neuromarketing.com.au

AOI 2 · The science

Eye tracking and attention: what gaze data shows

An eye tracker records eye movements or where a person is looking. From that record, researchers mark fixations, when the eye stays still for a while, and saccades, the quick movements from one fixation to the next. It shows where attention went. Whether looking at something means choosing it, or understanding it, takes more evidence.

General informationThis page describes research methods, not advice on any study. For the ethics rules on research with people, the official place to check is the NHMRC’s National Statement page.

A black and white close-up of a human eye: the patterned iris, the dark pupil at its centre, and the lashes above.
An eye in close-up. Photo by Pexels on Pixabay

Fixations and saccades are not fixed terms

The definitions above come from a 2011 methods book by Holmqvist and colleagues, quoted in a 2018 survey by researchers at Utrecht and Lund universities. The survey asked 124 eye-movement researchers how they define the two terms, and found a wider variety of definitions than the published literature suggests, which did not seem related to the researchers’ background or experience.

The authors asked researchers to spell out four things: how fast or slow the eye moves, what the movement is for, what the movement is measured relative to, and how the event is picked out of the tracker’s signal.

That last step is done by software, and the software matters. A 2026 study from the Hebrew University of Jerusalem compared seven detection algorithms on two sets of recordings that people had labelled by hand. One algorithm matched or beat the others throughout; several varied substantially from one data set to the other. The authors concluded that the choice of detector should suit the task and the data.

Does looking lead to choosing?

A drawn three-item choice screen with an invented scanpath Three plain packets side by side on a screen. Six numbered circles of different sizes sit on the packets, joined by thin lines: the eye starts on the middle packet, moves to the left one, back to the middle, to the right one, and ends with a large circle on the middle packet. It is an illustration, not recorded data. 1 2 3 4 5 6
Illustration only, drawn in the style of a three-item choice task. Each circle stands for a fixation, larger for a longer look; each line for a saccade between them. The last, longest look lands on the middle packet.

Readout · Proceedings of the National Academy of Sciences, 2011

Three snacks on a screen

Who
30 students at the California Institute of Technology, who regularly ate the snack foods used and were not dieting
Task
Pick one of three food items shown on screen, taking as long as they liked, while an eye tracker recorded fixations 50 times a second
Finding
The first look landed at random with respect to how much people liked each item. The more extra time people spent looking at an item, the more likely they were to choose it, and the last look tended to fall on the best-liked item.
Model
The authors, Ian Krajbich and Antonio Rangel, explained this with a model in which evidence for an item builds faster while it is being looked at

Two cautions come with the study. People still looked at the worst item at least a fifth of the time, so they were weighing all three. And the authors noted that where people look is not fully independent of how much they value things, which they left for future work to explain. The study shows looking and choosing travelling together in a simple task; it does not show that steering someone’s eyes will steer their choice.

Looking is not understanding

A study published in 2016 by researchers at American University and the Uniformed Services University tested a question that matters for any advertisement with a warning in it. A content analysis of US pharmaceutical television commercials found positive facial expressions more common during the spoken health warning than in the other parts of the commercials.

In an eye-tracking experiment with commercials made for the study, the researchers varied the faces shown during the warning. Happy faces reduced how well viewers objectively understood the warning. The lesson for reading eye-tracking results is that a look shows where attention went, and a separate measure is needed to know what was taken in.

What a careful study reports

In 2023 an international group of eye-tracking researchers published a minimal reporting guideline, developed over three years through an open, consensus-based process and due for review at least every four years. Its first group of twelve items, the ones every eye-tracking study should report, is a good test of any eye-tracking result, including one in a sales presentation. A careful report says:

  1. Which tracker, and which software

    The manufacturer and model, and the software and firmware versions used to record and process the data.

  2. How it tracks, and how often

    The tracking method, and the sampling frequency: how many times a second gaze was recorded, which affects how well timings such as the length of a fixation can be measured.

  3. How the head was held

    Whether the head was free or held still, for example on a chin rest, and how head movement could affect the data.

  4. What was recorded, and in what light

    Whether the data came from one eye, and which, or from both; which measures the tracker recorded, with their units and the frame they are measured against, such as a position on the screen; and the lighting in the room, which can affect data quality.

  5. How it was calibrated

    The calibration method and design, and the criteria for accepting it, including any checks during the recording.

  6. How uncertain the measures are

    An estimate of uncertainty for each quantity reported, such as the precision of gaze position.

  7. What was done to the data, and what was lost

    The processing steps, and the share of samples missing or discarded. The guideline’s own example: if a tracker meant to deliver 250 readings a second delivers 225 usable ones, 10% of the data is lost.

A coloured map of gaze, like the drawings on this site, is only as trustworthy as the steps behind it. If a study records video of people’s faces as well as their gaze, biometric data and privacy explains how the Privacy Act treats personal and biometric information.