Imagine you’re worried about vaccinating your baby. A message on Facebook tells you: “Protect the community and your kids by booking their vaccinations today.” Another asks: “Do you know all the effects of childhood vaccinations on your babies?”
Both link to the same reliable vaccination information. But in our recently published research, the second message was about 52% more likely to be clicked on.
We wanted to test whether changing the way that factual information is presented online could help it reach audiences who might otherwise overlook it. Our findings suggest the wording matters, even when the information itself stays the same.
What is misinformation?
Misinformation can appear in everyday discussions on social media, including conversations about childhood vaccination. In the Australian Facebook communities we studied, these discussions often connected with personal concerns about protecting children.
Misinformation means false information shared without an intention to cause harm. Disinformation is false information deliberately created and shared.
However, false claims are only part of the problem. Our earlier research found far-right users also shared articles from moderate news sources. They selected a narrow range of topics that fit their world views, meaning even mostly accurate news articles could reinforce those views.
Not only does this make it harder to get reliable information to people, it can also negatively affect their relationships. Our research describes people withdrawing from loved ones and mainstream information as they became more involved in fringe communities.
A misinformation minefield
Governments provide factual resources, while researchers and other organisations investigate misleading claims and provide the evidence. Fact-checking can help, but people need to trust the source and understand the explanation.
Social media platforms have also tried removing false or misleading content or making it less visible. Research on Meta’s pandemic response found many of the Facebook pages regulated by Meta and examined in the study declined in performance, although some continued growing.
Social media platforms have also tried adding warning labels to misleading posts and showing them to fewer people.
Users found ways around restrictions, such as coded language and links to less moderated platforms.
Meta stopped using independent fact-checkers in the United States in April 2025. It began introducing Community Notes, where users write and rate explanations that appear alongside posts.
So these approaches have had some impact, but they leave a question: how do you get reliable information in front of someone who isn’t looking for it?
Meet Jennifer, Patrick and Aaron
We spent months observing Australian Facebook communities sharing anti-vaccination and far-right narratives.
Alongside information, these communities offered companionship. Messages could sound like a friend sharing worries about their child.
We used these observations to develop three fictional audience profiles, known as personas.
Jennifer was a young mother concerned about vaccination. Patrick was a retired teacher worried about his grandchildren’s future. Aaron was a young man interested in fitness and technology. These profiles helped us select audiences for Facebook advertisements and write messages using familiar styles.
For Jennifer, we used personal, caring language, like a friend sharing worries about their children.
Related on the SLP
Patrick’s messages took an authoritative, explanatory tone. Aaron’s used a motivational tone, encouraging readers to take charge and check the facts for themselves.
We then compared those messages with neutral versions of the same ad. Each pair used the same image and linked to the same government or government-affiliated resource. The sponsor and advertising settings stayed the same, only the wording changed.
In 11 out of the 12 comparisons, people were more likely to click on an ad when its wording reflected the language used in the communities we studied.
Ads about school education received clicks 6% of the time when they used familiar language. The same ads with neutral wording only received clicks 1.1% of the time. This means the language inspired by communities the ads were trying to reach were roughly five times more likely to be clicked on.
For vaccination ads, those with familiar language were clicked on 0.98% of the time, compared to 0.65% for those with neutral wording.
Speaking their language
Our findings suggest communication style can help attract attention to reliable information.
They don’t show that we changed anyone’s mind. We don’t know whether people read the linked pages, and Facebook’s delivery system influenced who saw each advertisement.
We also changed several features of the wording together, so we can’t identify which feature made the difference.
But even with these caveats in mind, there are ways these findings can be used now to help combat mis- and disinformation online.
For example, public health teams could use this approach by listening to how people describe their concerns, then testing different ways of presenting the same evidence.
This doesn’t mean impersonating community members or amplifying fear to attract attention. It just means adopting language that speaks directly to questions and concerns of the people you’re trying to reach.
For a vaccination campaign, this could mean testing messages that acknowledge parents’ questions. For example, “You want to keep your little one safe. Wondering what vaccinations will mean for your baby?”.
A follow-up study could then examine whether readers understood the information and felt better able to make a decision.
Jooyoung Lee is a Research Fellow at the Behavioral Data Science Lab of the University of Technology Sydney. Find Jooyoung on LinkedIn
Emily Booth is a Research Fellow at the Data Science Institute of the University of Technology Sydney. Find Emily on LinkedIn
Marian-Andrei Rizoiu is a Professor of AI and Behavioral Data Science at the University of Technology Sydney. Find Marian-Andrei on X @andrei_rizoiu
A version of this article was originally posted at Conversation and has been reposted here with permission. Any reposting should credit the original author and provide links to both the GLP and the original article. Find Conversation on X @Conversation_US
