What percentage of worries never come true?

In two trials of worry outcome journals with people who have generalized anxiety disorder, 91.4% and 89.14% of the feared outcomes did not happen. When they did, people usually coped better than they expected.

How effective is journaling?

Across 146 randomized studies, writing about personal experiences had a real but small average effect (r = .075). A 2023 meta-analysis of 31 trials found a small effect on depression, anxiety and stress that tended to show up weeks later.

How many people keep using meditation apps?

Few. The median 30-day retention for mindfulness and meditation apps was 4.7% in one analysis of real usage, and fewer than 20% of new Medito users were still meditating after 14 days.

How long does it take to form a habit?

Between 18 and 254 days in the study behind the famous 66-day average, which came from 39 of 96 volunteers. In a 2021 trial, only 23% of participants reached the threshold for a formed habit within 12 weeks.

Journaling statistics: 38 numbers, every one sourced

These are the numbers behind journaling, worry, sleep, meditation and habits, each one checked against the original study and linked to it. Every figure below appears in one of our longer articles, where you’ll find the context, the caveats and the method behind it. The short version: journaling and meditation help, modestly; most people stop within weeks; and the way you do them matters more than how long you do them.

Last checked October 2026. If you cite a number, please cite the study it comes from; a link back to this page is welcome too.

Does journaling help? The effect sizes

r = .075 is the average effect of writing or talking about personal experiences across 146 randomized studies, positive and statistically significant, and small (Frattaroli, 2006).
d = 0.47 was the average effect of expressive writing on health and well-being in an early meta-analysis of 13 studies of healthy people (Smyth, 1998).
g = −0.12 is the pooled effect of expressive writing on depression, anxiety and stress across 31 randomized trials and 4,012 people with long-term follow-ups. It tended to appear at later follow-ups rather than right after the writing (Guo, 2023).
About 5% more reduction in symptom scores than control conditions is what journaling produced for people with a diagnosed mental illness, across 20 randomized trials (Sohal et al., 2022).
76% vs 42% of small skin wounds were fully healed at day 11 in older adults who did expressive writing, against the comparison group (Koschwanez et al., 2013). Blogs often quote this as 76% vs 58%; the paper says 42%.
15 minutes, four days in a row was the original expressive writing exercise, in which students who wrote about upsetting experiences later made fewer health-center visits (Pennebaker and Beall, 1986).

Worry and anxiety

91.4% of the outcomes people with generalized anxiety disorder worried about did not come true, in entries from a worry outcome journal (LaFreniere and Newman, 2020).
89.14% of feared outcomes did not happen in a 2026 smartphone trial with 117 people with GAD, and when they did, people coped better than they expected 84.35% of the time (Schwob and Newman, 2026).
About 15 minutes less daily worry was reported by people who added specific if-then plans to worry postponement, in a 14-day trial with 186 adults (McCarrick et al., 2025).
About a third of U.S. adolescents and adults have an anxiety disorder at some point in their lives (National Institute of Mental Health).
585 people across seven studies who thought about a stressful task using their own name or “you” instead of “I” felt less distressed and performed better under social stress (Kross et al., 2014).

Naming emotions and tracking mood

30 people in the original affect-labeling brain-imaging study showed reduced amygdala response when they named the emotion in a face rather than matching it to a name (Lieberman et al., 2007).
74,487 people’s tweets showed negative emotion dropping sharply after they posted an explicit statement of how they felt (Fan et al., 2019).
More than 40,000 momentary reports from 301 people showed that, for people who couldn’t tell their negative feelings apart, most coping strategies were more strongly tied to increased negative emotion (Kalokerinos et al., 2019).
r = −.15 is the average link between telling negative emotions apart and fewer harmful behaviors such as binge drinking or self-injury, across 17 studies (Seah and Coifman, 2022).
25 studies of mood-monitoring apps for young people produced evidence the reviewers called “promising but limited” (Dubad et al., 2018).

Sleep and racing thoughts

About 16 vs 25 minutes to fall asleep: young adults who wrote a specific to-do list before bed fell asleep faster than those who wrote about tasks already done, measured by polysomnography in a sleep lab (Scullin et al., 2018).
18 randomized trials and 1,654 participants: mindfulness meditation improved sleep quality compared with time-matched controls, with low-strength evidence of no difference from established sleep treatments (Rusch et al., 2019).
33 students with insomnia who did “constructive worry” (worries plus solutions, early in the evening) reported less pre-sleep arousal over five nights (Carney and Waters, 2006).
15 to 20 minutes awake in bed is when a standard CBT-I stimulus-control handout says to get up and do something quiet elsewhere (University of Pennsylvania); the American Academy of Sleep Medicine recommends stimulus control for chronic insomnia.

Meditation and mental health apps

3.3% is the median 30-day retention across 93 popular mental health apps, and 4.7% for mindfulness and meditation apps (Baumel et al., 2019).
300 minutes planned, 16 done: new Medito users intended a median of 300 minutes of meditation in their first month and did 16; fewer than 20% were still meditating after 14 days (Adams et al., 2026).
58% of people who paid for a year of Calm had stopped using it by day 350 (Fowers et al., 2022).
g = 0.24 and 0.28 are the effects of mindfulness apps on depression and anxiety across 45 randomized trials, small and real (Linardon et al., 2024).
64% of assigned home practice is what people in eight-week mindfulness courses actually did, and more practice went with better outcomes (Parsons et al., 2017).
8.3% of meditators reported an adverse event, most often anxiety or depression (Farias et al., 2020).

AI chatbots and journaling privacy

93% vs at least 20%: in a Stanford test of responses to mental-health prompts, human therapists answered appropriately 93% of the time, while AI models answered inappropriately at least 20% of the time (Moore et al., 2025).
981 people in a four-week randomized study: those who used a chatbot more had worse loneliness, emotional dependence and problematic use, whatever kind of conversation they had (Fang et al., 2025).
0.15% of ChatGPT’s weekly users have conversations with explicit indicators of suicidal planning or intent, by OpenAI’s own estimate (OpenAI, October 2025).
$7.8 million is what the FTC ordered BetterHelp to pay over allegations that it shared health questionnaire answers with advertising platforms (FTC, 2023).

Habits

18 to 254 days is how long people took to reach near-peak automaticity on a new daily habit; the often-quoted 66-day average comes from 39 of 96 volunteers (Lally et al., 2010).
23% of participants in a 12-week randomized habit trial reached the threshold for a formed habit (Keller et al., 2021, as reviewed by Singh et al., 2024).
d = 0.65 is the effect of if-then plans on reaching goals, across 94 independent tests (Gollwitzer and Sheeran, 2006).
4.2 percentage points more employees got a flu shot when asked to write down the date and the time they would go, from a 33.1% baseline (Milkman et al., 2011).
About 106 vs 154 days for a stretching habit to become automatic when done on waking versus at bedtime (Fournier et al., 2017).

Paper, typing and voice

No difference in the effect of writing by hand, typing or talking, across the 146 studies in the largest meta-analysis of disclosure; writing nobody would read did marginally better, r = .178 vs .050 (Frattaroli, 2006).
153 vs 52 words per minute: speech recognition on an iPhone was 2.93 times faster than the on-screen keyboard for English (Ruan et al., 2018).
168 students who wrote about stress by hand or on a keyboard felt equally upset and wrote about the same things (Sharp and Hargrove, 2004).

How to cite this page

Auserene (2026). Journaling statistics: what the research shows. https://www.auserene.com/blog/journaling-statistics

Please cite the original study for any number you use. Every figure here links to it.

Journaling statistics: what the research shows