Social media platforms provide spaces for individuals to disclose mental health experiences and seek support, yet how these disclosures evolve during major shifts in platform governance remains unclear. We examine mental health self-disclosure surrounding Elon Musk's October 2022 acquisition of Twitter, analyzing approximately 824K tweets across anxiety, depression, bipolar disorder (BD), borderline personality disorder (BPD), and suicidal ideation (SI) or related experiences, alongside 266K comparison tweets. Using Linguistic Inquiry and Word Count (LIWC) and an LLM-based intimacy classifier, we characterize changes in psycholinguistic expression and disclosure intimacy. Following the acquisition, we observe declines in affective and social language, alongside increases in first-person singular pronouns and future-focused expressions. Intimacy patterns vary by condition: anxiety, depression, and BPD show increases in low-intimacy and decreases in moderate-intimacy disclosures, whereas suicide-related posts show more high-intimacy disclosures. However, compared with broader Twitter discourse, only the relative decline in moderate- versus low-intimacy disclosure for BPD remains statistically significant. Our findings underscore the importance of considering both language and intimacy when studying platform transitions, informing the design of online environments that support sensitive self-disclosure.
Figures & tables
Figure 1: Overview of the study design.
Dataset
N.Users
N.Twitter posts
Twitter posts/User Mean (SD)
Depression
142,356
175,503
1.23 (7.31)
Anxiety
304,393
350,101
1.15 (2.59)
BD
26,794
39,194
1.46 (1.96)
BPD
34,740
52,171
1.50 (4.37)
SI-related
167,558
206,857
1.23 (4.67)
Control
248,374
266,364
1.07 (0.88)
Table 1: Descriptive statistics of the Twitter mental health dataset.
Condition
Pre N
Post N
Δ Low
Δ Moderate
Δ High
Mean Pre
Mean Post
χ2
Sig.
Anxiety
334,211
15,719
+3.36
-3.13
-0.22
1.681
1.645
82.41
BD
36,797
2,395
+1.67
-1.22
-0.46
1.826
1.805
4.68
BPD
46,313
5,844
+13.55
-13.38
-0.17
1.723
1.586
440.63
Depression
168,209
7,247
+5.14
-4.94
-0.20
1.611
1.558
75.26
SI-related
188,672
17,877
-0.77
-1.26
+2.03
2.288
2.316
55.41
Control
252,000
14,000
+0.05
+0.00
-0.05
1.014
1.013
1.90
Table 2: Changes in disclosure intimacy before and after the acquisition. Significance is based on FDR-adjusted p -values (* p<.05 , ** p<.01 , *** p<.001 ).
Condition
ΔΔ Low
ΔΔ Mod.
ΔΔ High
Mod. vs. Low RRR
Sig.
Anxiety
+3.31
-3.14
-0.17
0.871
BD
+1.62
-1.22
-0.40
0.915
BPD
+13.50
-13.38
-0.12
0.553
Depression
+5.09
-4.94
-0.15
0.813
SI-related
-0.82
-1.27
+2.09
0.821
Table 3: Changes in disclosure intimacy relative to control ( pFDR<.05 , pFDR<.01 , pFDR<.001 ).
Appendix figures & tables3 assets
Supplementary material from the paper’s appendix.
Appendix
Depression
BD
Anxiety
“I have depression” “I have been diagnosed with depression” “I was diagnosed with depression” “I suffer from depression” “I am suffering from depression”
“I have bipolar” “I have been diagnosed with bipolar disorder” “I was diagnosed with bipolar disorder” “I suffer from bipolar” “I am suffering from bipolar”
“I have anxiety” “I have been diagnosed with anxiety” “I was diagnosed with anxiety” “I suffer from anxiety” “I am suffering from anxiety”
BPD
SI-related
Control
“I have borderline personality disorder” “I have been diagnosed with borderline personality disorder” “I was diagnosed with borderline personality disorder” “I suffer from borderline personality disorder” “I am suffering from borderline personality disorder” “I have bpd” “I have been diagnosed with bpd” “I was diagnosed with bpd” “I suffer from bpd” “I am suffering from bpd”
“I have suicidal ideation” “I want to kill myself” “I wanted to kill myself” “I want to suicide” “I wanted to suicide” “I am going to kill myself” “I was going to kill myself” “I tried to kill myself” “I try to kill myself” “I tried to suicide” “I try to suicide” “I want to commit suicide” “I wanted to commit suicide” “I am going to commit suicide” “I was going to commit suicide”
Twitter API collection of 1,000 tweets per week. Observation period: 1 January 2018–1 February 2023.
Appendix
Table A1: Search phrases and collection method for the five mental health disclosure samples and the control sample. The six samples are arranged in two panels for readability. All disclosure searches covered English-language tweets posted between 1 January 2018 and 1 February 2023.
LIWC
Treated (Tr.)
Control (Ct.)
ATE
Tr. vs. Ct.
Pre
Post
Δ%
Pre
Post
Δ%
%
Pre t
Post t
Affective Attributes
Affect
0.0669
0.0557
-16.65
∗∗∗
0.0537
0.0525
-2.17
-15.17
66.196
∗∗∗
3.828
∗∗
Pos. Emotion
0.0192
0.0173
-10.28
∗∗∗
0.0400
0.0388
-2.86
-4.62
−123.821
∗∗∗
−30.190
∗∗∗
Neg. Emotion
0.0001
0.0001
-5.61
0.0009
0.0006
-34.87
∗∗
n/a †
−34.313
∗∗∗
−6.593
∗∗∗
Anger
0.0038
0.0037
-2.93
0.0039
0.0041
+5.26
-7.93
−2.130
−1.780
Appendix
Table A2 : Pre- vs. post-acquisition LIWC comparisons for treated (mental health disclosure) and control posts. Δ% : within-group change. ATE %=β/μcf×100 , where β=ΔTr−ΔCt and μcf=μTr,pre+ΔCt ( † undefined, μcf<0 ). Pre/Post t : between-group tests. Bonferroni-adjusted ∗p<.05 , ∗∗p<.01 , ∗∗∗p<.001 .
Level
Definition and Criteria
Examples
Low (1)
Little or no substantive disclosure of the author’s lived mental health experience. This includes bare diagnoses, lists of diagnoses, background or incidental mentions of a condition, general discussion of mental health, third-party information, ordinary situational worry, casual conversation, and figurative or humorous uses of mental health language. A mental health keyword alone is not sufficient for Moderate or High intimacy.
“I have depression.” → Low “I have anxiety and ADHD.” → Low “I think I might have BPD.” → Low “I have depression and I love playing video games.” → Low “That actor is so attractive I wanna die.” → Low
Moderate (2)
A personal disclosure describing how a mental health condition affects everyday life. This includes effects on work or school, relationships, sleep or eating, coping strategies, medication use, disclosure to family or friends, social interactions, and other functional or behavioral consequences. The post must describe a concrete lived experience rather than only naming the condition.
“My anxiety makes it difficult to concentrate at work.” → Moderate “Because of my depression, I sometimes struggle to get out of bed for class.” → Moderate “Watching streams helps me cope with my depression.” → Moderate “I take medication for my anxiety every morning.” → Moderate “My BPD sometimes makes me jealous in relationships.” → Moderate
High (3)
Disclosure of highly sensitive lived mental health experiences. This includes suicidal ideation or suicide attempts, self-harm, mania, psychosis or hallucinations, dissociation, other sensitive symptoms, personal experiences of stigma, and seeking emotional or social support around a mental health condition. High-intimacy information takes precedence when a post also contains Moderate-level information.
“I want to kill myself.” → High “I attempted suicide when I was younger.” → High “I’m hearing voices again and I’m scared.” → High “I’ve been experiencing manic episodes.” → High “I keep dissociating and losing track of what’s happening.” → High
Decision rule. Posts were interpreted in their full context. We first distinguished personal experience from jokes, quotations, general discussion, or third-party information. Highly sensitive personal disclosures were assigned High; otherwise, concrete everyday experiences related to functioning, relationships, coping, medication, or condition management were assigned Moderate. Posts containing only diagnostic labels, incidental mentions, casual discussion, figurative language, or general opinions were assigned Low.
Appendix
Table A3: Codebook for disclosure intimacy classification.
Aug 29, 2026·Renkai Ma, Lingyao Li, Shanting Chen +3Mental Health
College of Information Science University of Arizona Tucson, Arizona, USA · Department of Computer and Information Science and Engineering University of Florida Gainesville, Florida, USA