EEG, eye tracking, facial coding: a buyer's guide.

If you have shopped for neuromarketing research, you have heard pitches from at least three kinds of vendors who all sell the same thing in slightly different language. The EEG specialists will tell you that brain waves are the only true window into consumer response. The eye trackers will tell you that nothing matters until you know where the gaze landed. The facial coders will tell you that emotion is everything and emotion lives on the face. The biometrics shops will throw in skin conductance, heart rate variability, and a bundle of physiological signals nobody can quite explain to a marketing director without a slide.

All of them are right about their tool. All of them are slightly oversold. The honest version of the field is that each method measures something different, none of them measures everything, and the question you bring to the study should decide which combination you use, not the other way around.

Here is a practical, vendor-neutral walk through the four signals, what each is genuinely good at, and where you should be sceptical.

EEG: when the question is "what is the brain doing, when, and how strongly"

Electroencephalography measures voltage changes on the scalp produced by the synchronous firing of large populations of cortical neurons. A research-grade EEG cap reads electrical activity at thirty-two or sixty-four points on the head, typically sampling at 250 hertz or higher. From those readings, analysts derive measures of attention, cognitive load, valence (positive or negative), and approach-avoidance motivation, each with reasonable scientific backing.

What EEG is genuinely good at: temporal resolution. EEG can tell you within a few milliseconds when something changed in the brain's response. If you want to know whether the brand reveal at second 22 of an ad caused a spike in engagement or a collapse, EEG can answer that with precision that no other consumer tool matches. It also handles long-form content well, because you can map the engagement trace second by second across a four-minute video and find the moments where attention left the room.

Where EEG is overhyped: spatial resolution and "reading minds." EEG measures cortical activity, but it cannot localise that activity precisely, and it cannot tell you what a person is thinking. The marketing-trade-press picture of an EEG cap producing a readout that says "purchase intent: 73 percent" is, to put it kindly, a simplification. EEG also requires controlled conditions. You cannot do EEG in a busy store. You cannot do it well on a moving subject. It is a lab tool, primarily, with some emerging field applications.

The right use cases: pre-testing video creative, comparing alternative cuts of the same ad, evaluating long-form content, measuring engagement curves through a brand presentation, comparing alternative audio brand assets. The wrong use cases: anything in the wild, anything where you need to know exactly where in the visual field the consumer was looking, anything that requires individual-level prediction.

Eye tracking: when the question is "what did they actually see"

Eye tracking measures gaze position, fixation duration, saccade patterns, and pupil dilation. Modern systems do this with a small camera array that watches the eyes from a few feet away or, increasingly, through standard webcams with somewhat reduced accuracy. The output is a precise map of where the gaze landed, in what order, and for how long.

What eye tracking is genuinely good at: ground truth on attention. If you want to know whether anyone is actually seeing your brand asset, eye tracking gives you a clear answer. If you want to know whether a shopper's gaze went to the price first or the pack first, eye tracking tells you. Heat maps and gaze paths are some of the most actionable visual outputs in the field, because creative and design teams can look at them and immediately understand what to change.

Pupil dilation, which is part of most eye tracking systems, is a useful secondary signal for arousal and cognitive load. It is non-trivial to interpret cleanly, because pupils respond to light as well as to interest, but in controlled conditions the data has real value.

Where eye tracking is overhyped: emotional inference. Knowing where someone looked does not tell you how they felt about what they saw. A long fixation can mean love or confusion. A short one can mean dismissal or perfect comprehension. Eye tracking on its own cannot disambiguate. The vendors who sell "engagement scores" based purely on eye data are stretching the data past where it actually goes.

The right use cases: pack design optimisation, shelf layout testing, web and mobile usability, ad attention analysis. The wrong use cases: anything where the question is about emotional response without a second signal to triangulate.

Facial coding: when the question is "what did they feel"

Facial coding analyses video of a participant's face and classifies micro-expressions according to systems derived from Paul Ekman's work on the Facial Action Coding System. Modern computer vision lets this happen automatically on a recorded webcam feed, with reasonable accuracy on the basic emotional categories of joy, surprise, anger, disgust, sadness, and contempt.

What facial coding is genuinely good at: emotional response over time, especially for video and audiovisual content. A facial coding trace through a thirty-second ad will tell you where the laugh was, where the disgust was, where the smile that the viewer suppressed almost reached the lips before being corrected. The granularity is better than survey self-report by a wide margin.

Where facial coding is overhyped: precision of emotion categories and applicability outside video. The science underlying the six basic emotions is robust but contested in academic circles, and the boundaries between adjacent categories are fuzzier than the vendor outputs make them look. Coding also depends on the participant being recorded clearly, which limits applications in the wild and introduces some self-consciousness bias (people perform their faces slightly when they know they are being filmed).

The right use cases: ad testing, content evaluation, comparing emotional response across creative variants. The wrong use cases: anything where you need to know what someone was thinking rather than feeling, anything that requires natural unobserved behaviour over long periods.

Each method measures something different, none of them measures everything, and the question should pick the method.

Biometrics: when the question is "what is the body doing"

Biometrics is a catch-all category that usually includes galvanic skin response (GSR), heart rate, and heart rate variability. These are measures of physiological arousal, mediated by the autonomic nervous system, and they correlate with emotional intensity in well-defined ways. They do not, on their own, tell you whether the emotion is positive or negative. They tell you that something is happening, and how much something is happening.

What biometrics is genuinely good at: arousal detection. If you want to know whether a piece of communication produced a strong physiological response, GSR will tell you. Combined with facial coding for valence (positive or negative), the picture gets quite clear. Biometrics is also relatively unobtrusive once participants forget they are wearing the sensors, which makes it useful in more naturalistic settings than EEG.

Where biometrics is overhyped: standalone interpretation. A GSR spike means arousal of some kind. It does not, on its own, mean engagement or interest or purchase intent. The signal is useful as a confirmation or as part of a combined analysis. As a single number you report to a client, it is meaningless.

The right use cases: paired with facial coding or eye tracking, in any study where emotional intensity is part of the question. The wrong use cases: as a single signal for anything strategic.

Implicit reaction time testing

One more method belongs in any honest buyer's guide, though it is not strictly a neural measure. Implicit reaction time testing, often using paradigms derived from the Implicit Association Test, measures how quickly a participant pairs concepts under time pressure. It is a behavioural method that approximates what the unconscious mind has stored about a brand.

What implicit testing is good at: revealing brand associations that surveys cannot reach. If you want to know whether your brand is genuinely associated with "premium" or whether respondents are just being polite when they tell you it is, implicit testing answers that. It scales well, runs online, and produces stable results across thousands of participants.

What it cannot do: tell you why an association exists. The output is a map of mental connections. The interpretation of those connections, and the strategy that follows, is still a human job.

How to combine the signals

The single most important question in any neuromarketing brief is not which signal to use. It is what you want to learn. From that question, the right combination usually becomes obvious.

If you are pre-testing a thirty-second ad and want to know whether it works, the answer is usually facial coding plus eye tracking, with optional EEG for high-stakes decisions. The face will tell you the emotional shape, the eyes will tell you what was actually seen, and the brain data will give you the millisecond-level precision needed to recommend specific edits.

If you are testing a new pack design on shelf, eye tracking is the foundation. You need to know whether shoppers see the new pack faster than the old one, where their gaze travels, and which on-pack cues are landing. Implicit testing can supplement this by telling you whether the new design shifts category associations.

If you are doing brand health and want to go beyond survey scores, implicit testing alone is often enough, run at sufficient scale, segmented by your real customer groups.

If you are doing in-store or experiential research, the practical constraints push you toward eye tracking with mobile glasses, optional facial coding from a separate camera angle, and observation. Full EEG in those settings is usually more trouble than it is worth.

What to ask any vendor before you sign

Three questions that filter most of the noise. First, "How will you triangulate your conclusions across at least two signals?" Anyone who is selling you a single-signal solution for a multi-dimensional question is overselling. Second, "What is your sample size and how will you handle individual variance?" Neural data is noisy. Small samples can mislead. Third, "What will the deliverable look like, and what will it allow us to change in our work?" If the answer is a glossy report with no actionable edits, the study is decoration. Push for the brief that comes out the other side.

The honest summary

Neuromarketing is not a single tool and there is no best tool. There is a set of complementary signals, each with strengths and limits, and a research craft of combining them well. The vendors who tell you otherwise are selling a tool, not an answer. The studies that are worth paying for are the ones that start with your actual question, pick the right combination, and end with a recommendation specific enough that your creative or design or product team can act on it within the week.

None of this is exotic. It is just careful measurement. The reason it works is not that brains are mysterious. It is that humans are bad at reporting on themselves, and properly designed instruments are better at hearing the parts of the answer the human cannot speak. Pick the right instrument. Ask the right question. Read the result honestly. Make the change. Repeat.