Fake, AI-Generated Game Reviews Are Overrunning Steam

Today, it’s arguably harder than ever before for potential video game buyers to figure out whether a game is worth purchasing or not based on reviews alone.

You definitely can’t trust the “critic” reviews – usually written by gaming journalists who, more often than not, aren’t real experts on games and, even worse, tend to run defense for AAA studios by giving their games much higher scores than they deserve – but you also can’t blindly trust user reviews either, as it is, sadly, quite common for fans or haters to manipulate scores with waves of upvotes or downvotes, skewing and distorting the game’s actual reception.

One of the few remaining vestiges of genuine video game evaluation, as I’m sure many would agree, are Steam’s user reviews. Only people who’ve actually tried the game can leave them; no voice is louder than any other, meaning biased actors don’t get any more weight than anyone else; certain biases, like “Product Received for Free,” are marked for everyone to see; each review displays how much time its author has spent on the game, allowing you to judge how trustworthy they are; with just a few clicks, you can check each reviewer’s profile to see if they might be pushing an agenda with their assessment – the list of Steam’s system’s advantages goes on and on.

However, it seems that even this somewhat-reliable source for game scores may soon lose its credibility, with a recent study by Copyleaks, a team of experts in AI text analysis, highlighting a growing issue of fake, AI-generated reviews overrunning Steam.

As part of their research, the team analyzed numerous Reddit comments from users complaining about artificial reviews flooding the platform, connecting these complaints to multiple accounts responsible for dozens – sometimes hundreds – of fake-looking reviews and examining those reviews to assess whether they were AI-generated.

Using common signs of generative AI, such as “repetitive phrasing across different game genres, generic praise or criticism with minimal game-specific detail, and absence of personal insight or gameplay anecdotes,” Copyleaks ran the reviews through their own AI Detector tool, confirming they were not written by flesh-and-blood human beings. Honestly, judging by the reviews in question, they probably didn’t even need to bother copying and pasting them into the detector:

The most concerning part is that some of these AI-generated reviews date back to mid-2023, meaning Valve had plenty of time and ample opportunity to remove them and implement precautionary measures to automatically detect and delete such content, thereby protecting the credibility of Steam reviews, but simply chose not to.

As troubling as it may be, it is, unfortunately, hardly surprising. At this point, it’s clear which side Valve has taken in the AI vs anti-AI conflict, with the studio not being particularly strict about enforcing its own rules on AI disclosure and Gabe Newell now openly touting AI as the next major technological leap, comparable to computers or the internet.

In light of that, it’s hard to imagine anything will be done about artificial reviews, and the fall of Steam Reviews as a reliable source may well be inevitable, with only time revealing how Valve’s indifferent approach to AI – along with that other, arguably much more important Steam-related controversy – will ultimately impact the company’s goodwill. Keep in mind, just a decade ago, Ubisoft was universally beloved.