Summary: The disclosure research began with people talking into tape recorders, and it has never stopped testing speech. Across the randomised trials that compared the two, talking privately and writing come out about even, both beat thinking silently, and the effects are small either way. Voice also has a measured speed advantage. It does not have evidence that people finish more often.
What the evidence actually is
The big meta-analyses are mostly writing studies. Smyth's 1998 meta-analysis pooled 13 studies of written emotional expression. Frattaroli's 2006 meta-analysis pooled 146 randomised studies of disclosure and reported r = .075, roughly d = 0.15. That is the best estimate of the effect size, and it is small.
But the field did not start with writing, and it never stopped testing speech. In Pennebaker's 1987 experiments, people talked about their traumas into a tape recorder, alone or to a listener hidden behind a curtain. The listener made them hold back. Researchers later moved mostly to writing, which Slavin-Spenny and colleagues (2011) describe as a choice for experimental control, to avoid the awkwardness of talking to a tape recorder, not the result of writing winning a comparison.
These randomised trials had people talk privately, usually into a recorder:
| Study | Who and how | What it found |
|---|---|---|
| Esterling et al. 1994 | 57 students; talk vs write about stress vs write about trivia; three 20-minute sessions | Talkers had the best immune control of a dormant virus, then writers, then controls |
| Murray and Segal 1994 | Vocal vs written, 20 minutes, 4 days | "Similar emotional processing" in both; more overt emotion when talking |
| Kelley et al. 1997 | 72 people with rheumatoid arthritis; talked on 4 consecutive days | Worse mood at first; at 3 months less distress and better daily functioning; no effect on pain or joints |
| Lyubomirsky et al. 2006 | 96 students; write vs talk into a tape recorder vs think silently; 15 minutes on 3 days | Writing and talking both beat thinking at 4 weeks |
| Keefe et al. 2008 | 98 people with rheumatoid arthritis; private or clinician-assisted talking | No clear benefit. A null |
| Lumley et al. 2011 | Parallel writing (88) and speaking-into-a-recorder (93) trials; four 20-minute home sessions | Spoken: fewer swollen joints and lower doctor-rated disease activity at 6 months. Authors: modest benefits, "limited in scope and consistency" |
| Slavin-Spenny et al. 2011 | 214 people; one 30-minute session; written, private spoken, or talking to a listener | All methods produced more post-traumatic growth than control, and did not differ. No symptom reduction |
Put together: talking alone and writing perform about the same, which is modestly. One small trial favoured talking. None favoured writing. Several spoken trials, like several written ones, found nothing. And most of the spoken trials come from one research network studying arthritis patients, so how far they generalise to healthy people with a phone is an open question.
The comparison that matters most for a voice app is the third arm in Lyubomirsky's study. People who only thought about their worst experience did worse than people who said it out loud or wrote it down. Speaking is not the same as thinking it over.
The speed difference is measured
Ruan and colleagues (2018) tested speech recognition against the built-in touchscreen keyboard on an iPhone 6 Plus, under laboratory conditions favourable to both:
| Language | Speech | Keyboard | Ratio |
|---|---|---|---|
| English | 153 wpm | 52 wpm | 2.93x |
| Mandarin | 123 wpm | 43 wpm | 2.87x |
Speech corrected fewer errors during entry (5.30% vs 11.22%) but left slightly more in the final text (1.30% vs 0.79%).
This is a text-entry result, not a journaling result. It says nothing about whether the entries help you. It says that in a fixed five minutes you can produce roughly three times as much material by speaking.
The adherence argument does not hold up
This page used to argue that voice's real advantage is that more people finish. The evidence does not support that.
The worry is real: Baikie and colleagues (2006) ran the written protocol in a primary care clinic, recruited 53 people, and 14 completed all measures. But the one trial that ran written and spoken versions side by side, Lumley et al. (2011), found completion high in both. Nobody has shown that voice raises completion, so we no longer claim it.
What voice does have: a measured speed advantage, the hands-free situations below, and a format that the trials themselves used.
What voice plausibly changes
Emotional prosody
When you speak, you carry emotion through channels text does not have: pitch, rhythm, pace, volume, pauses, voice quality. Scherer, Banse and Wallbott (2001) found that emotion inferences drawn from vocal expression correlate across languages and cultures, which is evidence that the signal is in the voice itself rather than in the words.
Whether that signal does anything for a journaler reviewing their own recordings has not been studied. The content analyses are the closest thing: Murray and Segal's speakers expressed more overt emotion than writers, and Esterling's talkers showed the largest gains in cognitive change, self-esteem and coping.
Less filtering
Speaking is faster than your internal editor. When you write, there is a pause between thought and word where you edit, rephrase and second-guess. Speaking narrows that gap.
The Pennebaker protocol instructs people to keep going without worrying about grammar or structure. Voice makes that instruction easier to obey. This is a mechanism story, not a result.
Accessibility
Voice journaling works when writing doesn't:
- While walking or commuting
- When your hands are occupied
- With motor impairments that make writing painful
- In the dark (before sleep)
- When executive function is too low to open a notebook (relevant for ADHD)
What writing captures that voice doesn't
Deliberate processing
Writing is slower. That's usually framed as a disadvantage, but for certain kinds of journaling it's the point.
When you write, the speed constraint forces you to choose words. That choosing is cognitive work: you evaluate, categorise and structure as you go. Whether the slowness itself produces any of the benefit is an open question. No study has isolated it.
One caution against reading too much into linguistic markers. In the Baikie pilot, more positive-emotion words were associated with improvement in depression and stress, but increasing use of cognitive-process words was associated with worsening depressive mood. The markers are not a scoreboard.
Visual structure
Writing creates a spatial artifact. You can see your thoughts arranged on a page, draw connections, create lists and tables, mark important passages. Methods like fear-setting or bullet journaling rely on visual layout that pure voice can't replicate.
Privacy
You can write silently. You can't speak silently. In shared living spaces, offices, or public transit, writing offers privacy that voice doesn't.
Permanence of format
A written journal is immediately readable. A voice recording requires playback (or transcription) to review. The friction of reviewing voice entries is higher than scanning written pages.
What Pennebaker says
James Pennebaker, whose expressive writing protocol is the most-studied journaling intervention in psychology, used tape recorders in his own early experiments. His 1999 review is titled "the values of writing and talking about upsetting events", and his later handbook chapter with Cindy Chung (2012) summarises the head-to-head studies as finding "comparable biological, mood, and cognitive effects".
His text analysis tool, LIWC, runs on any text, transcripts included, so voice entries can be measured with the same instrument the written studies used.
The honest comparison
Speed and error rates below are from the Ruan text-entry study. The rest is a description of the two media, not a summary of findings, because the findings do not exist.
| Dimension | Voice | Writing |
|---|---|---|
| Entry rate | 153 wpm (English, measured) | 52 wpm on a phone keyboard (measured) |
| Uncorrected errors | 1.30% (measured) | 0.79% (measured) |
| Evidence base | Eight or more randomised trials of private talking; head to head, about even with writing | 146 randomised disclosure studies pooled by Frattaroli, mostly writing |
| Emotional signal | Prosody, tone, pacing | Deliberate word choice |
| Friction | Near-zero setup | Requires pen or device |
| Review | Needs transcription | Immediately scannable |
| Privacy | Audible to others | Silent |
| Accessibility | Hands-free, eyes-free | Requires motor control |
| Visual structure | None, audio is linear | Spatial layouts, lists, tables |
In the direct comparisons, neither came out ahead. The right choice depends on:
- What you're journaling for: brain dumps favor voice; analytical reflection favors writing
- When you're journaling: commuting, before bed, during walks → voice; at a desk → either
- Your constraints: shared space → writing; motor impairment → voice; ADHD → often voice
The hybrid approach
Most experienced journalers end up using both:
- Voice for capture: raw brain dump, emotional processing, on-the-go thoughts
- Transcript for review: read the transcription, highlight key insights, add structure
- Writing for synthesis: use the transcript as raw material for structured entries, lists, or action items
This gives you voice's speed and emotional fidelity at the capture stage, plus writing's structure and deliberateness at the review stage. The transcript is the bridge.
What's missing from the research
- A meta-analysis of spoken versus written disclosure. Not found. Whether Frattaroli (2006) coded modality as a moderator is something we have not been able to check in the full text.
- A trial of phone-based or automatically transcribed voice journaling. Not found. The spoken trials used tape recorders, mostly before 2012, mostly in arthritis patients.
- Any test of whether speaking improves completion. Not found, and the one side-by-side trial saw high completion in both formats.
- Whether reading back your own transcript helps, hurts or does nothing. Untested.
What is left is this. Disclosure has a small pooled effect. Talking privately into a recorder and writing perform about the same in the trials that compared them, and both beat silent rumination.
Use the format you will actually do. That is not a hedge; it is what the evidence supports.
Related: