Online reviews guide what customers buy for a wide range of products and services. Generative AI models can now help anyone write a review that reads as human, with the help of the likes of a general AI writing assistant or a specific AI review generator. That makes the star rating at the top of a product page harder to trust.

Bots and automation add to the problem, with reviews that can be generated around the clock and at scale – so a rating built on hundreds of glowing reviews may say more about a seller’s ability to use AI writing tools than about its customers.

UNSW Business School Professor of Marketing, Nitika Garg, is a consumer behaviour researcher with expertise in consumer judgment and decision-making. She explained that using AI for review generation makes it easier for brands or individuals to generate numerous, human-sounding reviews in mere seconds.

“With effective prompt management, this makes the whole process effortless and thus more tempting to engage in, while making it harder for consumers to discern them as AI-generated,” she said. “Conversely, sometimes a real consumer using AI to clean up their review might come across as less ‘human-like,’ blurring the lines between what a typical consumer understands as human and authentic versus AI and inauthentic.”

Yale School of Management Professor Balázs Kovács tested whether people could tell real Yelp restaurant reviews from reviews written by ChatGPT-4. He found that the people taking part were often wrong, even when they were paid for correct answers. They mistook human reviews for AI and, more often, mistook AI reviews for human ones.

Participants were also more confident that the AI-written reviews were real than they were about the human-written ones. Prof. Kovács also tested AI detection software, which was fooled by the AI-written reviews.

How fake online reviews influence buying decisions

Star ratings act as a quick and trusted guide to assessing reputation and credibility. When shoppers cannot see or test a product, other customers’ ratings and product reviews stand in for their own experience – a process marketers call “social proof”. Analysis of review data by Northwestern University’s Spiegel Research Center found purchase likelihood peaks when a product’s average rating sits between 4.2 and 4.5 stars. It then falls as the rating approaches a perfect five, which the researchers suggest reflects suspicion of manipulation. Number of reviews is another, often used heuristic by consumers where larger number of reviews increases consumer confidence in them.

Learn more: Marketing in the AI agentic enterprise: Lessons from Salesforce

”These judgment heuristics or mental shortcuts, typically help consumers in arriving at reasonable, if not always optimal, decisions quickly and efficiently,” said Prof. Garg. “However, the proliferation of fake online reviews calls the continued validity of such heuristics into question. Consumers may continue to rely on them even as the cues they depend on become less trustworthy, leaving them vulnerable to undue influence from fake reviews. This concern is compounded by research which shows that lower-quality products are more likely to use fake reviews to boost their ratings.”

Why fake reviews are hard to spot

Fake reviews were a problem even before generative artificial intelligence. In December 2023, the Australian Competition and Consumer Commission (ACCC) released the results of an internet sweep of 137 businesses. It found 37 per cent had engaged in concerning conduct with online reviews. The ACCC also examined 24 businesses that sell services to create fake reviews or remove negative ones.

Shoppers also tend to believe what they read. Dezhi Yin from the University of South Florida and his co-authors ran five experiments. They found that consumers tend to judge reviews as real, a tendency psychologists call truth bias. In one experiment, participants were told only 10 of 20 restaurant reviews were authentic. They still classed an average of 11.38 reviews as real, according to the university.

"Once verified, negative feedback needs to be acknowledged by firms and mitigating steps outlined in response to reviews"

NITIKA GARG

The study also found readers place more trust in negative reviews than positive ones. Yet the researchers note that platform review data shows negative reviews are more likely to be fake. “This is an illustration of the well-established ‘negativity bias’ where consumers trust negative reviews more because they are perceived as more diagnostic and less prone to deception,” said Prof. Garg. “In the online reviews context, this bias persists even for more experienced and familiar consumers.”

This leaves businesses exposed, because consumers may not only be more likely to accept a fake one-star review than a fake five-star one, they are also more likely to be influenced by it. “Thus, platforms and businesses should monitor negative reviews, real or fake, and respond promptly,” said Prof. Garg. “For example, platforms can regularly verify reviewer data with their user activity and purchase history to validate reviews. ‘Verified purchase’ labels accompanying reviews are one increasingly common way to signal this information to consumers.”

How to spot fake reviews

Four checks give shoppers a better guide than wording: evidence of a real experience, the pattern of reviews, disclosure, and the stakes of the purchase. No single check proves a review is fake, so it’s a good idea to go through a systematic review of all four and look for more than one warning sign before deciding.

1. Experience: look for signs of a real purchase or visit. A detailed review often mentions specifics, such as the model bought or how a problem was resolved. The ACCC lists generic reviews that lack detail as one sign a review may be fake. AI can invent detail too, so weigh even a detailed review alongside the other checks.

2. Pattern: check the reviews around it. The ACCC’s guidance for consumers flags a spike in highly positive or negative reviews over a short period. It also flags reviews that use language similar to other reviews for the same business. Clicking through to a reviewer’s profile can show whether the account has a history of reviews or appeared in the same week as dozens of others. Read beyond the top-rated product reviews, as the three-star and one-star reviews can show whether complaints repeat and how the business responds. Compare ratings across platforms, too. A business rated 4.9 on its own website and 3.8 on an independent site warrants a closer look.

Subscribe to BusinessThink for the latest research, analysis and insights from UNSW Business School

3. Disclosure: look for incentives and platform rules. Under ACCC guidance, any incentive offered for a review must apply whether the review is positive or negative, and must be disclosed. The ACCC also says businesses and platforms should display the number of reviews behind a rating (for example, three stars from 24 reviews). They should also display their policy for publishing, editing and removing reviews. Platforms that require proof of purchase may also be more reliable than those that do not.

4. Stakes: match the checking to the size of the decision. A $15 phone case does not need the scrutiny you would give a builder or a used car. For bigger decisions, look for evidence beyond star ratings, such as licensing records or seek advice from someone you trust.

What businesses and review platforms should do about fake reviews

Businesses can protect the value of their reviews by how they collect customer feedback. Ask every customer for feedback, including those who had a problem. Google’s Maps content policy, for example, bans businesses from paying for reviews or selectively seeking positive ones. It also bars businesses from setting staff targets for the number of reviews they collect.

It’s good practice to just leave the review writing to customers. A regular systematic review of incoming reviews gives a business actionable insights, such as repeated complaints or a sudden cluster of near-identical praise. If a business suspects a review is fake, the ACCC advises asking the platform to remove it or responding publicly to correct the record. The ACCC also warns businesses to have an adequate basis before removing or editing a review, since removing a genuine negative review may itself mislead consumers.

Platforms are also responsible. Research published in Marketing Science tracked Amazon products whose sellers bought fake reviews through private online groups. Roughly half of those products’ reviews were eventually deleted, but only after an average of more than 100 days. In the meantime, sellers kept the benefit of higher ratings and sales. Once the sellers stopped buying reviews, average ratings fell, and the share of one-star reviews rose.

The researchers suggest that platforms prioritise the prompt detection of fake negative reviews, given that readers are more likely to believe them. Platforms that show proof of purchase also give readers more to judge than the text alone, which matters when an AI review generator can draft that text in seconds.

Encouraging real, honest (human) reviews

“Businesses can also think about strategies to encourage real consumers to provide verifiable feedback such as through video testimonials or by connecting their feedback to their purchase in the system, to increase consumer confidence in them,” Prof. Garg explained.

“Once verified, negative feedback needs to be acknowledged by firms and mitigating steps outlined in response to reviews. Several companies do so directly on platforms by, for example, apologising for a consumer’s negative product experience and suggesting ways to address the same (such as a call back from a customer representative, reference to the user manual, etc.).”

Learn more: Understanding customer decision-making: The Amazon challenge

Both honest customers and fraudsters will use artificial intelligence to write reviews. Guessing who wrote a review offers little protection. Reviewing the evidence surrounding it and asking whether a real customer is behind it gives consumers a better basis for making a decision.

”If AI-generated content continues to erode trust in online reviews, consumers increase their reliance on other forms of feedback, such as word-of-mouth recommendations from people in their personal networks and communities, which may be less susceptible to manipulation,” said Prof. Garg.

“Brands can also seek to build consumer confidence by giving consumers more opportunities to experience a product or service before purchasing. Paradoxically, the erosion of trust in online information may bring marketers ‘full circle,’ prompting a renewed emphasis on personal recommendations and direct experience.”

3 recommendations for industry professionals

1. Marketing and customer experience teams should invite feedback from every customer, rather than selecting customers they expect to leave positive reviews.
2. Trust and safety teams should examine reviewer activity, timing, purchase evidence and repeated language. Text alone cannot establish whether a customer had a real experience.
3. Compliance and brand teams should keep a clear record of review policies, incentives and decisions to edit or remove reviews. ACCC guidance says businesses should not remove genuine negative reviews without an adequate reason.