Scroll through Airbnb for a few minutes and the language starts to blur together. Every place is “charming,” “cozy,” “stunning,” or a “gem.” Do those words actually persuade anyone, or do travelers tune them out because they know a host is trying to make a sale? That question sits at the heart of a long-running debate in marketing, advertising law, and economics, and until now it has been surprisingly hard to answer with real-world data.
A new study forthcoming in the Journal of Marketing Research takes on that question using more than $2 billion in Airbnb bookings. The result: gushing adjectives like “beautiful” and “luxurious” do move bookings, and by roughly the same amount as plain factual claims about a property.
The puzzle of “puffery”
In legal parlance, “puffery” refers to positive but vague claims sellers make about their products. Think Budweiser as the “king of beers” or Disney as a “wonderful world.” U.S. courts generally protect these claims from false-advertising lawsuits on the theory that reasonable consumers know to discount them. If someone says their apartment is “gorgeous,” you understand that’s an opinion, not a specification.
But that legal reasoning rests on an empirical assumption: that consumers actually do discount puffery enough for it not to matter. Most prior research on the subject has come from lab experiments, where participants react to hypothetical ads. Whether puffery moves the needle in a real marketplace where money changes hands has remained largely untested.
Michael Thomas, the author of the study, saw Airbnb as a rare chance to observe puffery in the wild. Hosts write short property descriptions and revise them from time to time. Those revisions offer a way to compare the same listing under different wording and see how bookings change.
Turning listings into data
Thomas worked with property snapshots collected by InsideAirbnb between 2015 and 2019 across 27 cities, from London and Los Angeles to Barcelona and Nashville. To analyze the language at scale, he used ChatGPT-4o to break each of 219,335 property descriptions into individual claims. A description like “Beautiful Big 3BR! Train/Bus 2min walk!” becomes claims like “the property has three bedrooms,” “the property is described as beautiful,” “the property is described as big,” and “the property is located near a train or bus station.”
The language model then scored each claim on how likely a court would consider it puffery. Those scores clustered into three natural groups. Purely objective claims (“the property has two bedrooms”) scored near zero. Weakly subjective claims that mix fact and opinion (“near downtown,” “ocean view”) landed in the middle. Puffed claims like “luxurious,” “charming,” and “stunning” scored near the top. Thomas cross-checked these groupings against actual court decisions and found broad agreement with how judges have historically classified similar words.
To measure how each type of language affected demand, Thomas built a statistical model that tracks how quickly a listing gets booked as a given stay date approaches. If a property attracts more interest, bookings tend to come in earlier. By comparing the same property before and after a description change (while holding photos and other features fixed), the model isolates the effect of the words themselves.
How much does language really move bookings?
Adding a single puffed claim to a description was associated with about 0.65 additional booked days per year for a typical property, roughly a 0.2% lift over baseline. An objective claim added about 0.52 days per year. Weakly subjective claims, the middle group, had the largest effect at about 1.00 additional days per year. Even a single exclamation point was linked to about 1.02 extra booked days per year.
As a placebo check, Thomas also counted “stop words” like “the” and “and,” which carry little meaning on their own. Their estimated effect was small and statistically indistinguishable from zero, offering some reassurance that the method isn’t just picking up noise.
The finding that puffery and objective claims perform similarly is notable because standard economic models of consumer search assume that shoppers place heavy weight on verifiable facts. Thomas offers a possible explanation for the modest effect of objective claims on Airbnb: much of that information is already visible elsewhere on a listing page or captured by search filters. When a description mentions “three bedrooms,” it often duplicates data the shopper already has. Weakly subjective claims like “near downtown,” which are harder to filter for directly, tended to add the most novel information and had the largest measured effect.
Moving a listing’s language from the 25th percentile to the 75th percentile of estimated effectiveness would add about 1.35 days of bookings per year for a typical property, corresponding to roughly $143 in extra revenue. That’s small at the individual level but potentially significant for hosts managing multiple properties or for large platforms in aggregate.
Does puffery backfire?
One worry about hype is that it raises expectations and then disappoints. If guests show up to a “stunning” apartment and find it merely fine, they might punish the host with lower ratings or unhappy reviews. Prior research on Airbnb photo quality has found evidence of exactly this kind of backlash for listings that upgrade their photography.
Thomas tested for a similar pattern with puffery and found little sign of it. Adding a puffed claim reduced numerical ratings by no more than 0.03 standard deviations, and the estimated effect on written review sentiment was slightly positive but not statistically significant. In fact, there was suggestive evidence that puffery increased reviewers’ mentions of trust.
Thomas is careful to flag a limit of this test. His data captures the average effect, not what happens to individual guests. If puffery attracts guests who were already inclined to like a place while turning off others who never book, the average masks whatever disappointment might be happening among the people it pulls in.
Implications for law, hosts, and search theory
The legal implications are nuanced. Some scholars have argued that courts overprotect puffery by assuming consumers ignore it. Thomas’s results push back on the “consumers ignore it” part: people do respond to language traditionally classified as puffery, and they respond about as much as they do to factual claims. At the same time, consumer protection law focuses on material deception, not mere influence. Because puffery doesn’t appear to reduce reported satisfaction on average in this setting, the author argues the findings are still consistent with current legal doctrine that treats puffery as protected but factual misrepresentations as actionable.
For hosts, the practical takeaway is modest. Puffery works, but weakly subjective factual claims work slightly better, especially when they add information not already available through Airbnb’s filters. A host debating whether to swap “luxurious retreat” for “two-minute walk to the beach” might get more from the specific detail. Exclamation points, essentially free, also appear to help.
Thomas also notes that Airbnb’s own search algorithm appears to reward listings for being updated, regardless of what changes. That “update reward” is a confounding factor his model controls for, and he suggests it may matter for other researchers using Airbnb data.
One caveat looms over the whole exercise. Airbnb shoppers arrive already searching, comparing structured listings side by side. That’s a very different environment from television advertising or highway billboards, where puffery competes for attention rather than differentiating similar options. Whether “the best a man can get” works the same way on a razor commercial as “cozy retreat” does on a rental listing remains an open question.
