Summary: Not really: research indicates it may be possible to detect PTSD risk by analyzing social media posts, but detection is limited to risk of PTSD, not identification of clinical PTSD.
Key Points:
- Researchers collected millions of social media posts for analysis
- Researchers trained a large language model, or LLM, what we now call AI, to classify posts as PTSD positive or PTSD negative
- The LLM identified presence of PTSD-like symptoms with over 80% accuracy
Identifying Early PTSD Symptoms
In 2025, a new tool called machine learning – a subtype of artificial intelligence (AI) that includes large language models (LLMs) designed use existing text to create human-like responses to inquiries or prompts – is revolutionizing the way we live and work, and a group of researchers based at the University of Birmingham in the U.K. recently conducted a study on the feasibility of using machine learning to detect post-traumatic stress disorder (PTSD) on social media.
In the paper “Identifying COVID-19 Survivors Living With Post-Traumatic Stress Disorder Through Machine Learning on Twitter,” data and health scientists published the results of a study that analyzed millions of posts on the social media platform Twitter, now called X, during the COVID-19 pandemic.
Here’s how the researchers describe the work:
“The goal of this study is to use classical machine learning approaches to classify tweets into COVID-PTSD positive or negative categories.”
In the wake of the pandemic, researchers knew the risk of mental health disorders associated with the pandemic could increase. We all knew that, because everyone warned us that the possible negative consequences of the pandemic itself, combined with the public health measures initiated to stop the spread and mitigate the impact of the disease, could have a detrimental effect on mental health. When we look at the pandemic from one perspective, we can see it as a collective trauma we all experienced together.
The research team realized this and saw a chance to leverage machine learning to serve public health by creating and teaching a machine model to classify tweets as PTSD positive or PTSD negative in order to identify people at risk of developing COVID-related post-traumatic stress disorder (PTSD).
PTSD, Social Media, and COVID
During the pandemic – the arrival, the pre-vaccine phase, and the long denouement that followed the arrival vaccines – it’s fair to say that engagement with social media increased for everyone, everywhere. During the early phase, it was one of the only ways we had to stay in touch with one another and the events of the world, and during the phases that followed, many of us kept up that level of engagement, sharing posts about COVID, mental health, and anything that we wanted to discuss.
To conduct the study, the research team gained access to a database of 3.96 million Tweets posted between March 2020 and November 2021 by people indicating they were COVID positive. Next, they filtered those posts for key words or phrases related to the following categories of PTSD symptoms:
- Re-experiencing, i.e. language related to flashbacks
- Hyperarousal, i.e. language related to agitation, irritability, excess worry, startle-prone
- Avoidance, i.e. language related to taking extreme steps not to encounter reminders of trauma
- Psychological symptoms, i.e. anxiety, depression, suicidality
Finally, they taught the machine learning model to place the posts into one of two categories:
- PTSD positive: posts by people diagnosed with COVID that included the term ‘COVID’ and language associated with PTSD-related symptoms
- PTSD negative: posts by people diagnosed with COVID that had no language associated with PTSD-related symptoms
How well did the machine learning model perform?
It classified the tweets with 83% accuracy.
We’ll discuss the implications of this finding below.
How This Can Help: Social Media Posts and Mental Health
We acknowledge that before deploying a machine learning model to assess mental health conditions in the general population using their social media posts, there’s likely a true tangle of privacy issues to address and overcome, which are far beyond the scope of this brief article.
We’ll assume – for the sake of discussion – there will be a way to use machine learning to analyze social media posts to detect PTSD or other mental health issues without compromising privacy.
Also, we should clarify that a tool like this would assess presence of increased risk. This would not be a clinical diagnostic tool – far from it. It would be a tool to identify people at risk for the purpose of reaching out to them to offer further evaluation and support, if needed.
Under that assumption, a tool like this could be valuable for several reasons.
Four Benefits of Early PTSD Detection Using Social Media
- People unaware of the symptoms of mental health disorders may not know that’s what they’re experiencing – and a tool like this could inform them.
- People with full awareness of what mental health symptoms are may have a hard time realizing that’s what they’re experiencing – and a tool like this could likewise inform them.
- Among people who know what they’re going through, many may not understand how to seek support – and a tool like this could identify them as potentially needing support.
- This tool could help identify people in need of support – by the content of their posts – before their symptoms escalate to the point of creating disruption in relationships, at school, or in the workplace.
That last point is perhaps the most salient: mental health disorders, including PTSD, can cause severe emotional and psychological distress, prevent people from performing well at school and work, and can reduce overall quality of life.
If a tool like this could help people get help before symptoms become severe and disruptive – without compromising privacy – then it would be an excellent addition to our current mental health monitoring techniques, because the earlier a person who needs treatment and support gets the treatment and support they need, the better the outcome.
Gianna Melendez
Jodie Dahl, CpHT