development guide.
- Data quality. Sampling rate, latency, accuracy and how the device reports confidence vary between headsets and SDKs. For interaction and foveation, modest accuracy is fine. For measurement, you need to know the numbers and test them yourself.
- Engine and SDK support. Check that the engine you choose exposes the eye data you need without fragile workarounds. This is part of choosing an XR engine.
The parts that are harder than the demo
Calibration and the users who fail it
Most systems ask the user to follow a few dots before tracking is accurate. Some people will struggle: certain glasses, contact lenses, makeup, eye conditions, or a headset that sits slightly differently each time. A demo with one practised user hides this. A rollout across hundreds of people does not.
Plan for it. Decide what the app does when calibration fails or confidence drops mid-session: retry, fall back to head direction, flag the session as unscored, or stop. For clinical use, the answer is usually to stop and tell the operator, because a quiet fallback produces a confident-looking wrong result.
Turning gaze into meaning
A gaze ray hitting an object for one frame does not mean the user saw it. You need rules for fixations, minimum durations, and what to do when two objects overlap in view. Small, distant or moving targets are harder to attribute reliably than large ones. Build the scene with that in mind: important things should be big enough to be looked at unambiguously.
Re-checking after updates
Headset firmware, runtime and SDK updates can change tracking behaviour. If your results feed decisions, run a check set after each update before it reaches users. That belongs in the plan for XR app maintenance.
Gaze data is sensitive. Treat it that way
Where someone looks, how their pupils respond and how their eyes move can reveal attention, interest and, in some cases, health information. It is far more personal than a click log. We are not lawyers, but in practice it is sensible to treat it as sensitive personal data from day one.
A few rules we build to:
- Collect the event, not the stream, unless you need the stream. Most training apps need "looked at the gauge for 0.8 seconds," not hours of raw gaze samples. Process on the device where possible and store only the derived events.
- Say what is captured, in plain words, before the session. Wearers should know their gaze is being measured and why.
- Keep clinical data separate and controlled. Where gaze is part of a medical assessment, it is health data and needs the access controls and retention rules that come with that.
- Decide who sees individual results. A trainer seeing one person's attention pattern is different from a manager browsing everyone's.
The wider picture, including body movement, room scans and passthrough cameras, is in XR data and privacy.
How we scope an eye-tracking feature
When a client asks for eye tracking, the first conversation is about which of the five jobs it is for. That answer sets the hardware, the accuracy you need, the data you keep and the effort involved.
- Name the job and the decision it supports. "Score whether trainees check three gauges in order" is buildable. "Add eye tracking" is not a requirement.
- Pick the device by the data, not the visuals. For measurement, tracking quality and raw data access come first.
- Prove the signal on the real hardware before building the experience. A short proof of concept with real users, including people with glasses, tells you whether the signal is good enough. On Nystag, getting the measurement right was the foundation everything else stood on.
- Design the failure paths. Calibration failure, low confidence and mid-session dropouts each need a defined behaviour.
- Decide the data model and retention up front. What is derived on the device, what is stored, who can see it, and for how long.
Eye tracking adds work in hardware, calibration handling, interpretation logic, testing across users and data handling. How much depends on the job: gaze-driven menus are a modest addition, while a measurement product makes eye tracking the core of the build. We break down the wider cost drivers in what an XR app costs.
Questions to answer before you build
- Which job is eye tracking doing: measurement, attention data, input, rendering or presence?
- Does your target headset have it, and does the platform give your app the gaze data you need?
- How accurate does the signal need to be, and have you tested that on the real device?
- What happens when someone cannot be calibrated, or tracking drops mid-session?
- What gaze events matter, and how are they defined?
- Will you store raw gaze, derived events, or nothing beyond the session?
- Who can see individual results, and have wearers been told?
- Who re-checks tracking after headset and SDK updates?
The short version
Eye tracking is worth adding when it answers a specific question your app needs answered, and expensive noise when it is added because the headset supports it. Pick the job, choose hardware that exposes the data that job needs, prove the signal with real users, and treat gaze as the sensitive data it is.
Planning a VR or XR app that depends on where people look, whether for diagnostics, training or interaction? Book a demo and we'll help you pick the hardware, prove the signal and scope the build honestly. See our XR work: Nystag clinical VR eye-tracking on the Vive Focus 3, ERIS XR and Planes XR for ARCortex, and MR Camera shared mixed reality.