Search Engine Manipulation Effect (SEME)

• Bias: Search Engine Manipulation Effect (SEME) – a phenomenon where biased search results significantly influence users' opinions, preferences, and decisions, especially those who have not yet formed a stance (S001).
• What it breaks: Objectivity of opinion formation, democratic processes, consumer choice, the ability to critically evaluate information.
• Evidence level: L1 – multiple randomized controlled experiments in various countries (S001), over 1,000 citations of the foundational study, reproducible results.
• How to spot in 30 seconds: You form an opinion about a candidate, product, or idea based mainly on the first 2‑3 search results, without considering why those results appear at the top.

Why does the order of search results rewrite our beliefs?

The Search Engine Manipulation Effect is one of the most powerful yet subtle cognitive phenomena of the digital age. First systematically described and studied by Robert Epstein and colleagues in 2015, SEME shows that biased rankings in search results can shift the preferences of undecided voters by 20 % or more in certain demographic groups (S001). This is not merely a statistical error – it is a difference capable of determining election outcomes, shaping public opinion on critical issues, or radically influencing the consumer behavior of millions.

A seminal study published in the prestigious Proceedings of the National Academy of Sciences presented evidence from five experiments conducted in two countries, confirming both the strength and durability of the SEME (S001). Since publication, the paper has received over 1,000 citations, underscoring its critical importance for understanding how digital technologies shape human thought and behavior. Subsequent research has not only replicated the original findings but also expanded knowledge of the effect’s mechanisms, its applicability across domains, and potential mitigation strategies (S002, S003).

A particularly troubling aspect of SEME is its invisibility to users. People generally do not realize when search results are biased and fail to notice that their opinions are being shaped by the order in which information is presented (S004). This invisibility makes the effect especially dangerous: unlike overt advertising or propaganda, which can be recognized and critically evaluated, SEME operates at a level that appears neutral and objective to users.

Search engines are perceived as tools for finding information, not as editors deciding which information to show first.

The mechanism of SEME is based on order effects – cognitive biases where the sequence of information presentation influences judgments and decisions. In the context of search engines this manifests as a primacy effect: the tendency to give greater weight to information encountered first (S002). Users disproportionately trust results that appear higher on the list, assuming that position correlates with quality, relevance, or credibility. This assumption is often wrong, yet it is so deeply ingrained in our interaction with digital interfaces that it operates automatically, without conscious analysis.

SEME is closely linked to availability heuristic, where we overestimate information that comes to mind more easily, and the anchoring effect, where the first search results become a reference point for all subsequent judgments. Moreover, confirmation bias amplifies the effect: users tend to click on results that confirm their existing beliefs, creating a closed loop of manipulation. This combination of cognitive biases makes SEME especially resistant to critical analysis.

Cognitive Architecture of Manipulation: How Algorithms Rewrite Our Beliefs

The search engine manipulation effect operates through several interrelated psychological and cognitive mechanisms that exploit fundamental features of human perception and information processing. At the core of SEME is the primacy effect—a well‑documented cognitive bias whereby information presented first exerts a disproportionately large influence on judgment formation and decision‑making (S001). In the digital context of search engines this effect is amplified by specific interface characteristics and user‑behavior patterns (S002, S003).

The neuro‑psychological basis of the primacy effect relates to how the brain processes and stores information. Early items in a sequence receive more cognitive resources for processing, are encoded more robustly into long‑term memory, and establish a cognitive frame through which all subsequent information is interpreted. When a user views search results, the top positions automatically attract more attention, are examined longer, and are perceived as more relevant and trustworthy.

Illusion of Objectivity: When We Trust the Black Box

A critical aspect of the SEME mechanism is the implicit trust in algorithms. Most users unconsciously assume that search engines operate as neutral tools that objectively assess the relevance and quality of information. This assumption creates a trust heuristic: “if it’s in the top spot, it must be the best or most correct answer” (S001).

The complexity and opacity of algorithms generate an illusion of objectivity: because we do not see how ranking decisions are made, we assume they are based on objective quality criteria. This heuristic operates automatically and swiftly, conserving cognitive resources while rendering users vulnerable to manipulation (S004).

Cognitive Economy and Positive Reinforcement

SEME appears reasonable and natural for several reasons. First, our experience with search engines usually confirms that top results are indeed relevant—most of the time we find what we’re looking for within the first few links. This creates positive reinforcement that strengthens trust in the ranking.

Second, SEME exploits cognitive economy—our natural tendency to minimize mental effort. Scanning and evaluating a large set of search results consumes time and cognitive resources. It is much easier and faster to rely on the first results, especially when we are under time pressure or cognitive load. This strategy generally works well in everyday life, making it an attractive heuristic; however, that very effectiveness renders us vulnerable when search results are skewed intentionally or unintentionally.

Experimental Evidence: When Order Rewrites Reality

A seminal study by Epstein and Robertson (2015) comprised a series of randomized controlled experiments in which participants used a purpose‑built search engine to investigate information about election candidates. The researchers manipulated the order of search results, showing one group results favorable to candidate A and another group results favorable to candidate B (S001).

Crucially, the content of the results was identical for both groups—the only difference was the order of presentation. The findings were striking: biased search results shifted voter preferences by 20 % or more among participants who were initially undecided. In some demographic segments the effect reached 80 %.

Even more concerning, the overwhelming majority of participants were unaware that the search results had influenced them and did not notice the ranking bias. When asked about factors that shaped their decision, they cited the content of the information, personal values, and rational considerations, but not the order of the results. This illustrates the bias blind spot—people’s inability to recognize external influences on their judgments.

Factor	Impact on Effect Strength	Mechanism
Familiarity with the topic	Inverse (less knowledge → stronger effect)	Lack of cognitive resources for critical evaluation
Trust in technology	Direct (more trust → stronger effect)	Illusion of algorithmic objectivity
Cognitive load	Direct (higher load → stronger effect)	Switching to fast heuristics instead of analytical thinking
Time pressure	Direct (greater haste → stronger effect)	Reduced time for critical evaluation of information
Decision uncertainty	Direct (greater uncertainty → stronger effect)	Relying on external relevance cues instead of personal judgment

Subsequent research has confirmed and extended these findings. Experiments conducted in various countries and with different decision types (not limited to electoral choices) have demonstrated the robustness of the SEME effect across contexts. Studies have also identified moderators of effect strength: it is stronger among individuals less familiar with the topic, among those who place greater trust in technology, and in situations where users experience cognitive load or time pressure.

Feedback Loops: How AI Amplifies Human Biases

An important line of inquiry examines how SEME interacts with other cognitive biases and how it manifests in artificial‑intelligence systems. Research has shown that the primacy effect is present not only in human interaction with search engines but also within AI systems themselves, creating feedback loops: biased human behavior trains a biased AI, which in turn amplifies human biases (S003).

This finding is critical for understanding how SEME can intensify and propagate within the digital technology ecosystem. When people click on top search results (often due to SEME), those clicks signal to the algorithm that the results are truly relevant. The algorithm then ranks those results even higher, creating a self‑reinforcing cycle that can entrench even erroneous or biased information.

Research also documents SEME manifestations in the context of gender and demographic biases. Biased autocomplete suggestions have been identified against certain groups, illustrating how SEME can contribute to the maintenance and amplification of social stereotypes. These findings demonstrate that SEME is not merely an abstract phenomenon affecting elections or consumer choices, but a mechanism that can shape and cement systemic biases in society, interacting with confirmation bias and the anchoring effect.

Digital environment, decision-making, information behavior

Examples of SEME in real life: how search engines shape our decisions

Scenario 1: Mayoral election in a midsize city

Imagine a voter named Alex, who is preparing for the local mayoral election. He has heard the names of two main candidates — Johnson and Peters — but hasn't closely followed the campaign and has no clear preference. A week before the election Alex decides to “Google” both candidates to make an informed decision.

He types the query “Johnson mayor” and sees search results (S001, S004). The first three results are articles describing Johnson’s successful projects as a city‑council member, her educational initiatives and support for local business. The fourth and fifth results contain criticism of her stance on urban development. Alex clicks the first two results, skims them and forms a positive impression.

Then he types “Peters mayor” and sees a different picture: the top results are critical articles about his past business projects and controversial statements, while positive material appears lower on the page. The search algorithm was unintentionally (or intentionally) biased toward Johnson because her campaign used SEO more actively, or because the algorithm favors certain source types that more often publish material about her.

Alex spends about 15 minutes reviewing the information and concludes that Johnson is the more competent candidate. He believes his decision is based on an objective assessment of facts, unaware that the order of search results played a decisive role. If thousands of voters like Alex see biased results, the effect could determine the election outcome (S001). This is an illustration of blind spot bias — Alex does not see that his judgment was distorted.

Scenario 2: Choosing a medical treatment

Mary, a 45‑year‑old woman, recently learned she has been diagnosed with an early stage of a disease for which several treatment options exist. Her doctor mentioned two main approaches: traditional therapy and a new experimental method. Mary wants to make an informed decision and turns to a search engine with the query “treatment [disease name] reviews” (S007).

The top search results are articles and forums where patients share positive experiences with the experimental method. These materials are well SEO‑optimized, contain emotional recovery stories and are actively promoted by the company producing the new treatment. Information about traditional therapy, its efficacy and long‑term outcomes appears lower on the page, in less prominent positions.

Mary reads the first three‑four results, is impressed by the success stories and starts leaning toward the experimental method. Research shows most users never go beyond the first page of results, making top positions critically important for opinion formation (S001). Traditional therapy has a broader evidence base, more predictable outcomes and fewer side effects, but this information is less “viral” and less aggressively promoted online.

The experimental method, while showing promising results in short‑term studies, has limited long‑term data and may not be suitable for all patients. Mary does not realize that the order of results reflects SEO effectiveness and commercial interests rather than medical consensus or evidence quality. Her decision, which she believes is carefully weighed, is actually shaped by SEME, an example of confirmation bias — she sees only information that confirms the appeal of the experimental method (S007, S010).

Scenario 3: Forming an opinion on a social issue

David, a university student, participates in a debate on a contentious social issue — for example, the impact of social media on teenagers’ mental health. He has no established view on the topic and decides to explore it using a search engine. He enters the query “social media impact on teenagers” (S008, S010).

The top results are articles emphasizing negative aspects: studies linking social‑media use to depression, anxiety and sleep problems. These pieces are well‑structured, contain striking statistics and expert quotes. Articles offering a more nuanced picture — e.g., that impact depends on type of use, individual characteristics and the quality of offline relationships — appear on the second page, which David does not reach.

David reads the first five results, takes notes and forms the opinion that social media is predominantly harmful to teenagers. In the debate he confidently presents this position, backing it with “facts” he found online. He does not realize his view was shaped not by the completeness of evidence but by the order in which the search algorithm presented information.

If the algorithm systematically favors certain content types — for instance, more sensational or emotional material — it can shape public opinion on important social issues, creating an illusion of consensus where diverse viewpoints actually exist (S008, S002). This relates to the availability heuristic, where information that comes to mind more easily is perceived as more common and important.

Scenario 4: Consumer choice and commercial decisions

Helen plans to buy a new smartphone and wants to compare several models. She enters the query “best smartphone 2024” and sees results that appear to be objective reviews and rankings (S007). The first three results are articles in which a particular model (let’s call it “Model X”) consistently occupies the first or second spot.

These reviews contain detailed technical specifications, professional photographs and persuasive arguments in favor of Model X. Helen clicks the top results, reads the reviews and begins leaning toward purchasing Model X. She does not know that these “independent” reviews are actually part of affiliate programs where authors earn commissions on sales through their links, or that the manufacturer of Model X invested significant resources in SEO and content marketing.

Alternative models that might better match her needs or budget appear lower in the search results and go unnoticed. Helen makes the purchase, relying on information that seemed objective but was actually carefully selected and ranked in the manufacturer’s interest.

Several months after the purchase Helen discovers that Model X has battery problems that are widely discussed on specialized forums, but this information was not visible in the top search results at the time of her research. Her consumer decision, which she considered well‑thought‑out, was shaped by SEME — an effect that operates not only in political contexts but also in commerce, influencing purchase decisions worth billions of dollars each year (S007). This demonstrates the anchoring effect, where the first information and ratings seen become the reference point for all subsequent judgments.