One thing you learn very quickly about being an Airbnb host is that you have to try and understand their search algorithm quite deeply to gain a solid edge.

To help you navigate this unique terrain of Airbnb, we've already narrowed down a list of major rankings factors to consider when trying to improve your search rankings.

However, beyond what we've already been experimenting and sharing on this blog, we still haven't yet dived into a one other crucial topic deeply enough and that's "personalization!"

Airbnb personalization is the adjustment of search results based on each guests' unique profiles. This includes what they clicked on, what they booked, and more.

For today's post, I think it's best if we split it up into two major sections:

  • Section 1 - We'll go over a small experiment we ran and the 5 major takeaways
  • Section 2 - We'll pull 9 insights directly from the Airbnb engineering team.

Before we go into the blog post, I've put together an accompanying video for those of you that prefer to watch:

I'm honestly really excited about this topic because RankBreeze looks at so much data that we want the results to be as realistic and accurate to what the common guest is experiencing during their search.

Which means...

We now have to get our hands dirty and get down with some testing now!

Our Experiment With Airbnb Personalization

Airbnb is in 191 countries and 81,000 cities around the globe - take a moment to realize how many travel experiences they truly impact.

It's huge and at that scale, there's no way their teams can visit every single place and know everything about them.

So what do they do? Simple, they use data.

From what we understand so far, every click a guest makes on the platform before booking will impact what they see in their search results.

To see how big of a change personalization can make, we decided to run a test.

Here are the details:

We used 4 different profiles, each replicating the same search... then restarted the search again to see what kind of results would show up and what was different.

  • Profile #1: Incognito & not logged in. This is essentially our baseline.
  • Profile #2: Logged in, but with no history of bookings.
  • Profile #3: 2 historical bookings, with 2 reviews (one positive & one negative).
  • Profile #4: 5 historical bookings, with 1 additional booking as a guest to another party.

The next detail we need to hammer out is the intent of the search. 

Ask yourself, "What is this guest looking for?"

This is the exact question Airbnb is asking and it should be the same question you ask yourself when performing any Airbnb SEO work.

In our scenario, the search intent of all four profiles was to "Find a place for 4 people in New Orleans, close to Jackson Square."

With that search intent in mind, we ran a regular search, clicked on 4 different listings on the second page, hung out on one of the listings for a longer time period and noted the effect it had on search results.

After doing this with all 4 profiles, clicking on 4 different listings each time, and increasing the length of time we hung out on one of the listings (dwell time), these were the new ranking changes afterward (i.e. how far did they move up or down):

Airbnb Personalization Test Results

In order to keep things simple, we decided to increase the dwell time of the listing which corresponded to the profile number.

For example, Listing 1 would get the most dwell time in Profile 1 and in Profile 2, we would increase the dwell time for Listing 2 (as indicated by the yellow boxes.)

With that in mind, a quick analysis of the table tells us is that most of the time when we hung out on a listing for longer it improved rankings.

What Did We Learn From This Test?

Here are the lessons we learned:

  • 1
    The longer the amount of time someone spends on your listing, the higher it appears in the rankings when they restart their search again.
  • 2
    Clicks don't matter as much when a profile has more historical bookings (at least in this experiment). We found it difficult to affect search rankings by clicking around in profiles that had more historical bookings.
  • 3
    Rankings on the first page were very similar across all profiles. However, once you got to the second page, there was more variance. This indicates that higher ranked listings are more stable within the first page.
  • 4
    In general, personalization lowers the rankings of most listings.
  • 5
    To keep the experiment more legitimate, we made sure to select listings with poor cover photos. However, no matter how many times we selected these listings, Airbnb refused to move these listings up the ranks. This could be because our own experiment was quite small in comparison to the number of people running legitimate searches, which could have affected our real-time results. Regardless, it still reinforces the fact that high quality, professional photos are an important side-factor.

What Airbnb Has To Say About Personalization

We've always been curious how personalization affects the success of a listing and recently we were able to gain more insight by examining a research paper by Airbnb's legendary data science and engineering team.

Here are our biggest takeaways from that research, translated into layman terms for the rest of us!

#1 Airbnb Weighs "Search Impressions" Highly

​Image Source

Search Impressions are the amount of views you get from the search results pages. (I.E. how many times someone sees your listing when it's loaded into results page.)

The simple reality of vacation rentals is that each listing can only be booked a maximum of 365 days a year.

This means that when a listing is booked, it stops collecting search data for those dates.

In turn, that makes it harder for Airbnb to make decisions because the data is sparse and essentially forces them to make assumptions based on few examples.

This explains why personalization is not as big of a factor as we originally thought, although it will grow as the company gets more data.

Airbnb recently began splitting up their performance metrics by "search impressions" and "listing views" so they're starting to give hosts a more refined look beyond just "views".

The consensus from the community about these new metrics has been quite positive but it also doesn't tell the whole story.

Actionable insights are still needed and we'll learn about this below.

#2 The Type of Impression Matters Even More

Did people just click on your listing and look at it or did they actually book it during their search?

When the Airbnb team started down this endeavour, they realized starting simple would be the best way to go about it.

Taking a page out of the Amazon A9 search algorithm, they started tallying up the results and giving "impressions with bookings" as high a rank as possible.

In simpler terms, whenever a listing is booked, that particular search results page is given a '+1', while the rest of the pages are given a '0'.

Sidenote: The example of 1 and 0 is the easiest way to characterize a positive and negative impression. Going a bit deeper, you can also separate positive impressions by "clicks" and "bookings". Whereas negative impressions can be categorized as "seen" or "unseen". An example of an unseen negative impression would be a page that never loaded.

#3 Higher Weighted Listings Are Mixed Up More Frequently

Every click a user makes, the more personalization can be crunched for that particular profile and search results.

Airbnb User Search Session

Airbnb will prioritize mixing up the top search results (within the first page) and try to "improve" those first, rather than mixing up the lower ranked listings. 

This means the top-ranked listings have the biggest opportunities to participate in personalization, most likely because Airbnb is more confident in those listings being able to satisfy the majority of guests' needs in that particular city. 

#4 Your "Historical Booking Count" Is Processed In Real Time

You know when the stars align and you magically get one booking after another?

That's personalization at work.

The more bookings you get, the higher your listing is weighted and this is part of what I call the "Circle of success". 

Listings that have more bookings will rank higher but since this is processed in real time, the velocity of those bookings also matters.

(This is essentially #4 in our Airbnb ranking factors list)​

#5 "Smart Pricing" Is Also Processed In Real Time

Whether or not you like Airbnb's Smart Pricing tool, I'm telling you now that if you use it, your data will be fed into their personalization algorithm faster.

But of course, the fact that it's not very good doesn't help it win any points in most host's hearts.

For the most part, their smart pricing tool prioritizes conversions across the board of all their hosts vs maximizing revenue of a specific host using it.

#6 Past Booking Helps Determine Initial Search Intent

Each time you click on a listing, the Airbnb algorithm asks itself "How does this current listing compare to your previous viewed listings"?

Knowing that, it'll decide how much of a re-arrangement of the listings they'll need to make.

From what we could tell,  Airbnb has predetermined what you might book when you have some historical bookings under your profile.

Therefore, the listings won't move around too much as Airbnb already know what listings you're more inclined to book if you booked for that number of same number of guests (at least at the beginning of your search when you're not clicking around as much.)

Also, the profile with 6 historical bookings showed similar top listings with those in the other profiles but as we went further down the pages, we realized more personalization was taking effect.

In the past, this profile booked smaller units and didn't book any large premium places, which is why we noticed smaller, cozier units showing up at the very bottom of the first page.

How historical bookings affect Airbnb rankings

#7 "Longer Views" Correlate With Bookings (But It's Still Not That Simple)

A longer view is also referred to as "dwell time" or "time on page".  It means the amount of time someone spends on your listing.

Longer views have been accounted for in Airbnb's algorithm updates but it did not make a difference in improving conversions on the platform.

This metric of dwell time, however, continues to be a point of research because after manual inspections of listings with long views, they found several factors which could have resulted in an increased view time. 

For example, some listings were high-end with high prices or had longer descriptions, while others had some humour to them, along with other various factors.

So although longer views didn't directly improve bookings for Airbnb, it's still being utilized within the personalization algo.

#8 Occupancy Rate is Seen as a Signal of High Quality

Airbnb uses occupancy rates in their personalization algorithm.

Occupancy rates to them is calculated as a fraction of available days - in which a listing is occupied and the assumption is that higher occupancy rates equals a more attractive listing.

This occupancy rate, along with the "Average length of stay", gives Airbnb an idea of how well you compare to the rest of the listings in your area and again, might be a reason why you get bookings one after the other since that metric is plugged in real time.

#9 Listings In Neighbourhoods With More Bookings Are Given a Higher Weight

Naturally, some areas of a city are more popular than others.

It can be assumed that the majority of travelers would prefer to stay in areas where plenty of historical bookings have occurred with a high velocity rate (aka a lot of bookings happening at a faster rate). 

With that in mind and a handful of data, Airbnb is choosing to show listings closer to the "hot" areas first because it's already proven to be a preferred choice amongst most travelers.

A heat map of the preferred listings when searching "San Francisco"  (Source here)

It's also not surprising that Airbnb would arrange their overall top-level rankings in this order.

When we conducted our own research, we came to the same conclusions about the importance of location.

Yes, we've all heard that in real estate it's all about "location, location, location" but it appears that if you're if you're using Airbnb's platform, it's important to consider too.

For now, that's all the info we have for you. If you're like more on the way or want to check out more of our work, please subscribe to our blog below.

Otherwise, if you have any  questions, let us know in the comments or see the good work we're doing at Rankbreeze.com



>