I Say So
What the research on self-affirmation actually says, and what it does not
Search for whether affirmations work and you will find page after page citing brain scans and university studies. Most cite one body of research to support a different practice. The research is about a writing exercise on personal values. The practice being sold is repeating statements about yourself, sometimes in your own recorded voice. The evidence does not transfer from one to the other.
This page separates them. Every study is named with authors, journal, year, sample size and what was measured. Findings that went against the practice are reported as carefully as the supportive ones. Claims we could not trace to a published study are marked as untraced. I Say So plays affirmations back in your own recorded voice, so it is worth saying at the start that we found no controlled trial testing whether own-voice playback changes any outcome.
Two different things share a name. Self-affirmation theory, a values writing exercise, has a real evidence base. Repeating positive statements, which is what affirmation apps sell, does not have the same support, and no controlled trial compares your own voice against anyone else's.
Two different things share a confusing name
Self-affirmation theory in social psychology is about reflecting on a core personal value. In the standard procedure a participant ranks a list of values, picks the top one, and writes a short paragraph about why it matters to them. The values are things like family, art or politics, not statements about being confident or lovable. Control participants usually write about a value they ranked low.
Positive affirmations, as the self-help practice, means repeating declarative statements about yourself, such as I am confident, sometimes read aloud, sometimes recorded and played back.
Most pages claiming research support for the second are describing studies of the first. The two differ in content, format and in what gets measured. Keeping them apart is the only way to read this literature.
What the value-affirmation studies measured
Cohen, Garcia, Apfel and Master (2006, Science) ran two randomised field experiments in a suburban north-eastern US middle school, involving 119 African American and 124 European American seventh graders. Students wrote a short in-class paragraph about their most important value, or in the control group their least important one. The outcome was course grades, and the paper reports a roughly 40 percent reduction in the racial achievement gap in the targeted course.
Creswell and colleagues (2005, Psychological Science) had 85 undergraduates complete a value-affirmation or control task before a laboratory stress challenge, and measured salivary cortisol. Control participants showed a significant rise 20 minutes after stress onset, t(32) = 4.49, p below .0001. Affirmed participants did not, t(38) = 0.85, p = .40. Heart rate and blood pressure did not differ.
Creswell, Dutcher, Klein, Harris and Levine (2013, PLOS ONE) tested 80 students from two Pittsburgh universities, 73 after exclusions, on 30 Remote Associates Task items under evaluative pressure. Affirmation improved performance overall, beta = .31, p = .005, with a stress by affirmation interaction, beta = .35, p = .041. The benefit appeared mainly in those reporting high chronic stress.
Cascio and colleagues (2016, Social Cognitive and Affective Neuroscience) scanned 67 recruited sedentary adults, 33 affirmed and 28 control. Affirmed participants showed more valuation-network activity for value versus everyday scenarios, t(57) = 2.43, p = .018, mainly for future-oriented values. The whole-brain analyses are described as exploratory. All four used a values writing or reflection task. None tested repeating statements about yourself.
Where self-affirmation did not hold up
Hanselman, Rozek, Grigg and Borman (2017, Journal of Educational Psychology) ran a large-scale replication across 11 middle schools in one Midwestern US district, with 1,269 consented students and a Black and Hispanic analytic sample of 449. The grade 8 GPA effect was d = -0.072 with a standard error of 0.058, and the grade 7 effect was effectively zero at d = -0.002. The authors state they found no evidence of effects, and that their estimates were precise enough to reject benefits of 0.10 standard deviations or larger. An earlier cohort in the same project had shown d = 0.152, and the two differed significantly, p = .013. They describe the intervention as potentially fragile at scale.
Pooled estimates are small. Epton, Harris, Kane, van Koningsbruggen and Sheeran (2015, Health Psychology) meta-analysed 144 experimental tests of self-affirmation and health messages, reporting d+ = .17 for message acceptance, d+ = .14 for intentions and d+ = .32 for behaviour. A 2025 meta-analysis in American Psychologist by Zhang, Chen, Wang and Hu pooled 129 studies with 17,748 participants and reported small positive effects on well-being, with average follow-up of about two weeks and 89 studies using college students.
The picture is a real effect in places, small on average, and one well-powered field replication that found none.
What happens when people repeat statements about themselves
The study that most directly tests the self-help practice is Wood, Perunovic and Lee (2009, Psychological Science 20, 860 to 866). In two experiments, participants with low self-esteem who repeated a statement such as I am a lovable person, or who focused on how it was true, reported worse mood than those who did neither. Participants with high self-esteem felt better, but only modestly.
The authors' proposed mechanism is that a statement conflicting with someone's self-view prompts contradictory thoughts that outweigh the positive ones. That finding is more than fifteen years old and is still the clearest direct test of repeated positive self-statements. It cuts against the practice, and deserves the same weight as any supportive finding.
What is known about hearing your own voice
There is a real literature on self-voice processing. Hearing your own voice engages self-related processing regions, and lesion work suggests right-hemisphere damage can impair discriminating it. Behavioural recognition studies are mixed. A recording also lacks the bone-conducted sound you hear while speaking, so it does not match the voice you are used to.
The one study that puts own voice and affirmation together is Jo and colleagues (2024, Brain Sciences 14(7), 637). Twenty-one adults, mean age 25.6, were scanned while listening to 60 sentences, each in their own synthesised voice or an unfamiliar voice, across self-affirmation, cognitive defusion and neutral sentence types. The behavioural measure was a rating of how strongly each sentence affected the listener's emotions. Sentence type mattered, F(2,61) = 14.72, p below .001. Voice identity did not, p = .84, and there was no interaction, p = .99.
Brain activity did show a voice by strategy interaction, stronger for cognitive defusion than for self-affirmation. The study measured no mood and no behaviour, and the authors note that synthesised speech in a scanner is a loose proxy for real self-talk.
What we could not find
We could not find a published controlled trial comparing affirmations in a person's own recorded voice against the same affirmations in another voice, on mood, self-esteem, stress or any behavioural outcome. If such a trial exists, we have not located it.
The nearest work is thin. Tzvieli and colleagues (2017, Advances in Mind-Body Medicine 31(2)) had 86 enrolled adults, 56 completing all tests, record statements of love for themselves and listen for two minutes each morning for 12 weeks. Mean clinical anger scores fell from 10.4 to 6.7, p below .001. There was no control group and no comparison voice, so the change cannot be attributed to the recordings.
Bauer, Schindler-Gmelch, Gerczuk, Schuller and Berking (2025, Behaviour Research and Therapy 184, 104667) randomised 144 participants and found that depressed participants told to speak anti-depressive self-statements with more vocal conviction showed a larger drop in depressed mood than controls, with no effects in the sub-clinical or non-depressed groups. That trial is about speaking aloud, not about listening to a recording. On recorded self-help audio generally, Greenwald, Spangenberg, Pratkanis and Eskenazi (1991, Psychological Science 2(2)) ran double-blind tests of subliminal tapes sold for memory and self-esteem and found no claimed effects.
The claims we could not trace
Two claims appear on almost every affirmation page and we could not trace either to a published study. The first is that research shows affirmations work better when they are personal. We found no study comparing personally worded affirmations with generic ones on an outcome measure. The idea is a plausible reading of self-affirmation theory, which does use a participant's own top-ranked value, but that is not a test of wording.
The second is that hearing your own voice produces a distinct brain response which makes affirmations more effective. The first half has support in the self-voice literature. The step from brain response to effectiveness traces only to vendor marketing. The one study to test own voice against another voice using affirmation sentences, Jo and colleagues in 2024, found no difference in how strongly the sentences affected listeners.
I Say So is built on own-voice playback. Nothing here is evidence that it works. We found no evidence either way, and we would rather say so than cite research that studied something else.
Questions
Is self-affirmation the same thing as positive affirmations?
No. Self-affirmation, as studied in social psychology, usually means a short writing exercise in which a person reflects on a core personal value such as family or art. Positive affirmations, as the self-help practice, means repeating declarative statements about yourself. The research base belongs to the first and is routinely cited for the second.
What did the famous self-affirmation studies actually measure?
Grades in a middle-school field experiment (Cohen and colleagues, 2006), salivary cortisol after a laboratory stressor in 85 undergraduates (Creswell and colleagues, 2005), performance on Remote Associates Task items under pressure in 73 analysed students (Creswell and colleagues, 2013), and brain activity in a valuation network in a scanner study of sedentary adults (Cascio and colleagues, 2016). All used a values writing or reflection task.
Have self-affirmation findings replicated?
Unevenly. Hanselman and colleagues (2017) ran a replication across 11 middle schools with a Black and Hispanic analytic sample of 449 and reported no detectable effect, with estimates precise enough to rule out benefits of 0.10 standard deviations or more. Meta-analyses find small positive average effects, around d+ = .14 to .32 for health-message outcomes in Epton and colleagues (2015).
Can repeating positive statements make someone feel worse?
That is what Wood, Perunovic and Lee (2009) reported in Psychological Science. Across two experiments, participants with low self-esteem who repeated a positive self-statement reported worse mood than those who did not. Participants with high self-esteem felt modestly better. The authors suggest a statement that conflicts with a person's self-view can prompt contradictory thoughts.
Is there evidence that hearing affirmations in your own voice works better?
We did not find any. The closest study, Jo and colleagues (2024) in Brain Sciences, scanned 21 adults listening to self-affirmation and other sentences in their own synthesised voice or an unfamiliar voice, and found no main effect of voice identity on how strongly the sentences affected listeners, p = .84. It measured no mood or behavioural outcome. We found no controlled trial comparing own-voice playback with another voice on any outcome.
Why does a voice affirmation app say the evidence for voice affirmations is thin?
Because that is what the literature shows, and a page that reviews the evidence and then exempts its own product would not be worth reading. I Say So records and plays back affirmations in your own voice. The research on value affirmation studied something else, and the specific own-voice claim traces to marketing rather than to published outcome data.
Sources, in order of appearance. Cohen, Garcia, Apfel and Master (2006), Reducing the racial achievement gap, Science 313, 1307-1310, doi 10.1126/science.1128317. Creswell, Welch, Taylor, Sherman, Gruenewald and Mann (2005), Affirmation of personal values buffers neuroendocrine and psychological stress responses, Psychological Science 16(11), 846-851, PMID 16262767. Creswell, Dutcher, Klein, Harris and Levine (2013), Self-affirmation improves problem-solving under stress, PLOS ONE 8(5), e62593. Cascio, O'Donnell, Tinney, Lieberman, Taylor, Strecher and Falk (2016), Social Cognitive and Affective Neuroscience 11(4), 621-629, PMC4814782. Cohen and Sherman (2014), The psychology of change, Annual Review of Psychology 65, 333-371, doi 10.1146/annurev-psych-010213-115137, cited for context only. Hanselman, Rozek, Grigg and Borman (2017), Journal of Educational Psychology 109(3), PMC5403146. Epton, Harris, Kane, van Koningsbruggen and Sheeran (2015), Health Psychology 34(3), 187-196, PMID 25133846, effect sizes read at abstract level and author order differs between database listings. Zhang, Chen, Wang and Hu (2025), American Psychologist, doi 10.1037/amp0001591, study counts verified from the APA release and effect sizes read at abstract level only. Wood, Perunovic and Lee (2009), Psychological Science 20, 860-866, PMID 19493324, read at abstract level so no per-experiment sample sizes are stated. Jo and colleagues (2024), Brain Sciences 14(7), 637, doi 10.3390/brainsci14070637, PMC11274574. Tzvieli and colleagues (2017), Advances in Mind-Body Medicine 31(2), 12-15, PMID 28659509, uncontrolled pre-post design, read at abstract level. Bauer, Schindler-Gmelch, Gerczuk, Schuller and Berking (2025), Behaviour Research and Therapy 184, 104667, doi 10.1016/j.brat.2024.104667, read from the repository record and abstract. Greenwald, Spangenberg, Pratkanis and Eskenazi (1991), Double-blind tests of subliminal self-help audiotapes, Psychological Science 2(2), read at abstract and secondary level. Self-voice background, Kaplan and colleagues on self-related processing of face and voice (PMC2566765), right-hemisphere lesion work on self-voice discrimination (PMC5988014), and Cheung and Babel (2022) in Frontiers in Psychology, all read at abstract level and not the source of any number here. Claims we could not trace, and therefore do not assert. One, that research shows affirmations work better when they are personal, for which we found no study comparing personal with generic wording on an outcome. Two, that an own-voice brain response makes affirmations more effective, where the brain response is documented but the step to effectiveness traces only to vendor marketing. Three, any controlled trial of own-voice affirmation playback against another voice or against no playback on mood, self-esteem, stress or behaviour, of which we found none. Four, Protzko and Aronson (2016), widely listed as a failed replication of Cohen and colleagues (2006), which we saw in a replication index but did not read, so it is not cited as evidence. Five, a frequently repeated claim that eleven minutes of listening to positive recordings raised self-compassion, for which we could not identify the underlying study or whether own voice was involved. This page is descriptive. It reports what studies measured and found. It is not advice, it makes no promise about outcomes, and nothing here is a reason to start or stop any treatment.