When a group of 20 adults living with psychosis gathered in a Yale recording space to co-write and record original songs, the researchers running the sessions were not measuring musical talent. They were tracking something more specific: whether the act of collaborative songwriting could shift the cognitive patterns that fuel paranoid thinking. Over four weekly group sessions, paranoia scores on a validated clinical paranoia measure declined. Hallucination severity, as assessed by a separate clinical scale, did not decrease significantly across the cohort. That split result, published in the journal Psychosis in early 2026, offers the first clinical evidence that structured music-making may target a distinct cognitive pathway rather than dampening psychotic symptoms across the board.
Inside the SING trial
The study, formally called Song-making In a Group (SING), recruited participants between the ages of 18 and 65 who had been diagnosed with schizophrenia or who experienced distressing auditory hallucinations at least once a week. Over a six-week longitudinal protocol, they attended four group sessions where they wrote lyrics together and recorded original tracks. Researchers measured symptoms before, during, and after the sessions using validated clinical scales for both hallucination severity and paranoia. The published paper reports that paranoia scores declined across the songwriting arm, though the specific magnitude of change and the particular paranoia scale used have not been detailed in publicly available summaries of the results. Hallucination scores, measured on a separate validated instrument, did not reach a statistically significant decrease across the cohort, and the paper does not report whether there was a non-significant trend in that direction.
The paranoia reduction stood out precisely because hallucinations held steady. In roughly 3.5 million American adults living with schizophrenia-spectrum disorders, according to the National Institute of Mental Health, paranoia is one of the most socially disabling symptoms. It erodes trust, strains relationships, and often persists even when medications control other features of psychosis. An intervention that specifically eases persecutory thinking, even modestly, would address a gap that current treatments fill unevenly.
Senior author Philip Corlett, a psychiatry researcher at Yale, has spent years building a case that psychosis can be understood through the lens of predictive coding: the brain constantly generates expectations about incoming sensory data, and symptoms emerge when those expectations, or “priors,” overwhelm actual evidence from the senses. Earlier experimental work by Corlett and colleagues demonstrated that conditioned hallucinations can result from overweighted perceptual priors, using computational modeling and neuroimaging to show how the brain tips toward false perception when expectations grow too strong (Powers, Mathys, and Corlett, Science, 2017). A broader theoretical synthesis in Trends in Cognitive Sciences by Corlett, Horga, Fletcher, and others laid out the predictive-processing account of hallucinations in detail. The SING trial is the team’s attempt to translate that framework into a practical, low-cost intervention that could sit alongside medication and psychotherapy.
What the language data suggest
Beyond symptom questionnaires, the researchers used a computational text-analysis tool called Linguistic Inquiry and Word Count (LIWC) to track how participants’ language shifted across sessions. The SING trial registration on ClinicalTrials.gov confirms that language-use assessment via LIWC and computational tasks tied to predictive-processing theory were prespecified outcomes, not measures chosen after the data came in.
One pattern under scrutiny was pronoun use. Published research has established that people with schizophrenia tend to use more first-person singular pronouns (“I,” “me,” “my”) in spoken narratives compared to controls, a linguistic marker associated with reduced social cohesion and heightened self-focus. Under the predictive-processing model, a decrease in “I” statements and a shift toward group-oriented language during songwriting would signal that participants were updating their internal models to better incorporate social context. That kind of recalibration could, in theory, weaken persecutory beliefs that depend on an exaggerated sense of personal threat.
The linguistic findings remain preliminary. Full quantitative data on the magnitude of pronoun shifts has not yet appeared in the published paper, so it is not clear whether the changes were consistent across most participants or concentrated in a few, or whether they tracked closely with the paranoia reductions. Those details will matter for determining whether language analysis can serve as a reliable proxy for cognitive change in future trials.
Where the evidence has gaps
Twenty participants and no control group receiving a non-musical intervention place hard limits on what this study can prove about causation. The broader trial design, as described by Yale’s Belief Lab, included random assignment to multiple conditions: songwriting, karaoke, listening to a personal playlist, and listening to someone else’s playlist. But the published paper focuses on the songwriting arm, and comparative data across conditions has not yet appeared in peer-reviewed form. That leaves a central question unanswered: is it the creative act, the social interaction, or simply structured time with music that drove the paranoia reduction?
Detailed statistical outputs for the LIWC analysis and the behavioral tasks tied to predictive processing have not been released in a publicly accessible format. Without effect sizes, confidence intervals, and individual-level trajectories, independent researchers cannot fully assess whether the paranoia drop was clinically meaningful or shaped by a handful of strong responders. No qualitative interview data or participant accounts have been published alongside the numbers, which means the lived experience behind the results, including how safe or demanding the sessions felt and whether participants noticed changes in how they interpreted ambiguous social situations, remains absent from the public record.
A related and expanded SING study, registered under trial ID NCT05929352, explicitly ties the musical intervention to predictive-processing mechanisms and references prior meta-analytic evidence on the effects of music-based therapies on negative and cognitive symptoms. That trial’s prespecified endpoints and interim results have not yet been posted as of May 2026, so whether the larger design confirms or complicates the initial findings is still unknown. The expanded study may also carry enough statistical power to detect changes in hallucinations, which did not shift significantly in the smaller cohort.
What this means for psychosis treatment
Three tiers of evidence support the claims here. The strongest is the peer-reviewed SING paper itself, which documents the study design, participant criteria, and the split outcome. The second tier consists of the two ClinicalTrials.gov registry entries, which confirm that outcomes were prespecified and protect against the concern that researchers selected their measures after seeing results. The third tier is the broader body of predictive-processing research by Corlett and colleagues, which provides the theoretical scaffolding but does not, on its own, prove that songwriting engages those mechanisms.
The gap between theory and direct evidence matters. The claim that songwriting recalibrates prediction errors is a hypothesis supported by indirect markers, including paranoia reduction and potential pronoun shifts, not by neuroimaging data from the SING participants themselves. Corlett’s earlier work, including the 2017 Science paper on conditioned hallucinations, used brain scans and computational models to establish the overweighting-of-priors mechanism, but the SING trial did not include neuroimaging. Until the expanded trial produces that kind of data, the link between group music-making and prediction processing rests on behavioral proxies.
For people living with psychosis or their families, the practical signal is cautious but specific. The early data suggest that structured, collaborative songwriting may help ease paranoid thinking for some individuals, even when it does not quiet voices or other hallucinations. That positions it as a potential complement to established treatments like antipsychotic medication and cognitive-behavioral therapy, not a replacement. Anyone considering a similar program should do so with clinical guidance, and the researchers themselves emphasize that larger, controlled studies are still needed before firm recommendations can be made.
How a cognitive theory is reshaping creative intervention design
What the SING trial demonstrates most clearly is a method: how a precise cognitive theory can shape a creative intervention and generate testable predictions. By grounding an arts-based program in predictive-processing models and preregistered outcomes, the Yale team has moved past the vague claim that making music is good for mental health. They are asking a narrower, harder question: can writing a song together change the way a person weighs evidence against expectation? The first answer, incomplete but pointed, suggests it might.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
