A temporary restriction on ChatGPT in Italy gave researchers a way to examine whether access to generative AI changed the reviews travelers posted about hotels.
Weilong K. M. Gu and Martin Spann compared TripAdvisor reviews for 1,930 hotels in Italy with those for 2,979 hotels in Germany, France, Spain and Greece, where the same restriction did not apply. Their study in the International Journal of Research in Marketing was published online in June 2026.
The researchers report that access to ChatGPT was associated with a greater volume of reviews and longer reviews, particularly on desktop devices. Review text also became more subjective and expressed more extreme sentiment in their analysis.
The findings concern a particular platform, set of hotels and period. They do not establish that every review written after the restriction was lifted used AI, or that an AI-written review describes a fabricated stay.
A temporary restriction created a comparison
Generative AI tools can produce or revise text in response to instructions. A traveler might use one to translate a genuine account, organize thoughts or change its tone. A tool can also generate statements that are inaccurate or not based on an actual experience.
Italy’s temporary restriction lasted about four weeks in 2023. The researchers used a method called difference-in-differences, comparing changes in reviews for Italian hotels with changes over the same period in the comparison countries.
The design offers a closer comparison than looking only at review counts before and after the restriction. It still depends on assumptions about how the groups would have changed without the policy. Countries can differ in travel patterns, seasonality and the composition of visitors, and other writing tools may have remained available.
The authors interpret the pattern as consistent with AI reducing the effort needed to write a review. Their results do not directly identify the authoring process behind individual reviews.
Longer feedback needs careful reading
For a hotel or another business, an increase in review length is not automatically an increase in reliable detail. A longer account can give readers more context about service, cleanliness or location. It can also repeat general phrases or exaggerate an experience.
Businesses analyzing customer feedback can separate a change in language from a change in the service itself. Repeated complaints about room temperature, for example, can be checked against maintenance records and support requests. A shift in the emotional wording of reviews alone does not show why customers feel differently.
Our earlier reporting on AI-generated hotel review summaries considered a different step in the process: how a platform presents existing comments to potential guests. The Italian study concerns the content customers contribute.
Trust depends on the experience behind the words
A business should avoid assuming that polished language proves a review is false, just as a short or grammatical review is not proof of a genuine purchase. Review platforms can use purchase verification and moderation processes, but these checks have their own limitations.
In the United States, the Federal Trade Commission’s endorsement guidance says endorsements should reflect honest opinions and relevant commercial connections should be disclosed. Rules differ by jurisdiction and platform.
For managers using reviews to decide what to fix, the strongest approach is to compare comments with other evidence and track whether the underlying complaint recurs. The language used to express a complaint can change faster than the service being described.
