Groundbreaking research published in the Proceedings of the National Academy of Sciences (PNAS) has uncovered compelling evidence that ordinary people unconsciously contribute to the growth of their own information bubbles, simply through the way they type search queries online. This large-scale study, encompassing 21 experiments and nearly 10,000 participants, demonstrates that even without any intent to seek confirmation, our habitual online search patterns—and the algorithms designed to respond to them—subtly guide us towards ever-narrower realities. These findings have huge implications for how Thais access information, understand national debates, and engage with global topics in a time when digital literacy is crucial for an informed society (PsyPost).
In a digital world where nearly 90% of Thais aged 15 and above access the internet daily, knowing how our searches shape our worldviews matters deeply. Social media and search engines are often accused of creating isolated echo chambers for users. However, this research shows the problem goes beyond technology alone: we, through our own search behavior, help seal ourselves in informational silos. Understanding this dynamic can benefit everyone from high school students preparing for university entrance exams to public health officials battling misinformation, to everyday social media users seeking trustworthy sources.
The study was inspired by a simple, relatable situation. As recounted by one of the lead researchers, an assistant professor at Tulane University, the authors began asking questions about how typing different search terms yielded sharply different results during a routine Google search for cold medicine: “Searching for ‘cold medicine side effects’ gave me a much more alarming set of results than searching for ‘best medicine for cold symptoms’,” she explained (PsyPost). This experience highlighted the immense power of phrasing: which words we type shape the information we see.
To test just how far this effect reached, the scientists designed experiments on a wide range of topics—from health concerns like caffeine benefits and food risks, to social issues like crime, energy, and finance. Participants reported their beliefs about each topic, then chose their own search terms. Reviewers who were unaware of the study’s aims categorized these queries. Consistently, participants chose search words that confirmed their existing beliefs. If a person already saw caffeine as healthy, phrases like “benefits of caffeine” were chosen; those who thought caffeine was dangerous used terms like “caffeine dangers.”
But the effect didn’t stop there: after reading the search results generated by their queries, participants’ beliefs shifted further in the direction of their initial bias—even if their original stance was only slight. Notably, those prompted to search “nuclear energy is good” ended up with a more positive view after the search, while those searching “nuclear energy is bad” came away more negative, despite being randomly assigned. A related experiment using identical search results for both groups showed there was no belief change unless the search content actually differed. This clearly linked the algorithm’s output—and our own choice of input—with reinforcement of bias.
The real-world implications were tested in another experiment involving Dutch university students. Those who searched for the benefits (instead of risks) of caffeine, not only reported more positive attitudes but also opted for caffeinated energy drinks over decaf—demonstrating how search-induced belief changes can translate to behavior.
What can be done? Encouraging users to conduct additional searches did not shake their biases; most simply repeated their initial search logic, still picking search terms that confirmed their beliefs. Behavioral “nudges,” like asking participants to imagine how different search words might have altered their results, had only modest impact.
The breakthrough came when the researchers intervened in the algorithms themselves. Using a custom-built search engine, they presented users with either results tailored for the specific, belief-confirming query, or with a blended set spanning both positive and negative viewpoints. Participants shown these broader, more balanced results were more willing to reconsider their beliefs. The same result was replicated using AI chatbots: answers that included pros and cons regularly encouraged more nuanced beliefs, and were rated just as relevant and useful as narrower responses.
As the lead researcher summarized, “Our existing beliefs unconsciously influence the words we type into a search bar, and because search engines are designed for relevance, they show us results that confirm our initial belief… While telling people to do follow-up searches doesn’t fix it, our research shows that designing search algorithms to provide broader, more balanced viewpoints is a very effective solution.” She suggested something as simple as a “Search Broadly” button on Google—mirroring the “I’m Feeling Lucky” option—could be a practical step toward a society better equipped to evaluate issues (PsyPost).
The researchers stress that these effects aren’t limited to political debates—where making space for nuanced discussion is already challenging. They span nearly all modern topics: from health to finance, energy, and even everyday issues on which many Thai internet users regularly seek guidance.
There are, of course, limits. The “narrow search effect” is most pronounced when users begin with some prior belief, and when technology returns different answers for different queries. In cases where everyone uses the same neutral query (such as a breaking news event), or in highly resistant identity-based topics, search bias may have less impact.
For Thai society, the timing of these findings is highly relevant. Thailand’s rapid internet expansion has meant digital tools play an increasing role in personal health decisions, political awareness, and consumer culture. The country’s digital literacy rates continue to lag behind regional neighbors (UNESCO report), and the Ministry of Digital Economy and Society has repeatedly warned about the risks of “fake news” and misinformation online. These new lessons suggest that bias originates at the very moment a user types into Google or an AI chatbot—the digital front door for most Thais seeking authoritative health advice, making travel plans, or searching for education resources.
Educational leaders at Thailand’s top universities have long promoted critical thinking classes to help students recognize and avoid filter bubbles. But as this research points out, even a highly educated user will likely fall into the trap of their own search habits. Practical solutions now include appealing directly to the design of Thailand’s most-used platforms. For example, Thai-language search providers and local AI companies might adopt a broader-response protocol for controversial or nuanced issues, ensuring that users always see the spectrum of evidence, not just what matches their expectations (PsyPost).
Culturally, Thai society places a high value on avoiding conflict and maintaining harmony—traits that sometimes discourage heated debate but may also reinforce groupthink online. In classrooms, the preference for rote learning over open-ended exploration can interact with the “narrow search effect,” reinforcing rather than challenging preconceptions. It is not uncommon for major health scares to spread rapidly on social media as students, parents, and the elderly seek confirmation using search terms that echo their fears.
Thailand is not alone in this challenge, but the scale of mobile internet access—over 90% of the population according to the National Statistical Office—means every citizen is both a potential victim and a contributor to the persistence of echo chambers. In 2024, a government-backed campaign urged Thais to learn “fact-checking” skills for the new digital age. While information campaigns are important, this research underscores how systemic changes—such as algorithmic design—may offer a more potent weapon against belief polarization.
Technological companies and digital policymakers in Thailand could take a cue from these findings, pressing for regulations or guidelines that require search engines and AI chatbots to default to broader, more even-handed answers, especially in areas like public health and civic issues. Civil society organizations, media literacy non-profits, and education leaders should all consider how to support Thais in developing awareness not just of external manipulation, but of how our own choices—down to which Thai words we type—lock us into a particular informational world.
As AI becomes more woven into everyday life and work—for example, in student essay writing, health self-diagnosis, and even news consumption—understanding and counteracting the “narrow search effect” becomes more essential. The researchers themselves are now studying ways to tailor digital interventions for the most vulnerable populations and to test these strategies on divisive, misinformation-prone topics. Their goal: to create a new, human-centered standard for search and artificial intelligence design, and to foster a more robust public discourse.
Practical takeaways for Thai readers are clear. First, reflect intentionally on your own online search behavior. Before typing a query, ask: “Am I confirming my own beliefs, or am I searching for comprehensive truth?” Second, when seeking important information (on health, finances, education, or politics), deliberately rephrase queries to cover possible alternatives (“risks of X” instead of only “benefits of X,” and vice versa). Third, pay close attention to news sources that provide balanced reporting, and support policies or platforms that incorporate features for broader search and exposure. For parents and teachers, encourage students to try viewing an issue from multiple angles—and to understand that even unconscious choices can have lasting effects on what they learn.
With Thailand’s rapid push toward “Thailand 4.0” and the rise of AI-powered education and service platforms, now is the time to build digital habits and systems that help all citizens—regardless of age, background, or region—become more open-minded, better informed, and resilient in the face of polarization. As the study shows, small changes in search habits or algorithmic design can open the door to a richer, truer understanding of the world.
For further details and original research, see the PsyPost report on the PNAS study.