Skip to main content

Do listing photos that stand out win more bookings? What a Journal of Consumer Research study found

On Airbnb and other peer-to-peer sites, buyers judge one-off offers mostly from pictures. A model trained on New York listing photos, then checked against people’s eyes and choices, shows that looking different pays only up to a point, and that strong ratings let bolder listings go further.

Feng et al. scored 481,747 Airbnb photos for visual uniqueness: bookings rose with it up to a point, then fell, and good ratings softened the fall.
Date

7 November 2026

Marketing in 3 · 2025 in research · No. 19 · Paper published 10 May 2025

Three pressed-glass goblets with a thumbprint pattern in cobalt blue, olive green and amber on a white background, the green one standing upright between two turned upside down
Three of a kind, and the eye goes straight to the one that breaks the pattern. Photo: Meg MacDonald on Unsplash.

The short answer

Up to a point. A model rated how distinct each New York Airbnb photo looked next to all the others. Set against occupancy, the scores traced an arc: listings with fairly ordinary pictures gained as those pictures grew more distinctive, but beyond a middle level demand slipped again. Very distinctive homes lost less when guests rated them highly and hosts replied quickly. Tested as a simple straight line, uniqueness showed no link to bookings at all.

Evidence level
  1. One study
  2. Several studies
  3. Meta-analysis

Key takeaways

  • For listings on the rising side of the arc, photos one standard deviation more distinctive went with 15.9% higher occupancy, worth roughly $9,810 in extra yearly revenue for an average New York listing, by the authors' reckoning, if it stays open all year.
  • Editing out three or four objects the model had flagged, such as lamps, a pillow or wall art, made online viewers about 22 percentage points less likely to choose that photo as one of the most unique.
  • Each extra point on the model's scale raised the chance that lab participants put a photo among their three most unique, out of 12, by about 4 percentage points.
  • In the authors' worked example, lifting the guest rating from 80 to 90 out of 100 was worth roughly nine times more to a listing past the sweet spot than to one still short of the peak.

What the researchers did

Xiaohang (Flora) Feng of Carnegie Mellon University in Pittsburgh, Charis X Li of Grenoble Ecole de Management in France and Shunyuan Zhang of Harvard Business School looked at a dilemma that every host and second-hand seller knows. On a peer-to-peer marketplace no two offers look alike, and buyers size them up mainly from photos. Should your pictures look different from everyone else's, and if so, how different? The paper grew out of Feng's doctoral dissertation, which Li and Zhang supervised.

The hard part was measuring it. Earlier work judged a product against a typical example of its category, but there is no typical Airbnb flat. So the team built a model that compares every photo with many others and scores how far each one stands out, without any human ratings to learn from.

The work then ran in three steps: train the model on the New York pictures, check its scores against people in three studies, and link the scores to 13 months of occupancy for more than 14,000 listings.

Photos
481,747 images from New York listings in 192 zip codes, gathered every month from April 2022 to April 2023
How the model learns
By comparing each photo with randomly drawn others, so no human labels were needed
Checks with people
Eye tracking with 212 students and a choice task with 288 students, both at European universities, plus an online test with 399 adults in North America
Booking data
14,121 listings over 13 months from the data firm AirDNA, 95,673 listing-months in all
Held constant
Price, ratings, reviews, amenities, size, local supply, county and month, plus how attractive and how cluttered the photos were

How a machine judges what stands out

The model never saw a human rating. It took each photo, made two slightly altered copies (flipped, blurred, turned grey or shifted in colour) and tried to pick that pair out of a batch of 256 randomly drawn photos. A picture whose two copies stay easy to match among strangers is one that looks unlike its neighbours, so it earns a higher uniqueness score. On 61,012 photos held back for testing, the model's first guess was the right partner 73.1% of the time, within the 60% to 80% range of leading models of this kind.

Each score comes with a heat map showing which parts of the picture push it up (red) or down (blue). In one bedroom that scored 14, the red sat on bookshelves either side of the bed, an orange blanket and a bicycle hung on the wall. In a bathroom that scored 2, the blue gathered around the toiletries.

Uniqueness in this sense is neither beauty nor busyness. Across all the photos, the scores were essentially unrelated to an algorithm's rating of how attractive a picture is (a correlation of minus 0.01) and to the number of objects in it (0.0088).

Do people see what the model sees?

Three studies put the scores in front of people. In an eye-tracking lab, 212 students wore tracking glasses and browsed 12 real listing photos, eight bedrooms and four living rooms, as if planning a New York trip. Photos with higher scores held their gaze longer and drew more glances, with correlations of 0.65 and 0.60 across the images. The aesthetics rating did not predict where they looked, and simpler pictures pulled in more attention than busy ones.

In the second study, 288 students chose the three most unique photos from the same sets and then clicked five spots that made each choice special. Their clicks landed where the heat maps were red, for example on a plush toy, a curtain and a lamp in one bedroom. Photos the aesthetics algorithm rated as better looking were actually less likely to be chosen.

The third study tested cause and effect. The team removed the objects flagged in red from three photos that had each scored 14, and the model's scores for them collapsed. Shown collages of six photos online, 399 adults were much less likely to call the edited versions unique. Removing objects from blue areas, the authors note, would raise a score instead.

Model uniqueness score for three listing photos, before and after editing out flagged objects
Model uniqueness score for three listing photos, before and after editing out flagged objects0481216Each photo as listed: 14.0 Each photo as listed14.0A: pillow, lamps, wall art out: 4.4 A: pillow, lamps, wall art out4.4B: toy, lamps, carpet, ottoman out: 2.5 B: toy, lamps, carpet, ottoman out2.5C: TV, books, wall art out: 3.1 C: TV, books, wall art out3.1
Model uniqueness score for three listing photos, before and after editing out flagged objects
Group Value ()
Each photo as listed 14
A: pillow, lamps, wall art out 4.4
B: toy, lamps, carpet, ottoman out 2.5
C: TV, books, wall art out 3.1

Study 3, the three photos edited for the online test with 399 adults; all three originals scored 14. Scores from the same model, as reported in the paper.

Bookings follow an arc

With the model checked, the team gave every listing a monthly score by averaging its photos and set that against occupancy. Tested as a straight line, uniqueness had no link with demand. Allowed to curve, it traced an inverted U: occupancy climbed as photos grew more distinctive, peaked at a score of 12.61 and then declined. The authors read the decline as uncertainty. A home that looks like nothing else is intriguing, but guests are less sure what they will get.

The average listing scored 12.85, close to the peak, and listings ranged from about 8 to 16. Because the curve is steepest at its ends, the biggest gains and losses belong to listings far from the middle in either direction.

The shape held when demand was counted in reserved days, when the score was calculated in two other ways, and when the analysis also allowed for how far each photo sat from the average New York listing photo. Listings whose photos were consistent with one another did better than sets that swung between very plain and very unusual shots.

Good ratings make bold photos safer

If the decline comes from uncertainty, anything that reassures guests should soften it. The team tested two such signals: the share of enquiries a host answers within 24 hours, and the listing's overall guest rating. Both flattened the arc, so distinctive listings lost less at the far end when hosts were responsive and past guests were happy.

To put money on it, the authors compared two otherwise average listings, one scoring 10, below the peak, and one scoring 14, beyond it. The same improvement in either signal was worth far more to the bolder listing, and a better rating counted for more than faster replies.

Estimated extra revenue a year from stronger trust signals, in US dollars
Estimated extra revenue a year from stronger trust signals, in US dollars050010001500200025003000Replies 80% to 90%, below peak: 91 Replies 80% to 90%, below peak91Replies 80% to 90%, past peak: 394 Replies 80% to 90%, past peak394Rating 80 to 90, below peak: 291 Rating 80 to 90, below peak291Rating 80 to 90, past peak: 2659 Rating 80 to 90, past peak2659
Estimated extra revenue a year from stronger trust signals, in US dollars
Group Value ()
Replies 80% to 90%, below peak 91.4
Replies 80% to 90%, past peak 394.1
Rating 80 to 90, below peak 291.3
Rating 80 to 90, past peak 2658.8

Demand model, 14,121 New York listings. The authors' worked example for two otherwise average listings scoring 10 (below the peak) and 14 (past it), open all year.

What the authors suggest

For platforms, the authors see room to feature listings with the right level of distinctiveness in campaigns and on social media, to add a uniqueness score to search and recommendations, and to show the most unusual homes to guests who are looking for something new.

For hosts, they picture a tool that does not exist yet: an Airbnb feature that scores each photo and shows its heat map, to guide redecorating or the choice of pictures. In their data, simply dropping each listing's least distinctive photo lifted the average listing score by 0.53 standard deviations.

The method should travel to other places crowded with one-off images, such as Instagram, property sites like Zillow or Amazon's marketplace. The authors insist the model would need retraining on pictures from each new setting; the New York scores cannot simply be reused.

Marketing in 3 steps

Ready. Set. Market.

ReadyDecide and prepare
  • Put your photos in the line-up. Search your marketplace the way a buyer would and screenshot the first 20 listings that compete with yours. Drop your own photos into the same grid and note which ones you would scroll past.
  • Rate every shot for difference. Score each of your photos from 1 to 5 for how different it looks from the other listings in the grid. Circle any picture that could belong to anyone, such as a bare hallway or a plain bathroom.
SetBuild it
  • Add one memorable object per room. Give each main room one or two things a buyer will remember, such as a bright throw, a full bookshelf or a bike on the wall. Keep the rest of the frame calm so the eye finds them.
  • Cut the dullest shot and even out the set. Remove the photo you scored lowest. Then look at the set as a whole: no picture should be far plainer or far stranger than the others, and the later photos deserve the same care as the cover.
MarketLaunch and measure
  • Back an unusual look with reassurance. If your listing now looks out of the ordinary, answer every enquiry within a few hours and put your rating and best reviews where buyers see them first. Ask each happy customer for a review within two days.
  • Change one photo at a time. Swap a single photo, then compare views, saves and bookings over the next four weeks with the four weeks before. Keep the change only if the numbers improve, and stop adding quirks once they start to slip.

Get it by email

New issues every Monday and Thursday, in your inbox. Free, no ads, unsubscribe anytime.

Your address is used only to send Marketing in 3. You will get one email to confirm.

What to expect

  • An arc, not a ladder. Distinctive photos help only until a listing hits the sweet spot. Beyond that, extra quirkiness costs bookings rather than winning them.
  • Room for trial and error. What makes a photo stand out is particular to that photo, so the authors expect hosts to test changes rather than follow a fixed recipe.
  • Reassurance has to come with it. On this evidence, a very unusual look sits best with listings that already have strong ratings and quick replies; without them it carries more risk.
  • Illustrations, not forecasts. The dollar figures assume a listing open every day of the year and an effect that lasts all year, so read them as a sense of scale.

Where it does not work

  • One city, one platform. Every photo and booking came from New York Airbnb listings, so other cities, other marketplaces and other kinds of product may behave differently.
  • Listings, not clicks. Demand was measured per listing per month. The data cannot tell whether a single photo drew more clicks, views or messages.
  • Pictures only. Titles, descriptions and reviews were left out, although they also shape how unusual a listing seems.
  • Observational bookings. The authors used instruments and many controls to rule out other explanations, but hosts chose their own photos, so the link with demand is not an experiment.
  • Student panels in the lab. The eye-tracking and choice tasks used students at European universities; only the edited-photo test drew on a broader online sample.
  • The theory guided the model but was not tested. The idea that people judge uniqueness by comparing whole images with one another was assumed in building the model, not examined directly.

Questions marketers ask

Do Airbnb listings with unique photos get more bookings?

Up to a point. Across more than 14,000 New York listings, occupancy rose with the distinctiveness of the photos only until the middle of the range; the most unusual listings did worse again (Feng et al., 2025).

How is visual uniqueness defined here?

How much an image stands apart from the other images a viewer is comparing it with, judged on the whole picture rather than a checklist of features. It is a separate quality from how attractive or how cluttered a photo is.

Is Airbnb's search ranking behind the effect?

Probably not. Without access to the ranking algorithm, the authors ran searches across different prices and dates and looked at 270 results spread over 15 pages each time. A listing's uniqueness was not linked to where it appeared.

Does the cover photo matter most?

Not by itself, at least in this data. Once the rest of the set was allowed for, the cover photo's distinctiveness made no measurable difference, while the average of the other photos followed the same arc. The authors think guests weigh the full set before they book.

Can sellers on other marketplaces use this?

The approach, yes; the scores, no. A model would first need fresh training on images from the new setting, such as a resale app, a property site or an online store, before its scores meant anything there.

Also new at the time

Journal of Consumer Marketing · 2025

Brand posts with memes on social media: perceived humor and consumer engagement

Sewak, Lee, & Haderlie

Six studies, five experiments plus an analysis of real brand posts on X, compared posts built around a meme with posts without one. Meme posts came across as funnier, and that humour brought more likes and shares. The format backfired in cause-related campaigns, where memes lowered engagement. A joke can carry a light message, while a serious cause may need a straighter voice.

International Journal of Retail & Distribution Management · 2025

Will consumers pay for e-fashion? A multi-study investigation

Pino, Pichierri, & Sit

Three studies, one of them in a real buying situation, asked what people will pay for digital clothes, garments that exist only on screen. The more unique shoppers judged a digital piece to be, the more they were willing to pay. The effect was stronger among sensation seekers and among people who normally like to handle a product before judging it. For virtual fashion, uniqueness looks like the thing worth charging for.

Source

Feng, X., Li, C. X., & Zhang, S. (2025). Visual Uniqueness in Peer-to-Peer Marketplaces: Machine Learning Model Development, Validation, and Application. Journal of Consumer Research, 52(4), 800-825. https://doi.org/10.1093/jcr/ucaf021

This brief is written for practitioners from the published article. Open access (CC BY 4.0): free to read at the publisher. Numbers are as reported by the authors; read the article for full methods.

Keep reading

More issues on Creative · AI · Retail

2026 · No. 45 ·

AI ad backgrounds that click and stay on brand

Daviet and Nishimura trained an AI on nine weeks of Instagram ads: its background images averaged 0.98% link clicks, against 0.65% for a designer's.

Bruno Schivinski

Bruno Schivinski is Associate Professor of Marketing at Gdańsk University of Technology. He studies how consumers engage with brands in digital and social media.

About Bruno · All Marketing in 3 issues · Subscribe