If you search for ECT, what you find will probably frighten you. Researchers set out to measure exactly that, analysing tweets in English and Spanish that mentioned ECT over a 22-month period.
Two findings stand out. First, more than half of all the posts were jokes or trivialising content — and those were liked and shared more than anything else. Second, among posts expressing a genuine personal opinion, around 75% were negative.
There was also an imbalance in who was talking. Healthcare providers produced about a quarter of the posts and most of the medical content. People with lived experience of ECT, and their families, made up only around 10% combined.
So the loudest voices online are largely not the people who have actually had the treatment.
This matters because that impression is doing real work. ECT is one of the most stigmatised treatments in medicine, and fear shaped by jokes rather than evidence can stop someone considering something that might genuinely help them.
None of this means you should ignore concerns about ECT — there are real trade-offs worth discussing. It means the discussion is better had with a psychiatrist than with a search engine.
What the study did
This was an observational study, meaning the researchers watched and measured a naturally occurring pattern (public tweets) rather than manipulating anything themselves. de Anta et al. collected every English- and Spanish-language tweet mentioning electroconvulsive therapy (ECT) between January 2019 and October 2020.
They then ran a mixed quantitative-qualitative content and sentiment analysis. In plain terms: 'content analysis' means sorting each tweet into meaningful categories (who posted it, and what kind of content it contained - medical, personal opinion, joke/trivializing, and so on), while 'sentiment analysis' means judging whether the emotional tone of a piece of text is positive, negative, or neutral. To do this at scale, the team combined manual coding (humans reading and classifying tweets) with semi-supervised natural language processing (NLP) - a machine-learning approach where an algorithm learns to classify text after being trained on a smaller set of human-labelled examples, then applies that pattern to a much larger dataset. Alongside this, they tracked 'public interest' using retweets and likes per tweet, as a rough proxy for how far each type of content actually spread.
What it found
Healthcare providers were the single most active group, producing about 25% of tweets, and they generated most of the medical content, largely focused on ECT's main clinical indications (the conditions it's used to treat). By contrast, people with lived experience of ECT and their relatives combined made up only around 10% of tweets.
More than half of all tweets were classified as joke or trivializing content - and this category had the highest like-and-retweet ratio, meaning it was disproportionately rewarded with engagement compared to medical or personal-experience content. Among the smaller subset of tweets that expressed a genuine personal opinion about ECT, roughly 75% carried a negative sentiment.
The authors note that this picture is more negative than what's typically found in studies using traditional methods, like surveys asking people directly about their attitudes toward ECT.
Why it matters
ECT remains one of the most stigmatized treatments in psychiatry, and this study suggests social media may be amplifying rather than correcting that stigma. Two mechanisms seem to be at play: an imbalance in who is speaking (clinicians and, implicitly, critics or casual commentators dominate over people who've actually undergone treatment), and an engagement bias, where platforms tend to reward provocative or trivializing content with more visibility regardless of accuracy. Traditional surveys can't capture this dynamic, which is why this social-media-based approach adds something genuinely new to the picture.
Limits to keep in mind
This study describes what's posted on one platform (Twitter/X), during one specific 22-month window, in two languages - it doesn't tell us what people who don't tweet believe, and it can't establish that social media content causes negative attitudes, only that it correlates with them. Sentiment classification, even when machine-assisted, involves judgment calls that don't capture nuance perfectly.
If you're weighing up ECT for yourself or someone you care about, a conversation with a psychiatrist will give you a far more reliable picture than a social media feed - and reaching out for that conversation is always a reasonable thing to do.