To Gain Trust, Write This
Imagine you’re planning a trip and two strangers offer to host you in their homes for a fee. Both have a one-line profile. The first profile says, “Life is beautiful, so let’s enjoy it.” The second says, “We look forward to hosting you.” All else being equal, which of these strangers would you feel more comfortable staying with?
This kind of process — choosing whom to trust people based on their online profiles — determines far more than just where you sleep on vacation. Online profiles affect how fast your ideas spread, whether you get hired, and sometimes whom you end up marrying. Unfortunately, profile designs are often flawed, and at least partially contribute to redlining on Uber and Lyft and racial discrimination on Airbnb. We therefore need a better understanding of how profiles are perceived.
This is what we, researchers at Cornell Tech and Stanford, set out to study in a recent paper, to appear later this month at the 20th ACM Conference on Computer-Supported Cooperative Work and Social Computing (CSCW 2017). Specifically, we looked at the perceptions of trustworthiness in Airbnb host profiles. We examined over 1200 unique host profiles across 12 U.S. cities in order to determine which topics engender trust and ultimately influence choice of host.
Our first question was: what do hosts talk about in their profiles?
Hosts were most likely to talk about where they’re from and where they live right now. About 70% of host profiles mention this topic. Origin and residence is followed by talking about what they do for a living, or where they go to school, about 60% of the time. Next, people begin to add colors, and write about what they do for fun, what their personality is like, and their significant others and kids. There are two topics that are Airbnb specific. About 50% of the time, hosts will say that they love to travel, and make sure that you know they look forward to hosting you. Finally, 8% of the hosts who have a profile talked about their life motto and values (as prompted by Airbnb). We will see in a second that this may not be a great idea after all.
Now, not all hosts have the same relationship with their guests. On-site hosts share living space with their guests, while remote hosts live somewhere else. What we found is that on-site hosts write 18% more words than remote hosts (66 words v.s. 56 words on average), and they are more likely to talk about their interests, tastes, and personality — perhaps angling for a better match.
Next, do different topics make a difference in how trustworthy these profiles are perceived to be?
To answer this question, we recruited participants from Amazon Mechanical Turk to rate Airbnb host profiles for trustworthiness. We found, perhaps unsurprisingly, that longer profiles are perceived to be more trustworthy. Similarly, profiles that touch on more topics inspire more trust. But not all topics are created equal. We conducted an analysis on the relative effectiveness of topics, we found that the language of hospitality (e.g. “We look forward to hosting you.”) is the most effective in establishing the perception of trustworthiness; while listing life motto (e.g. “Life is beautiful, so let’s enjoy it.”) is much less likely to make them seem trustworthy.
“You will find staying with me an enriching experience.”
There is something uncannily similar about this promise of hospitality to a similar promise that we’ve been hearing so much in the past year in the political discourse by Donald Trump — “I will be the greatest jobs president that God ever created.” You won’t find it in our paper, but we refer to the promises that are highly effective in conveying perceived trustworthiness as Trumping Promises.
So what’s next? Our research relies on manual assessment of every single profile, which isn’t scalable and doesn’t allow us to study the specific words, phrases, and linguistic styles that help establish trust. In our next project, we will discuss the possibility of automatic evaluation of profile text using techniques from natural language processing and machine learning. Drop us a line to hear more.
The research described in this post was accepted for publication after peer review and will be presented at CSCW 2017, in a paper titled “Self-Disclosure and Perceived Trustworthiness of Airbnb Host Profiles”. The paper received an honorable mention for Best Paper from the conference and will be available here along with the dataset.
See also coverage on this research on Cornell Chronicle.
This blog post received much valuable input from Mor Naaman, Kevin Simler, and Ali Alkhatib.
