Does Eye Tracking Really Outperform the Polygraph? What the 2026 Scientific Reports Study Actually Found

A peer-reviewed study published in Scientific Reports in 2026 compared an eye-tracking Concealed Information Test with a polygraph-based CIT in the same participants. The eye-tracking method achieved a numerically higher accuracy rate. Some commentary around the study has suggested a decisive result. The reality is considerably more nuanced — and considerably more interesting.

What Did the Researchers Actually Test?

The study by Celniak, Słapczyńska, Jarszak and Augustyniak, published in Scientific Reports (2026), compared two methods of assessing concealed recognition in a controlled laboratory setting. Both were variants of the Concealed Information Test — not conventional specific-issue polygraph examination.

CIT Is Not Ordinary “Lie Detection”

The Concealed Information Test is a recognition-based psychophysiological methodology. In a CIT, an examinee is presented with multiple-choice stimulus sets. Each set contains one item corresponding to a genuine, concealed case detail — the probe — alongside several plausible but incorrect alternatives. The examiner records physiological or behavioural responses to assess whether the individual produces a differential response to the probe relative to the alternatives. This is a test for concealed recognition, not a test for deception.

CIT does not ask “Is this person lying?” It asks: “Does this person appear to recognise specific information that only someone with knowledge of the events would be expected to recognise?”

In the Celniak et al. study, 39 participants took part in a simulated mock-crime involving a mock theft. Twenty were assigned to the guilty condition (they committed the offence and possessed concealed knowledge of the critical details). Nineteen were assigned to the innocent control condition. All 39 underwent both the eye-tracking CIT and the polygraph CIT, in randomised order.

The Study Design in Brief

The Polygraph CIT Condition

• Instrument: Lafayette LX4000 polygraph
• Format: Sequential presentation — each alternative shown one at a time
• Scope: Four CIT subtests
• Channel: Electrodermal activity for numerical classification (conventional Lykken-style scoring)
• Examiner: Blind to participant condition
• Exclusions: Five recordings excluded (physical countermeasures or artefacts identified)

The Eye-Tracking CIT Condition

• Instrument: Tobii Pro Fusion eye tracker
• Format: Simultaneous presentation — four alternatives displayed together on screen
• Scope: Seven CIT subtests, each repeated across three series
• Channels: Fixation behaviour, visit counts, glances, pupil diameter (multiple ocular measures)
• Classification: Support Vector Machine (SVM) with Radial Basis Function (RBF) kernel
• Exclusions: None — all 39 recordings analysable

The Headline Result: 94.9% versus 88.2%

The eye-tracking CIT correctly classified 37 of the 39 participants: 94.9% accuracy. The polygraph CIT correctly classified 30 of the 34 evaluable examinations: 88.2% accuracy.

The polygraph denominator was 34, not 39. Five recordings were excluded because the examiner identified physical countermeasures or artefacts. Those examinations could not be scored using the standard protocol. The two accuracy figures are therefore not calculated on the same base population — an asymmetry that must be taken into account when comparing them. The eye-tracking condition produced no exclusions.

The Most Important Number in the Study

The numerical difference between 94.9% and 88.2% is real. But the authors report that the difference in classification accuracy between the two methods was not statistically significant.

This is not a footnote. It is one of the study’s central findings. A statistically non-significant difference means the data are insufficient to conclude that the eye-tracking CIT is genuinely more accurate than the polygraph CIT. The numerical gap observed could plausibly have arisen by chance given the sample size.

A numerically higher accuracy figure is not the same as a demonstrated superiority. Where the difference is not statistically significant, scientific caution requires treating the two methods as not yet shown to differ in classification performance.

Was This Really an Equal Comparison?

Multiple variables changed simultaneously between the two conditions. The polygraph condition involved sequential presentation, four CIT subtests, and conventional electrodermal scoring by a trained human examiner. The eye-tracking condition involved simultaneous presentation of four alternatives, seven CIT subtests, three repetitions of each subtest, multiple ocular features, and machine-learning classification.

The comparison was not simply eye movements versus polygraph physiology. It was closer to:

a simultaneous, multi-feature, machine-learning CIT with seven items and three repetitions
— versus —
a sequential, electrodermal-scored polygraph CIT with four items

It is not possible, from these data alone, to determine how much of the numerical performance difference was attributable to the measurement modality, as opposed to the number of items, repetitions, presentation format, classification algorithm, or some combination. A methodologically cleaner future experiment would hold those factors constant and vary only the measurement channel.

What the Study Does Show

None of the above qualifications diminish the genuine importance of this research. The study demonstrates four things of clear scientific value: (1) concealed recognition produces measurable differential responses in ocular behaviour; (2) 94.9% laboratory accuracy is an impressive result for an eye-tracking CIT; (3) all 39 eye-tracking recordings were analysable, compared with 34 of 39 polygraph recordings; and (4) the within-participant design provides strong methodological control.

What About Countermeasures?

Five polygraph recordings were excluded because the examiner identified physical countermeasures or artefacts. Physical countermeasures in polygraph CIT typically involve deliberate movements — muscle tensing, controlled breathing alterations — timed to inflate physiological responses to alternative items and distort the differential pattern CIT scoring relies upon.

The authors suggest deliberate manipulation of eye movements may be more difficult, since saccadic eye movements and fixation behaviour are largely automatic. This is a plausible hypothesis. But participants in this study were not trained in specific eye-tracking countermeasures, the study does not test whether sophisticated rehearsed eye-movement strategies could defeat the classification, and deliberate countermeasure testing should form part of any future validation programme. Neither technology should be described as countermeasure-proof.

The Practical Strengths of Eye Tracking for CIT

• Speed: Eye-tracking condition approximately 10 minutes; polygraph condition approximately 30 minutes.
• Sensor simplicity: No peripheral sensors attached to participants.
• Complete datasets: All 39 eye-tracking recordings were analysable.
• Automation potential: Machine-learning classification offers automated, reproducible scoring once classifiers are validated at scale.
• Complementary measurement: Eye movements may capture a different facet of the recognition process than autonomic physiology — not merely a different technology for measuring the same thing.

This Is Not Really a Contest Between “Lie Detectors”

Neither a polygraph instrument nor an eye tracker contains a sensor for deception. Both measure the physiological and behavioural consequences of processes associated with concealment, recognition, attention and inhibition. In a CIT context, when a concealed memory is activated by a probe stimulus, an orienting response may be triggered, the probe acquires heightened salience relative to alternatives, and inhibitory processes may operate as the individual attempts to suppress recognition. These processes can produce measurable changes in autonomic physiology and in ocular behaviour. The polygraph CIT and the eye-tracking CIT may therefore be measuring different manifestations of the same underlying recognition process.

This suggests the more productive question is not “which technology is better at detecting lies?” but rather: which measurement channels, individually or in combination, most reliably capture the differential responses that recognition memory produces?

CIT Is Not the Same as Conventional Specific-Issue Polygraph Testing

CIT and the Comparison Question Test (CQT) — used in most conventional specific-issue polygraph examinations — operate on different principles and address different investigative questions.

The demonstration that eye tracking can function as a measurement modality within a CIT framework does not address CQT methodology. Any extrapolation from CIT eye-tracking data to claims about conventional polygraph examination would be scientifically unsupported.

Laboratory Accuracy Is Not Field Accuracy

Limitations of This Study

• Small sample: 39 participants. Confidence intervals around the accuracy figures are wide.
• Simulated crime: Deliberately constructed, memorable critical details may not reflect naturally occurring cases.
• No real-world consequences: Responses in genuine jeopardy may differ materially from a low-stakes laboratory context.
• Controlled information environment: In real cases, information leakage through media, disclosure or investigative processes is a genuine risk.
• Countermeasure naivety: Participants were not trained in eye-tracking countermeasures.
• No independent replication: A single study should not be treated as definitive.
• Machine-learning validation: The SVM classifier was optimised on these data. Generalisation to real forensic cases requires prospective validation on unseen data.

Promising does not mean operationally validated. The distinction between laboratory accuracy and forensic field accuracy is one of the most persistent challenges in applied psychophysiological research. It applies with equal force to both methods examined in this study.

A Note on Machine-Learning Classification

Machine-learning classifiers offer genuine advantages — integrating multiple features, identifying complex non-linear patterns — but their application to forensic classification raises important considerations: transparency and explainability for courts; reproducibility; overfitting risk on small datasets; and the need for prospective validation on unseen data from real forensic cases. These are not reasons to reject machine-learning approaches. They are the questions a rigorous validation programme would need to answer before a machine-learning CIT classifier could be considered suitable for operational forensic use.

Could Combining the Technologies Be the More Interesting Question?

Given that polygraph CIT and ocular CIT may capture different manifestations of the same underlying recognition process, a multimodal CIT framework — simultaneously recording electrodermal activity, cardiovascular measures, respiration, eye movements, pupil responses and behavioural variables — could, in principle, provide richer and more robust discrimination than any single channel alone. The incremental contribution of each channel could then be assessed empirically. This is a future research direction, not a current operational capability — but it may be among the most productive framings of what the next generation of CIT research should examine.

Conclusion

The 2026 Scientific Reports study by Celniak and colleagues is a genuine and valuable contribution to the science of recognition-based credibility assessment. The eye-tracking CIT achieved an impressive 94.9% classification accuracy under laboratory conditions. The within-participant design is methodologically sound.

But the study does not demonstrate that eye tracking has supplanted the polygraph. The numerical difference was not statistically significant. The conditions differed substantially in design. The sample was small, the crime was simulated, and field validation has not been conducted.

On the basis of this study, eye tracking should be regarded as a promising emerging measurement modality for CIT — one that merits serious scientific attention and further rigorous investigation. It should not yet be regarded as a validated replacement for forensic polygraph examination in operational settings. The appropriate response to promising laboratory findings is further research, independent replication, and careful translation into professional practice.

Reference

Celniak, W., Słapczyńska, D., Jarszak, B., & Augustyniak, P. (2026). Evaluating eye-tracking as a method for deception detection against the polygraph as deployed in practice. Scientific Reports, 16, 25352. https://doi.org/10.1038/s41598-026-55344-9

This article was written by Dr Keith Ashcroft, investigative psychologist and principal forensic polygraph examiner at the Centre for Forensic Neuroscience. Dr Ashcroft is a Chartered Psychologist registered with the British Psychological Society and a full member of the American Polygraph Association. He has received specialist training in Concealed Information Test methodology at Japan’s National Research Institute of Police Science (NRIPS) and has served as External Examiner for doctoral research on CIT validity at the University of Huddersfield.