Evidence Review - What Music Actually Does (and Doesn't Do)

On this page

This document is the scientific foundation for every playlist spec and every engine recipe in CognitionFM. Each claim gets one of four verdicts:

Tier Meaning
WELL-SUPPORTED Multiple RCTs or meta-analyses, consistent direction, plausible mechanism
MODERATE / MIXED Real signal but inconsistent, small, or heavily moderated by person/task
WEAK / SPECULATIVE A few small studies, methodological problems, or effect likely non-specific
DEBUNKED / MARKETING No credible evidence, or evidence directly contradicts the claim

Every citation below was verified against the source in July 2026. Where marketing language circulates ("neuroscience-backed," "healing frequency"), the gap between the actual finding and the claim is stated explicitly.

1. The mechanisms that actually matter

Four mechanisms explain most reliable music effects. Everything else in this document is a special case of these.

1.1 Arousal and mood, not "smart music" - WELL-SUPPORTED

The famous "Mozart effect" dissolved under scrutiny. Thompson, Schellenberg & Husain (2001) showed the spatial-task benefit appears after any pleasant, energetic music and disappears when arousal and mood are controlled - a slow, sad Albinoni adagio produced no benefit. Pietschnig, Voracek & Formann's meta-analysis (2010, Intelligence) of ~40 studies / 3,000+ subjects found publication bias, inflated effects from a single lab, and "little evidence for a specific, performance-enhancing Mozart effect."

Design consequence: music helps performance by putting you at the right arousal level for the task - not by transferring information into your brain. Choose music by target arousal, not by composer prestige.

1.2 Attention capture: the irrelevant sound effect - WELL-SUPPORTED

Sounds that change unpredictably ("changing-state" sounds: speech, lyrics, staccato melodies, novel timbres) disrupt serial recall and verbal short-term memory even when you ignore them (Salamé & Baddeley 1982; Jones & Macken's changing-state work - overview, recent evidence). Two robust details:

Design consequence: for any task holding words or ordered items in working memory (writing, reading, debugging a specific line), background audio must minimize acoustic change: no lyrics, no sharp transients, no dramatic melodic contour, high self-similarity. This is the single most load-bearing finding in this system.

1.3 Task complexity × music salience - WELL-SUPPORTED

Kämpfe, Sedlmeier & Renkewitz's meta-analysis (2011, Psychology of Music) found a global null: averaged across everything, background music does nothing. Disaggregated, it hurts reading and memory and helps emotion and sports. Gonzalez & Aiello (2019, JEP: Applied) sharpened this: salient music (louder, more layers) improved simple vigilance tasks and impaired complex ones - and boredom-prone people who most want music while working were the most distracted by it. A 2022 systematic review (Cheah et al., Music & Science) confirms the same task/music/person interaction structure.

Design consequence: there is no one "focus playlist." Simple/boring work tolerates (and benefits from) stimulating music; complex work needs minimal or no music. The system therefore splits work playlists by cognitive load.

1.4 Preference, familiarity, and personality moderate everything - WELL-SUPPORTED

Furnham & Bradley (1997) found introverts suffer more from background music than extraverts on recall and reading. Preference cuts both ways: liked music improves mood and persistence, but Perham & Currie (2014) showed preferred lyrical music impaired reading comprehension as much as disliked lyrical music - liking a song does not protect you from its lyrics.

Design consequence: the system calibrates to the individual (see 03-testing-protocol.md) and never assumes a universal prescription. But preference is not a free pass for lyrics during verbal work.

2. Verdicts by use case

2.1 Deep work: writing, reading, language-heavy tasks

Verdict: music with lyrics or high variability - WELL-SUPPORTED as harmful. Quiet, steady instrumental music - MODERATE (roughly neutral; helps mood/persistence, may slightly cost accuracy vs. silence).

Silence is the performance ceiling for complex verbal work in most lab studies (Kämpfe 2011; Perham & Currie 2014). Steady instrumental sound earns its place not by boosting cognition but by masking worse noise (office chatter is a changing-state sound), stabilizing mood, and making long sessions tolerable. That is a real, defensible benefit - just not "music makes you smarter."

2.2 Deep work: coding and analytical tasks

Verdict: MODERATE. Coding is heterogeneous: reading unfamiliar code is language-heavy (rules above apply); routine implementation in a familiar codebase is lower verbal load and tolerates a steady pulse. No strong coding-specific literature exists; this is extrapolation from the task-complexity findings (Gonzalez & Aiello 2019), flagged as such. Steady, predictable, instrumental pulse music is a reasonable default; drop to ambient or silence when reading/debugging hard code.

2.3 Boring/admin work

Verdict: WELL-SUPPORTED. The clearest positive case for stimulating music: simple, repetitive tasks benefit from added arousal (Gonzalez & Aiello 2019; the arousal-mood mechanism). Lyrics and favorites are fine here - the task doesn't compete for verbal working memory.

2.4 Creative thinking

Verdict: MIXED. Some studies find positive-mood music broadens ideation; others (e.g., Threadgold et al. 2019) found background music impaired verbal insight problem-solving. The honest summary: moderate arousal and good mood help divergent thinking; verbal creative work is still verbal work. Treat creative-session music as mood scaffolding before/between thinking, not during verbal ideation.

2.5 Stress reduction and recovery

Verdict: WELL-SUPPORTED - one of the strongest literatures. de Witte et al. (2020, Health Psychology Review) ran two meta-analyses over 100+ studies: music interventions produced significant small-to-medium reductions in both physiological (heart rate, blood pressure, cortisol) and psychological stress outcomes. A companion meta-analysis on therapist-delivered music therapy (de Witte et al. 2022) found a medium-to-large effect (d ≈ 0.72). Slow tempo (~60-80 BPM), low dynamic range, and predictability recur as the effective ingredients. Note the effect does not depend on any special frequency or genre - preferred calm music works.

2.6 Emotional processing

Verdict: MODERATE - real but person-dependent, with a known failure mode. Taruffi & Koelsch (2014, PLOS ONE) (n=772) found sad music reliably provides emotion regulation, consolation, and imaginative reward - nostalgia, not raw sadness, is the most common evoked emotion. Benefits are larger for high-empathy listeners. The failure mode: for ruminators, sad music can entrench low mood rather than process it (mood-regulation styles literature, Saarikallio and others). Use deliberately, time-boxed, not as all-day background.

2.7 Walking and light exercise

Verdict: WELL-SUPPORTED - the strongest positive case in the whole literature. Terry, Karageorghis et al.'s meta-analysis (2020, Psychological Bulletin, 139 studies) found music improves affective valence (g = 0.48), physical performance (g = 0.31), reduces perceived exertion (g = 0.22), and even improves oxygen-use efficiency (g = 0.15). Fast tempo beats slow; personally motivational, familiar music beats generic. This is where your favorite energetic vocal music belongs.

2.8 Sleep

Verdict: WELL-SUPPORTED for subjective sleep quality; mechanism is mundane. The Cochrane review (Jespersen et al. 2022, 13 RCTs, n=1,007) found moderate-certainty evidence that music listening improves subjective sleep quality in insomnia, with low-certainty evidence for faster sleep onset and better efficiency. The plausible mechanism is relaxation + pre-sleep routine/conditioning - not "delta-wave entrainment." Sedative-music criteria across trials: slow (~60-80 BPM), smooth onsets, narrow dynamics, no lyrics, low volume, used consistently at the same point in the wind-down.

2.9 Deliberate mood lift

Verdict: WELL-SUPPORTED for short-term mood, MODERATE for lasting effect. Music is one of the most reliable laboratory mood-induction tools, and Ferguson & Sheldon (2013) found the combination that matters: upbeat music + the intention to feel better improved mood, while passive listening alone did not (and over-monitoring your own mood undermines it). Two studies, so the durability claim is moderate - but the short-term effect rests on a much older mood-induction literature.

2.10 Anger processing - matching beats soothing

Verdict: MODERATE (small studies, consistent with music-therapy practice). Sharman & Dingle (2015) anger-induced 39 extreme-metal listeners: their own aggressive music did not make them angrier - hostility fell as much as in silence, while positive activation rose. This supports the music-therapy "iso principle": match the current arousal first, then shift - don't force calm music onto a furious brain. Caveats: small n, fans only; matching works with music you identify with.

2.11 Intimacy and sexual arousal

Verdict: WEAK-to-MODERATE - plausible mechanisms, thin direct literature, preference and context dominate. The only well-identified mechanism is excitation transfer: arousal from one source (music) is misattributed to another. Marin et al. (2017, PLOS ONE) showed arousing music increased sexual-attractiveness ratings of faces (in women, in a lab). Beyond that: conditioning (music tied to shared history carries the association), self-consciousness masking (background sound reduces inhibition the same way it masks office noise), and the folk consensus on slow, bass-heavy, smooth-timbred grooves - which is genre convention, not a tested prescription. No study validates any specific "aphrodisiac" genre, tempo, or (obviously) frequency. Honest design guidance: personal/shared associations first, general arousal + low awkwardness second, everything else is taste.

3. Skeptic's corner

3.1 Binaural beats - MIXED, and wildly oversold

The honest reading is messier than either camp admits. Garcia-Argibay et al.'s meta-analysis (2019, Psychological Research) reported a medium pooled effect (g = 0.45) across cognition, anxiety, and pain. But Basu & Banerjee's 2022 meta-analysis found mixed results for memory and attention with major methodological variability, and recent well-powered work (e.g., a 2025 parametric study in Scientific Reports) found no convincing sustained-attention enhancement. Underlying studies are small, blinding is nearly impossible, and the "brainwave entrainment" mechanism is far weaker at the cortex than marketing implies. Verdict: possibly a small real effect, mechanism unclear, nowhere near the "40 Hz unlocks genius mode" claims. CognitionFM does not use binaural beats and does not need to.

3.2 432 Hz / 528 Hz / solfeggio frequencies - DEBUNKED / MARKETING

No credible evidence that any specific tuning frequency has healing or cognitive properties. The "ancient solfeggio scale" is a 1990s numerological construction; ancient cultures could not measure Hertz. The handful of small studies (e.g., on 432 Hz tuning) are underpowered, poorly controlled, and unreplicated. When a slow 528 Hz drone relaxes someone, it's because slow, predictable, low-variability sound relaxes people (§1.2, §2.5) - the number is irrelevant. CognitionFM tunes to A=440 like everyone else and says so publicly.

3.3 "Dopamine music" - MARKETING (a real finding, misread)

Salimpoor et al. (2011, Nature Neuroscience) did show striatal dopamine release during musical chills - for music the listener already loved, at peak emotional moments. That is a finding about musical pleasure, not a productivity lever. "Boost dopamine to focus" playlists misappropriate it; intensely pleasurable music is, if anything, more attention-capturing during complex work (§1.2).

3.4 Lo-fi hip-hop - the properties are fine, the mythology is not

No special "lo-fi mechanism" exists. Lo-fi works for many people because it is incidentally well-designed for the changing-state findings: no lyrics, narrow dynamic range, repetitive structure, moderate tempo, familiar harmonic language. Any music with those properties does the same job. (That property list is, not coincidentally, close to CognitionFM's deep-work recipe.)

3.5 White/pink noise and "brown noise for ADHD" - MODERATE for noise-masking, WEAK for enhancement

Steady broadband noise is the theoretical ideal steady-state sound: maximum masking, zero changing-state content. Evidence supports it for masking disruptive environments; evidence for cognitive enhancement (including the stochastic-resonance/ADHD line) is small and inconsistent. It's a legitimate, boring alternative to music for verbal deep work - worth including in personal A/B tests.

4. Claims we may publicly make (and their honest phrasing)

This table feeds every public title and description, wherever the mixes are published. Nothing stronger than the right column ever ships.

Underlying finding ✅ Honest public phrasing ❌ Never say
Changing-state / lyrics disrupt verbal work "Designed around research on why lyrics and sudden changes break focus" "Scientifically proven to boost focus"
Arousal-mood hypothesis "Built to hold a calm, steady level of alertness" "Activates your brain's focus network"
de Witte stress meta-analyses "Slow, predictable music measurably reduces stress in controlled studies" "Melts cortisol away"
Jespersen Cochrane sleep review "Music as part of a wind-down routine improves sleep quality in clinical trials" "Delta waves put you into deep sleep"
No frequency magic "Tuned to standard A=440 - no healing-frequency claims here, because the evidence isn't" anything with "432 Hz," "528 Hz," "solfeggio," "Hz healing"
Preference/person moderators "Test it against silence - this works for many people, not everyone" "Works for everyone"

5. Reference list