This post explores a fascinating fMRI study by Ninaus and colleagues (2013), who asked a simple but profound question: what happens in the brain when people believe they are doing neurofeedback, even if the feedback is not real? This work from the archives, steps back from specific neurofeedback (NF) protocols and looks instead at the more general cognitive machinery that is activated whenever a person sits in front of a feedback display and tries to control their own brain.

Neurofeedback is a form of biofeedback in which real-time measures of brain activity (for example EEG rhythms or BOLD signals in fMRI) are transformed into sensory feedback, allowing a person to learn—through practice and reinforcement—how to modulate those signals over time. The promise has been to support self-regulation in conditions such as ADHD, depression, and anxiety, and to enhance cognitive functions like attention or working memory.

Yet every neurofeedback session is about more than just the targeted brain rhythm or ROI. It also recruits self-referential processes ("am I doing this right?"), interoceptive attention (tuning into subtle inner cues), expectations of control, and broader cognitive control systems that calibrate effort, sustain focus, and adjust strategies. This study isolates those general mechanisms by using a sham feedback paradigm: participants think they are controlling a moving bar with their brain activity, but in reality the bar is a prerecorded EEG neurofeedback session.

By comparing active attempts to control the bar with simply watching it move, the authors identify the neural substrates of cognitive control and self-related processing that are automatically engaged in NF-like contexts, regardless of whether genuine learning is happening. As we will see, this has direct implications for clinical practice, client expectations, and how we design and interpret neurofeedback interventions.

Methods

Participants and overall design

Twenty healthy adults (10 female, 10 male, aged 40–63 years) with no neurological or psychiatric history took part in the study. All were neurofeedback novices and naïve to the true purpose of the experiment. During fMRI scanning, participants viewed visual feedback in the form of three colored bars and were told that these bars represented their own brain activity in real time.

Critically, no real neurofeedback was provided. Instead, the bar movements were based on prerecorded EEG sensorimotor rhythm (SMR) neurofeedback sessions taken from prior trainings. These recordings were cleaned for artifacts by two independent EEG experts, smoothed with a 1-second moving average, and then replayed at 20 updates per second to generate a naturalistic NF-like display. The middle bar corresponded to the "target" signal, while the two side bars represented activity to be suppressed, mimicking classic SMR training instructions.

Experimental conditions

The paradigm used three block-designed conditions:

1. Get control: participants were instructed to voluntarily control the movement of the bars using their brain activity. Their goal was to increase the height of the central bar and reduce the two lateral bars. They were told to relax, stay focused on the feedback, and avoid overt movements—very similar to typical EEG-NF instructions.

2. Watch moving bars: participants passively watched the same moving bars without attempting to control them.

3. Watch static bars: participants viewed static bars, providing a low-level visual control condition.

Each condition was repeated five times. A typical trial began with a fixation cross (around 17–20 s), followed by a cue (“control” or “watch”; 3 s), and then a 20 s block of the relevant bar display. In the get control condition only, participants rated their perceived success after each block on a 5-point scale ranging from "no control" to "full control."

Questionnaires and individual differences

After scanning, participants completed measures of:

• Locus of control for technology (KUT), indexing how confident they generally feel in dealing with technological systems.

• Rumination (Ruminative Response Scale, short version), with subscales for reflective pondering and brooding.

These were used to examine relationships between personality-style variables and perceived control over the neurofeedback-like task.

MRI acquisition and analysis

Functional images were acquired on a 3T Siemens Skyra using a T2*-weighted sequence (TR 920 ms, TE 30 ms, 23 slices, 4 × 4 × 4 mm voxels) with whole-brain coverage. Standard preprocessing (motion correction, slice-timing correction, normalization to MNI space, 8 mm smoothing) was conducted in SPM8.

Block regressors were created for fixation, cue, each of the three bar conditions, and the rating period. The main contrasts of interest at the group level were:

• Get control > watch moving bars (cognitive control and self-referential engagement over and above processing of movement).

• Watch moving bars > get control (regions more engaged when bars are watched without active control).

• Watch moving bars > watch static bars (general response to movement and to seeing one’s supposed brain activity).

Whole-brain analyses were thresholded at p < .001 uncorrected with cluster-level FDR correction at p < .05 and a minimum of 10 voxels.

Results

Behavioural findings: perceived control and control beliefs

Participants, on average, reported a moderate sense of control over the bars during the get control condition (mean rating 2.69 on a 1–5 scale, where 3 represented “medium control”; range 1–4). During debriefing, none of the participants suspected that the feedback was sham, suggesting that the NF illusion was convincing.

Interestingly, there was a significant negative correlation between perceived control and the KUT score (locus of control for technology). Those who generally felt more confident with technology (higher KUT) rated their ability to control the bar as lower. In other words, the more someone usually trusts their ability to handle devices, the more accurately they seemed to detect that this particular device was not responding to them.

Scores on rumination (reflection and brooding) were not meaningfully related to perceived performance.

Neural correlates of attempting to control the feedback

The main contrast—get control > watch moving bars—revealed a broad frontoparietal and cingulo-opercular network, including:

• Bilateral anterior insula (strongest peak on the left), extending into operculum.

• Anterior cingulate cortex (ACC).

• Bilateral supplementary motor area (SMA).

• Dorsomedial and dorsolateral prefrontal cortex.

• Right superior parietal lobule and left supramarginal gyrus.

• Left thalamus.

Table 2 and the activation map on page 6 show these clusters distributed across classical cognitive control and interoceptive/self-related regions. The authors interpret this as a neural signature of cognitive control under the belief of doing NF.

In the reverse contrast, watch moving bars > get control, only the left angular gyrus showed stronger activation. This region is commonly associated with the default mode network and self-related processing during rest, but also with external-agency attribution.

Processing moving vs static bars

The contrast watch moving bars > watch static bars—essentially "what happens when my ‘brain activity’ moves on the screen"—showed extensive bilateral activation in:

• Inferior and superior parietal cortex.

• Temporal cortex.

• Posterior insula.

• Inferior frontal gyrus.

• Middle occipital regions and fusiform gyrus.

• Precentral gyrus and supramarginal gyrus.

The brain slices in Figure 3 (page 6) illustrate this widespread visual and attention network, which appears regardless of whether people are trying to control the display.

Discussion

This study elegantly disentangles two layers of what happens in a neurofeedback context: simply seeing feedback that is purported to represent your brain, and actively trying to control it. Even without genuine contingency between brain and display, the intention to self-regulate is enough to recruit a robust cognitive control network.

The bilateral anterior insula stands out as a central hub. This region is closely linked with interoceptive awareness and self-related processing, integrating internal bodily states with external cues. In a neurofeedback setting, the insula seems to serve as a comparator: "How do my inner cues match what I see on the screen?" When participants try to control the bars, this comparison process appears to intensify.

The anterior cingulate cortex and dorsomedial/dorsolateral prefrontal cortex, also strongly activated, fit with a classic role in monitoring discrepancies between actual and desired states, allocating attention, and adjusting strategies. In NF language, these areas are likely helping to track “am I getting more of the desired signal?” and adjust effort or mental strategy—even when the feedback is actually unrelated to the person’s brain.

The supplementary motor area and thalamus add to this picture of a centralized control system that supports sustained attention and preparation for responses (including the performance ratings). The superior parietal and supramarginal regions suggest the involvement of spatial attention and integration of visual feedback into body- and self-representations.

In contrast, when participants simply watched the moving bars without trying to control them, the left angular gyrus became relatively more active. This structure is often associated with the default mode network and with attributing events to external rather than self-generated causes. When the instruction removes the need to exert control, the system appears to shift toward a more passive, externally attributed interpretation of the display.

The behavioural relationship between perceived control and locus of control for technology is clinically and practically relevant. Individuals who typically feel competent with devices may be more sensitive to a lack of real contingency in NF, and thus report lower perceived control when the system does not respond as expected. In real clinical NF, this could translate into early frustration or dropout if the training is not designed to provide clear, meaningful feedback and early experiences of success.

Taken together, the findings suggest that:

• A large amount of the brain activity we see during NF is about general cognitive control and self-referential processing, not just the specific rhythm or region being trained.

• The belief that the display reflects one’s brain, and the instruction to control it, are sufficient to engage these networks—even when genuine learning cannot occur.

• Individual expectations and control beliefs modulate how people experience their own performance.

For people considering neurofeedback, this underscores that NF is not just a “brain gym” but also an experiential process involving attention, self-awareness, and the felt sense of agency. For clinicians and referring professionals, it highlights the importance of psychoeducation around expectations, the role of early successes, and how the training environment can either support or undermine self-efficacy.

From a professional NF standpoint, these data invite us to think more deliberately about how we harness and shape these control networks. For example, some protocols deliberately encourage effortless attention and low cognitive load (e.g., SMR uptraining with simple reward animations), while others may inadvertently promote over-effort and frustration, especially in highly driven or tech-confident clients. The present study suggests that trying too hard, in the absence of clear contingencies, may chronically engage control networks without producing efficient learning.

More broadly, the work fits into an emerging model of neurofeedback as a layered process: implicit learning within the targeted networks, embedded in a scaffold of explicit cognitive control, interoception, and agency. Future studies combining EEG-NF with simultaneous fMRI could help clarify how these layers interact—and how best to design training that respects the limits of each.

Brendan’s perspective

One of the things I love about this study is that it quietly exposes something every neurofeedback clinician has seen in the chair: sometimes, people work incredibly hard in training, feel like they’re doing everything “right,” and still say, “I don’t think I’m controlling this.” Other times, people float through sessions, claim they’re not doing much at all, and yet their brain metrics and symptoms shift beautifully.

What Ninaus and colleagues show is that the moment you tell someone, “That thing on the screen is your brain, try to control it,” you light up a whole set of regions dedicated to effort, monitoring, self-awareness, and error correction. And you do this even before you know whether the feedback is actually contingent.

In clinical EEG neurofeedback, this has a lot implications (and here are just a few).

First, there is always a background level of cognitive control “cost” baked into any training. When I ask someone to uptrain SMR at C3/C4 (12–15 Hz over the sensorimotor strip) for better behavioural inhibition or sleep, I’m not just shaping SMR. I’m also asking their anterior insula, ACC, and frontoparietal networks to hold a particular relationship between inner state and outer feedback. If I pile on complex instructions (“try to relax but focus, think positive but not too hard, imagine your body is heavy but stay awake…”), I may be overloading that system.

Second, client expectations and technological self-confidence matter. Someone who scores high on a KUT-like measure (the person who troubleshoots everyone’s Wi‑Fi at work) is used to systems responding predictably when they press the right buttons. If their brain does not seem to drive the display quickly, they may conclude, “This isn’t working,” even if subtle learning is occurring under the hood. For these clients, I would:

• Use very clear reward structures (e.g., thresholded SMR or low-beta increases with immediate audiovisual reinforcement) and avoid long stretches with ambiguous feedback.

• Front-load psychoeducation on what “learning” looks like (small, sometimes noisy changes over sessions) and why their subjective sense of control might lag behind genuine neuroplastic change.

• Consider briefer blocks with more opportunities to see wins; say, 30–60 second trials with explicit summary feedback, rather than long continuous runs.

Third, the study nudges us toward lighter, more permissive instructions. Because the cognitive control network will engage as soon as we create a control task, we can afford to simplify the client’s job. Instead of “try to raise your SMR and lower theta and high beta while staying perfectly still and focused,” we might say, “Notice what it’s like when the spaceship flies smoothly; let your brain find more of that.” In practice, that might mean:

• For SMR training at C4 in a hyperactive child: emphasising stillness and a calm but alert body, with a simple game where smoother, less jerky movements indicate success.

• For high beta downtraining at Fz or Cz in an anxious adult: using feedback that rewards small reductions in 22–30 Hz while pairing it with slow breathing or body-based cues, so the inner sensations that co-occur with success become easier to notice.

Even though this study uses fMRI and sham feedback, it resonates with EEG choices. If the insula and ACC are acting as comparators between inner state and outer signal, I want my EEG feedback to be as clean and interpretable as possible. That means:

• Good impedance control and artifact management, so the feedback is genuinely about the targeted rhythm, not eye blinks and jaw tension.

• Thoughtful thresholding and adaptation rules to avoid long periods where the client cannot produce reward events.

• Stimuli that are engaging but not cognitively demanding (e.g., simple animations, natural scenes, or auditory tones), leaving the heavy lifting to implicit learning rather than explicit strategy use.

I also see a clear link to protocol individualisation. Not everyone needs the same level of cognitive engagement. A ruminative, overthinking adult with high frontal beta and insomnia may benefit from protocols and instructions that deliberately de-emphasise effort—perhaps alpha uptraining at POz with eyes-closed auditory feedback, combined with mindfulness coaching to notice but not chase sensations. An underaroused, inattentive teenager, by contrast, might respond well to slightly more task-like, gameified feedback during SMR or low beta uptraining at C3/Cz/C4, where engagement itself is part of the therapeutic target.

The angular gyrus finding also whispers something interesting: when we explicitly tell clients “just watch,” the brain seems to shift toward an externally attributed mode. In practice, I sometimes use this dyad deliberately within a session: alternating blocks where the client actively “plays” the feedback with blocks where they simply observe. This can help people who otherwise try too hard to let the system work on them a bit, reducing cognitive strain while still exposing the brain to contingent feedback.

Finally, clinical reality is messier than the tidy sham setup in the scanner. Our clients see symptom changes, shifts in sleep, emotional resilience, and performance; outcomes that depend on many hours of repeated, contingent training. But this paper is a useful reminder that when research studies report “no difference between real and sham NF,” we should look carefully at what, exactly, was trained and for how long, and how much of the observed activation might simply reflect the shared cognitive control scaffold rather than specific self-regulation of a target rhythm or network.

In day-to-day practice, the take-home for me is this: design neurofeedback so that the brain can learn implicitly, while the person feels gently supported rather than under exam conditions. If we respect the cognitive control systems as finite resources—not the sole engine of change—we can build protocols that are more sustainable, more humane, and ultimately more effective.

Conclusion

By using sham feedback in an fMRI paradigm, Ninaus and colleagues show that the mere belief of doing neurofeedback is enough to recruit a wide network of regions involved in cognitive control, interoceptive awareness, and self-related processing. Anterior insula, anterior cingulate, SMA, and dorsomedial/dorsolateral prefrontal areas all come online when people attempt to “drive” a feedback display they think reflects their brain, whereas more passive, externally attributed processing engages the angular gyrus.

These findings remind us that neurofeedback is always a layered process: targeted modulation of specific brain signals sits on top of a more global scaffold of motivation, agency, expectation, and effort. For clinicians and trainers, this means that how we frame the task, how early we provide experiences of success, and how we titrate cognitive load are not cosmetic details—they directly shape the neural context in which learning occurs. When we get this right, neurofeedback can become less of a struggle to control and more of a guided conversation between brain, body, and feedback.

References

Ninaus, M., Kober, S. E., Witte, M., Koschutnig, K., Stangl, M., Neuper, C., & Wood, G. (2013). Neural substrates of cognitive control under the belief of getting neurofeedback training. Frontiers in Human Neuroscience, 7, 914. https://www.frontiersin.org/articles/10.3389/fnhum.2013.00914/full