The instruments
A 2010 review by Dan Ariely of Duke University and Gregory Berns of Emory University set out what each brain-imaging method can and cannot record. The methods differ in how sharply they show where activity happens, and when.
| Method | What it records | Strength | Limit |
|---|---|---|---|
| fMRI | Changes in blood oxygenation that broadly follow activity in the brain | Can resolve small structures, including deep ones, to about 1 to 10 millimetres | Slow, in steps of one to ten seconds, and some regions suffer signal artefacts |
| EEG | Changes in the brain’s electrical field, through electrodes on the scalp | Timing in milliseconds, and equipment that can be inexpensive | Spatial detail of about a centimetre, and poor sensitivity to deep structures |
| MEG | Changes in the magnetic fields that brain activity creates | EEG’s fast timing, with better spatial detail | Expensive, and most sensitive to activity near the surface |
| Eye tracking | Eye movements or the direction of gaze, by methods such as video-oculography | Shows the order and length of looks | See eye tracking and attention |
Two hopes behind the interest
Ariely and Berns described two reasons marketers were drawn to brain imaging. The first was cost: the hope that a scan could replace slower studies. The second was hidden information: the assumption that people cannot fully put their preferences into words, and that brain activity might reveal them, even before a product exists.
Their verdict on the first was blunt. They thought it unlikely that imaging would be more cost-effective than traditional marketing tools. On the second they were more hopeful, expecting the bigger pay-off in product design rather than in selling, while warning that the field could become a passing fad, as subliminal advertising had.
Three studies still cited
Readout · Neuron, 2007
Brain activity before a purchase
- Who
- 26 healthy adults aged 18 to 26, studied by researchers at Stanford, Carnegie Mellon and MIT
- Task
- In a scanner, each person saw a product, then its price, then chose whether to buy it
- Finding
- Liking a product went with activity in the nucleus accumbens; prices that seemed too high went with activity in the insula and reduced activity in part of the prefrontal cortex. Each region helped predict the purchase beyond what people said in ratings.
- How well
- The brain measures predicted purchases about 60% of the time, against 50% for chance
Readout · a wine study, as Ariely and Berns describe it
Price changes the experience
People tasted wine they were told cost $90 a bottle or $5 a bottle. Activity in the medial orbitofrontal cortex was higher for the “expensive” wine, and tracked how much people said they liked it, even though the wines were the same. The reviewers read this as suggesting that price shapes the experience of a product itself.
Readout · a cola study, as Ariely and Berns describe it
A brand is more than a taste
Without brand information, people showed no significant difference in preference between two colas in a taste test. When people who called themselves drinkers of one brand were told that brand was coming, their brains showed activity the other group’s did not. The study suggested the difference was cultural rather than in the drink.
The reverse inference problem
A common step in neuromarketing is to see activity in a region linked to reward and conclude that a person liked something. Ariely and Berns called this reverse inference, and explained its weakness: one brain region takes part in many processes, and choices draw on many regions.
That is why they rejected the idea of a “buy button” in the brain. On their reading of the evidence, the processes behind a purchase cannot be reduced to one area of activation. When a claim names a single brain region as proof that an advert works, this is the question to ask of it.
What later reviews found
A 2022 systematic review from Keele University looked at 174 papers that used EEG in neuromarketing. It found two measures the most consistent markers of preference: frontal alpha asymmetry, which separated positive from negative responses, and a brain response called the late positive potential. It also found limited consistency across papers, with each measure giving mixed results. Prediction improved when machine learning was used, especially alongside eye tracking or facial-expression analysis.
A 2026 review from the universities of Lleida and Jaén looked at a quieter part of method: which images, videos or products researchers show people. It found that many studies rely on stimuli made for one experiment, and argued that when stimuli are unique, poorly documented or out of reach, this makes it hard for other researchers to repeat the work.
The ethical questions raised early
The 2010 review also listed ethical concerns about bringing brain imaging into selling. In summary:
- Privacy of thought. Imaging might reveal preferences beyond the task at hand. The authors’ answer was transparency of purpose: people should know what they are helping with, and their data should be used only for it.
- Private preferences. People need control over what they reveal, and revealing a preference outside the research question is a privacy breach.
- Exploiting traits. Targeting groups by a neurological trait, or timing prices to a biological state, would strike many people as repugnant.
- Small samples. Results from a small group are generalised to whole populations.
- Abnormal findings. About 1% of people have an abnormality on an MRI scan, so the authors described a written policy for handling them as standard practice.
Australia’s rules on the first three are covered in consent in research and biometric data and privacy.