Have you ever sat down to write an online review, stared at the blank text box, and typed out a quick “Nice place, would stay again” just to be done with it? Writing a thoughtful review takes effort. So what happens when a tool arrives that can do a chunk of that work for you? A new study offers evidence that the arrival of generative artificial intelligence is already reshaping how much people write, how long their reviews are, and even the emotional tone of what they say.
The research, published in the International Journal of Research in Marketing, takes advantage of an unusual real-world event to isolate the effect of one popular AI tool on consumer reviews.
A sudden ban creates a natural experiment
Weilong K.M. Gu and Martin Spann of the LMU Munich School of Management wanted to answer a question that has been hard to study directly. Generative AI, meaning software that produces new text, images, or other content, is now widely used. ChatGPT, the chatbot built by OpenAI, reached a million users within five days of its late-2022 launch. But measuring how such tools change everyday behavior is difficult, because you cannot easily compare a world with the tool to a world without it.
An event in the spring of 2023 gave the researchers something close to that comparison. On April 1, 2023, Italy’s data protection authority blocked ChatGPT over privacy concerns tied to European data rules. The ban was sudden, users did not see it coming, and it lasted four weeks before OpenAI addressed the regulator’s concerns and service resumed on April 28. During that same stretch, neighboring European countries were aware of the situation but did not impose their own bans.
That setup let the authors treat Italy as a “treatment group” that lost access to the tool, and compare it against a “control group” of countries where the tool stayed available. This kind of design is called a difference-in-differences analysis. In plain terms, the researchers looked at how reviews changed in Italy after the ban, then subtracted out whatever changes happened over the same period in the comparison countries. Whatever difference remains is attributed to the ban itself.
Counting reviews across five countries
The authors turned to TripAdvisor, one of the most visited travel websites in the region, and gathered reviews for 4,909 hotels. Of those, 1,930 hotels were in Italian cities including Milan, Rome, Florence, Venice, and Naples. The remaining 2,979 hotels sat in comparison cities across Germany, France, Spain, and Greece, chosen for their regional and cultural closeness to Italy and their high tourism volumes.
One reason they picked TripAdvisor is that it records both the date of a person’s hotel stay and the date the review was written. That detail let the researchers keep only reviews written during the ban window and drop ones posted long after a trip, reducing the risk of counting a traveler who had already left Italy and was no longer affected. They organized the data by hotel and by week, covering four weeks before the ban through four weeks after, which produced 117,816 observations once mobile and desktop reviews were separated.
That split between mobile and desktop turns out to matter. At the time of the ban, ChatGPT had no mobile app, so using it on a phone was awkward. The authors expected any effect of the tool to show up mainly among people writing on desktop computers, where ChatGPT was easier to use.
What the numbers showed
The pattern matched that expectation. When ChatGPT became unavailable in Italy, the number of reviews written on desktop devices fell by roughly 2 percent compared with the control group. The average length of desktop reviews dropped by about 9 percent. In other words, when the tool disappeared, people wrote slightly fewer reviews and noticeably shorter ones.
The researchers ran several checks to test whether the result was real or a coincidence. They performed a “placebo test,” pretending the ban had happened at an earlier date when nothing actually changed, and found no effect, which is what you would hope to see. They also extended the time window, swapped in different statistical controls, and used an alternative matching method that pairs Italian hotels with similar foreign hotels based on price, star rating, and chain membership. The desktop effect held up across all of these.
Why access to the tool seems to matter
The authors propose that the main reason is effort. Writing a review by hand takes work, and an AI assistant lowers that cost. To test this idea, they looked at review ratings, building on earlier research linking the effort a task requires to the scores people give. They reasoned that when writing is harder, some frustration may bleed into a lower rating. Consistent with that, the analysis found that desktop review ratings dipped when ChatGPT was unavailable in Italy.
To supplement the hotel data, the researchers ran an online survey through the platform Prolific with 95 participants in the United States and 88 in Italy, all of whom had written reviews before. Among U.S. respondents, 63 percent said they had used generative AI tools to help write reviews, as did 32 percent of Italian respondents. Usage was higher on desktop than on mobile in both groups. Among those who used the tools, 73 percent of U.S. respondents and 68 percent of Italian respondents said they did so to reduce effort. Most did not simply copy and paste the AI’s output; larger shares said they made minor edits or used it as inspiration and wrote the review themselves.
Changes beneath the surface
Beyond volume and length, the authors examined two features of the actual text. The first was sentiment, or how positive or negative the language sounded, measured with a tool called VADER. Focusing on one- and two-star reviews, they found that when ChatGPT was unavailable, those negative reviews were slightly less negative in their wording. The size of the shift was small, but it was statistically significant. The authors suggest that access to the tool may let unhappy reviewers express criticism in a more pointed yet less blunt way.
The second feature was subjectivity, meaning how much a review reflects personal opinion versus plain facts, measured with a library called TextBlob. Here the analysis found that ChatGPT availability was linked to more subjective, opinion-driven review text on desktop devices.
What it might mean for businesses
The authors note that earlier research has connected higher review volume and longer reviews to higher sales. Read alongside that work, their findings suggest that wider adoption of generative AI could expand the pool of reviews a business receives and give firms more detailed feedback to act on, whether for product improvement or spotting problems early. They also point out that platforms may need better ways to filter and summarize a growing flood of content, citing features such as Amazon’s AI-written review summaries.
A few caveats deserve attention. The study covers a single type of purchase, hotel stays, over a short four-week window, so the results may not carry over to restaurants, physical products, or longer time frames. The authors also cannot fully rule out that some Italian users got around the ban with a VPN, or that the ease of AI writing may eventually raise the number of fake reviews. And because the data was grouped at the hotel level, they could not track individual users switching devices or changing habits. The researchers frame their work as an early look at a fast-moving phenomenon rather than a final word, and they flag whether AI-assisted reviews become more standardized and less personal over time as a question worth watching.
