It’s no secret that AI has taken over search results, and while there are critics of this trend, the majority of end users are generally accepting of the summaries generated whenever they use Google. While not a problem in the abstract, there are some cases where summaries created by LLMs actually pose real-world risks because they miss some vital piece of information in their efforts to be as short and snappy as possible. Review summaries are the perfect example of this, so let’s talk through what problems they pose and why searchers should still click through to read the full coverage of a product or service before committing to a purchase.
The Filtering Effect
The AI models used by search giants are trained on vast volumes of data, and that process of taking a lot of information and sieving it down to its core components means a lot can be lost along the way, especially aspects that only crop up rarely in the bulk of the training data. In the case of reviews, that can mean negative feedback or flaws don’t find their way into an AI summary.
Take the example of online casino site Planet 7. The AI summary from Google correctly identifies that some users and professional reviewers have complained about lengthy payout times. However, it doesn’t pick up the lack of live casino or roulette, while also missing out the upsides, including the impressive welcome bonuses and the variety of slots available. That’s why reading full Planet 7 Casino reviews is still necessary, so that you know what you’re signing up for.
The Potential for Sterility
Another drawback of review summaries composed by AI is that they may not convey the human emotion behind the general sentiment expressed by users and journalists for a given product or service. This applies to good and bad reviews alike, but matters more when there are risks involved.
So, let’s say you’re looking to buy a car. Reading "The brake pads failed on a steep hill and I almost crashed" triggers an immediate, emotional "do not buy" response. Reading the AI summary version, which might read something like "A small percentage of users noted braking inconsistencies" because of how it pares down and neutralizes the intent and general consensus, also completely sterilizes that sense of danger.
The Problem with Looping Feedback
One last worrying aspect of AI review summaries that needs to be known to the wider public is that because LLMs are now responsible for so much content generation, both at a professional level and by individual web users, a lot of what’s created is now feeding back into the training data. In other words, AI models are regurgitating talking points and opinions that other AI models put together, rather than reflecting human-written opinions.
In short, people must be cautious when checking out AI review summaries, and not simply assume that the short overviews and bullet points provide a full picture of the positives and risks of any product or service.
