Eye Tracking

Design reviews kept stalling at the same place. A user failed to find the signup button, and there was no way to tell whether the button never registered visually or whether it registered and simply didn't read as the thing to click. The first calls for a change in position and contrast, the second for a change in wording. Not knowing which, teams changed both and moved on without learning which one worked.

The records on hand didn't settle it either. Clicks mark where a hand went and say nothing about where it didn't. Asking doesn't help much more: "I never saw that button" can mean the eyes truly never landed there, or that they did and nothing stuck. People don't remember the path their own eyes took.

Eye tracking fills that gap with hardware. It came over to screen design from psychology and advertising research. Because it records seeing rather than clicking — and, when someone did look, in what order and for how long — it separates "invisible" from "visible but ignored" for the first time.

Seat a participant in front of the equipment, align it with their eyes, then give them a task and record where they look while they work. Six steps.

1. Pick the hardware. A bar-shaped unit mounted below a monitor covers desktop screens; glasses-mounted trackers cover situations where the head moves, like a retail shelf or a physical device. Webcam-based tracking also exists, but the error is wide enough that you should only trust it to split a screen into four or five rough zones.
2. Decide in advance what you want to know about. Name the regions you care about — the price table, the signup button, the banner at the top. Skip this and you still get a recording, but with no basis for reading it you end up cherry-picking whatever stands out.
3. Calibrate. Five to nine dots appear and the participant follows them in turn so the tracker can map eyes to coordinates. This fails often enough — glasses, contacts, lighting — that a few participants drop out right here.
4. Give the task. Something with a clear goal: "find the monthly price for a ten-person team on the pricing page." Let people browse freely and gaze scatters until nothing is shared across participants.
5. Read the output. You get a map with gaze intensity layered in color, a plot linking fixations in the order they happened, and three numbers: the share of participants who looked at a region at all, how long it took them to look at it first, and the total time spent there. If only 40 percent looked at the price table and it took them 3.2 seconds to get there, that table is barely on the page.
6. Ask why. Right after the recording, walk the screen with the participant and ask what they were looking for at each point. Gaze data reports where the eyes went and stops there, so this usually runs bundled with a moderated test, where a facilitator sits alongside and asks.

How many participants depends on the claim you want to make. Five to eight will surface the places people get stuck, but to point at a color map and say a region goes unread, a recommendation of around thirty has been in circulation for a long time. Maps built from a handful of people swing hard on the viewing habits of one or two of them.

The situations where this method earns its cost are narrow. Information is dense enough that reading order decides the outcome; the gaze that matters is off-screen, as with a shelf or a kiosk; or you need to know whether a brand registers in an ad. All three share one thing — no clicks are recorded there.

There are four reasons to drop it instead.

1. Cost. Hardware alone runs from about $2,000 to $25,000, and even renting it, calibration and analysis eat real hours. Budget two to three weeks from plan to readout.
2. No reasons. Seven seconds on the price table may mean the table is compelling, or may mean nobody understood it the first time and had to reread.
3. Looking is not reading. Ask afterward what people remember and it's common for them to draw a blank on a line their eyes rested on for more than two seconds.
4. The setting is unlike real use. Sitting at unfamiliar equipment under an instruction to find something makes people scan more carefully than they ever would at home while scrolling fast.

So small companies rarely run it. If the question is what a first screen communicates, a five-second test answers it in a day. If the question is where real users struggle, session replay answers it from recordings that already exist. Eye tracking comes out when neither of those answers the question and the answer is attached to a decision big enough to justify the equipment.

"I don't think this rises to eye tracking. Let's start with a five-second test."

A suggestion to filter with the cheap method before reaching for the expensive one.

"In the eye tracking data, only 40 percent of people ever looked at the price table."

There's a region sitting on the screen that gaze never reaches.

"Calibration wouldn't hold, so we lost two participants."

The tracker couldn't map their eyes, and their sessions dropped out entirely.

At Pagedeck the method came up during a rework of the pricing page. Paid team conversion was running below expectation and nobody could separate a pricing problem from a visibility problem, so Noa, the product designer, proposed eye tracking.

The quote ended it. A research vendor with the equipment wanted two to three weeks and a substantial fee, which was heavy for a 45-person company spending it on one page. Noa instead showed the screen to twelve people for five seconds each and asked what they remembered. Three recalled a price. The answer arrived in two days.

The most common mistake is reading long gaze as strong interest. The red areas on the map look like the parts that worked, which makes the assumption a natural one. But gaze lingers for two different reasons: attraction, and rereading something that didn't parse. Eight seconds on a pricing line may be evidence that it doesn't land the first time rather than evidence that it lands well. Leave the red areas untouched without separating the two and a sentence users trip over every time gets preserved as a sentence that works.

Assuming the equipment supplies the answer is the other frequent slip. Precise recording feels like it should yield a precise conclusion. What decides the quality of the output, though, is the task. A recording taken under "have a look around" and one taken under "find the monthly price for a ten-person team" are different pictures, and the first varies so much per person that layering them produces no shape at all. What to settle before renting hardware is not the participant count but what you will ask people to do.