Can You Really Trust Those High Ratings? A New Training Method to Detect Fake Reviews

Can You Really Trust Those High Ratings? A New Training Method to Detect Fake Reviews

Can You See Through the Five Stars? The "Consumer's Decoding Skills" Needed in the Era of Fake Reviews

When choosing a product on online shopping platforms, what do we look at? Price, photos, delivery date, brand name. And often, the final push comes from the reviews. If there are many stars and positive comments, we feel "it's probably safe to buy." Conversely, if low ratings stand out, no matter how attractive the product photos are, we hesitate to purchase.

But can we really trust those stars?

Online reviews have become the "trust infrastructure" in modern consumer behavior. The experiences of strangers influence our shopping decisions. While there is the power of collective intelligence, the problem of fake reviews intentionally manipulating purchasing decisions is also spreading.

Michelle Walther, who conducted doctoral research at the University of Twente in the Netherlands, approached this problem from an intriguing angle. While research on automatically detecting fake reviews using AI is increasing, Walther focused on the judgment process of the "consumers themselves" reading the reviews. How do people read reviews, what clues do they trust, and at what stage do they decide "this review is not useful"? She clarified this mechanism and examined whether training could enhance the ability to detect fake reviews.


Consumers Don't Read Reviews "To Find Fakes"

A key point of this research is that when consumers read online reviews, they are not necessarily trying to "spot fake reviews." For many, the purpose of reading reviews is more practical.

Does this product suit me? How is the sizing? Are there any drawbacks not mentioned in the description? Are users satisfied?

In other words, consumers are primarily looking for "useful product information." Determining whether a review is fake tends to be a secondary task that arises during the decision-making process, not the main goal of shopping.

This is where the nuisance of fake reviews lies. Consumers, not in a cautious mode, encounter reviews written to seem like natural experiences. If the star rating is high, the text seems plausible, and the poster's name and profile picture are well-crafted, readers may accept it without deep suspicion. Fake reviews infiltrate precisely this "normal reading mode."


What is the CREM Model?

Walther organized the process by which consumers evaluate reviews by combining literature reviews, observations, and think-aloud methods. This led to the creation of the "Consumer Review Evaluation Model," or CREM model.

The CREM model shows that consumers do not trust reviews all at once but make judgments in stages. Broadly speaking, consumers first look at the relevance of the review. They check if it relates to the product information they want to know, if it suits the purpose they are considering for purchase, and if specific usage scenarios are described.

Next, they evaluate the trustworthiness of the poster. Clues include whether the person seems to have actually used the product, if there is anything unnatural about their posting history or profile, and if there is a mismatch between the review content and the rating.

Finally, they assess the credibility of the review text itself. Is the content specific, not overly exaggerated, free from repetitive expressions, and does the emotional expression naturally match the star rating? Through these multiple elements, consumers decide whether to include the review in their purchasing decision.

Interestingly, this process is not necessarily a rigorous examination like that of an expert. Many consumers rely on intuition and experience to read reviews within limited time. Therefore, detecting fake reviews depends not only on "knowledge" but also on "reading habits."


Training Can Enhance the Ability to Detect Fake Reviews

Another important point highlighted by Walther's research is that consumers' ability to detect fake reviews can be improved through training.

Training based on the CREM model helps consumers organize their perspectives when viewing reviews and learn what clues to pay attention to. For example, instead of just looking at the star rating, they check the specificity of the text, the trustworthiness of the poster, and the consistency between the review content and the rating. Extreme positive or negative expressions, abstract praises that do not touch on product features, and unnatural phrasing common to multiple reviews can also be warning signs.

The study found that participants who underwent such training significantly improved their ability to identify fake reviews. This suggests that fake review countermeasures can be advanced not only by relying on platforms but also as consumer education.

Of course, consumers should not bear all the responsibility. Businesses that post fake reviews, companies that buy and sell reviews, and platforms with insufficient countermeasures each have clear responsibilities. However, in reality, it will take time to completely eliminate fake reviews. In the meantime, consumers equipping themselves with the ability to "read reviews" becomes one of the defense strategies.


Why AI Detection Alone Is Not Enough

In fake review countermeasures, automatic detection by AI is also an important theme. Related research reports models that determine fraudulent reviews with high accuracy by combining linguistic features of the review text, posting behavior, consistency with star ratings, and the length of the text.

However, AI detection has its limitations. Those writing fake reviews also evolve. With generative AI, they can create more natural texts in large quantities than before. If you only look at simple keywords, unnatural language, or similar writing styles, you might miss cleverly crafted fake reviews.

Moreover, the authenticity of a review is not determined solely by the text. Even without actually buying the product, one can write a seemingly genuine experience. Conversely, real reviews can also be short and abstract. A review flagged as suspicious by AI is not necessarily fake, and a review passed by AI is not necessarily genuine.

This is why it is necessary to combine platform detection, legal regulations, business transparency, and consumer education. Walther's research is valuable for focusing on the "cognitive process of the reader" among these.


Reactions on SNS—"Calls for Severe Punishment" and "Expectations for AI"

Regarding the Phys.org article itself, being newly published, large-scale SNS discussions were still limited within the confirmable range. On Phys.org, the number of shares was small, and comments were not prominent. However, looking at the reactions to the fake review issue as a whole on SNS and business communities, interest is quite high.

In related LinkedIn posts, news about AI models detecting fake reviews introduced methods combining language analysis with signals from posting behavior. Although the number of reactions was not large, the direction of the discussion included both "welcome if AI can detect it" and "caution that the fake review side might also become more sophisticated using AI."

Additionally, industry posts dealing with review investigations showed strong consumer voices demanding penalties against companies that engage in fake reviews. Especially, opinions calling for exclusion from review sites, removal from search results, fines, and even criminal penalties were introduced, indicating a deepening distrust among consumers.

Summarizing the SNS atmosphere, fake reviews are no longer seen as "slightly exaggerated advertising." They are viewed as actions that steal consumers' time and money, disadvantage businesses that conduct honest trade, and destroy the review system itself.

On the other hand, there is also fatigue among consumers. The feeling of "every review seems suspicious," "neither five-star nor one-star can be trusted," and "in the end, I can only look at official information and return policies" is something many can easily share. The more reviews there are, the greater the burden of selecting trustworthy information. The fake review problem is also part of the "trust fatigue" in an era of information overload.


Fake Reviews Are Also a Risk for Companies

Fake reviews are not just a problem for consumers; they are a significant risk for companies as well.

In the short term, fake high-rating reviews might boost sales. However, if fraud is discovered, trust in the brand will be greatly damaged. Furthermore, in the United States, the FTC has issued rules prohibiting fake reviews and false endorsements, and in the UK, the Competition and Markets Authority is demanding stronger fake review countermeasures from companies like Google. The trend of legal regulation is clearly becoming stricter.

Companies need to be cautious not only about not writing fake reviews themselves. Issues can also arise from handling incentivized reviews, reviews by employees or affiliates, unjust removal of negative reviews, and self-promotion disguised as review sites. Although reviews may seem to lie between advertising and word-of-mouth, given their significant influence on consumers, transparency and fairness are required.

For honest companies, fake review countermeasures are both a burden and an opportunity for differentiation. Purchase verification, clear review posting policies, sincere responses to low ratings, reporting of unnatural reviews, and reflection in product improvements. Such steady efforts build long-term trust more than short-term star ratings.


How Consumers Can Read Reviews from Today

So, what should consumers be aware of when viewing reviews? Applying the CREM model to everyday shopping, first look at whether the review answers the information you want to know. Reviews that describe usage situations, duration, comparisons, and drawbacks are more helpful than simple "it's great" or "I recommend it."

Next, look at the trustworthiness of the poster. Is there purchase verification, is the posting history extremely biased, or are similar high ratings posted in succession on the same day? A perfect profile does not necessarily mean authenticity, but unnatural patterns can be warning signs.

Furthermore, check the balance of the review text. Actual users often write not only good points but also small complaints. Conversely, reviews that list only virtues like an advertisement or praise without touching on specific product features should be approached with caution. Discrepancies between star ratings and the content can also be clues. If it's five stars but the content is too thin, or one star without specific issues, it's better not to judge based on that review alone.

Finally, it's important to look at the overall distribution rather than a single review. Read both high and low ratings and look for common criticisms. If multiple people specifically mention the same flaw, it might be important information. Conversely, if similar phrasing of high ratings is concentrated in a short period, it's better to view them cautiously.


From "Believing" Reviews to "Decoding" Them

Online reviews are convenient. In fact, for many people, online shopping would be more unsettling without reviews. The problem is not in believing reviews unconditionally or doubting everything. What is needed is an attitude of decoding reviews as information.

Walther's research visualizes how consumers evaluate reviews and shows the potential to enhance judgment skills through training. This provides a perspective that sees fake review countermeasures not just as a technological issue but as a societal challenge involving consumer education, platform design, corporate ethics, and legal regulation.

The number of stars will continue to influence our shopping in the future. That's why it's necessary to have the ability to discern how those stars are created, who tells them, and how trustworthy they are.

The next challenge in the review society is not to increase reviews but to create an environment where trustworthy reviews can be found and suspicious reviews do not sway us. And the first step is for each of us to look a little more carefully behind the stars.


Source URL

Related Articles