Hi guys,

Sorry for the delay, but we put together a high-level FAQ addressing some questions raised in this and other threads.  Doing this FAQ- or article-style is sometimes more practical than addressing lots of individual variations of similar questions, but hopefully this sheds some light on the questions raised.  Part of this is a re-post from a thread at SS; but I've also added info specific to the topics in this thread.

Understanding Search
 
We often get asked how our search rankings work.  The first important thing to note is that there isn’t one singular search and discovery experience at Shutterstock.  In addition to the core search experience at Shutterstock.com, we have Palette, Instant, Spectrum, and People search; mobile apps like our iPad app; as well as different sort orders such as “New,” “Popular,” “Relevant,” and “Undiscovered.” We also have Bigstock and Offset. There are many ways for customers to get to your images, including direct visits from Google, and many ways for them to filter the collection for specific file types like vectors or video. We localize our results for specific countries as well - i.e., it makes sense for Japanese customers searching in Japanese to see the most appropriate results for their own language. 
 
Our search rankings are complex and based on a number of factors. Keywords, freshness, and popularity (downloads) all factor into search results.  We focus on providing customers with the best image that matches their needs and we don’t favor specific contributors or portfolios. Location can be a factor when the search query calls for it (for example, ‘streets of New York’), or when region-specific results are appropriate.   
 
We believe that if we always match the best possible media with a customer, over time this will lead to more satisfied customers, and in the end more downloads (and more royalties) for our contributors. That belief has delivered over 400 million paid image licenses to date. 
 
How is my media ranked in each sort order? 

The Shutterstock search engine uses a wide variety of factors to match a customer with the media he or she is looking for.  This engine is tuned to generate four types of search experiences, or sort orders: New, Popular, Relevant, and Undiscovered.  You can think of a sort order as a customized version of the core search experience.   

The algorithm that powers the Relevant sort order tries to find the best match between what a customer’s needs and the media that best serves those needs.  To do this we leverage a wide variety of factors, including how well the query matches the keywords, but also factors such as downloads or other data.  There is no one factor that dictates the position media will appear in the search results; rather, it is a combination of many different factors.

The New sort order is a subset of the Relevant sort order, but with an emphasis toward newer content.  Older content can and does appear in the New sort order, but will be ranked lower than relevant newer content. 

The Popular sort order is similar to Relevant, but it factors downloads much more strongly than the Relevant sort order.  However, highly relevant content with less downloads will still rank well than less relevant content with lots of downloads.

Finally, the Undiscovered sort order is similar to the Relevant sort order, filtered so as to only include results with zero downloads. 

Why is my media not showing up higher in the popular sort order even if I have more downloads?

The Popular sort order leverages downloads and other customer signals to boost popular results higher than they would appear in the Relevant sort order.  While downloads is a significant factor, it is not the only factor, and a very good relevance match will take precedence.  Also, media that is ranked well in one query will not necessarily be ranked well in another.  For example, a family at the dinner table might rank quite highly for the query “Family”, but not as high for the query “Family at the Beach”, even if it has more downloads than higher ranked photos in that query. 

Why is my photo with zero downloads not showing up in ‘undiscovered’? 

The Undiscovered sort order is similar to the Relevant sort order, filtered so that all results have zero downloads.  Depending on whether the results above have downloads, for each query media with zero downloads will appear in the same or better position as the ‘Relevant’ sort order.  If your media doesn’t appear in the Relevant sort order it won’t appear in the Undiscovered sort order either. Like all sort orders, the search result position will depend primarily on how well the query matches to the media. 

Sometimes there’s a delay in the time it takes for our algorithms to register a download, and media will briefly remain in the undiscovered sort order even after it has been downloaded. 

Image Gallery Stats

The Image Gallery Stats page as been in beta and can include “preview” or “comp” images provided to trusted companies like major advertising agencies for consideration for purchase.  We’ve discussed those images elsewhere in public forums, but those preview downloads are part of enterprise agreements where customers are paying up to hundreds of dollars for a license once they select an image for actual use or purchase (with royalties of up $120 or more).  There is no usage license granted until the buyer purchases the image.  Preview or “Comp” images for these types of enterprise buyers are very common in the stock industry. Shutterstock contributors earn much higher royalties for downloads when an actual purchase is made.

It’s important to note that your Earnings Summary is the primary place to find your paid downloads; the Image Gallery Stats page should not be used for that purpose.

Download Map

The Download map shows an assortment of recent customer downloads of your images, which could include preview images (see above) and redownloads.  The location on the map is an approximation, due to the size of the map and other considerations. Like gallery stats, the download map should be looked at as a source of interesting information, not the place to derive highly precise information about your downloads and earnings.

About Testing, and Making Changes Thoughtfully
 
Search is powered by a set of services that include machine-learning algorithms: a type of software that is able to improve over time. In a very simple sense, the goal of an algorithm is maximize downloads. To get the right media in front of a customer, the algorithm takes many factors into account, such as how well the keyword matches a search query, or the past success of this particular photo, music clip, or video.  For some search queries, customers are looking for newer content or content that has not been used before, and the algorithm will leverage factors such as the age of the media; for other queries, age is less important.  There are many, many other factors that the algorithm takes into account, and we are adding more every day.  One of the ways we improve our algorithm over time is to add new factors.  We are also frequently making small changes in how the algorithm weighs these factors.
 
In practice, some of the changes we make dont always work out for the best, so when we make a change, we need a way of knowing if it’s a good idea.  We achieve this through an experimentation process called A/B testing.  In an A/B test, a typically small amount of our customer base is put into an experiment in which we’ve changed the algorithm, and we’ll measure how well it performs relative to a control group.  There are often many such experiments going on at Shutterstock, and its probable that you’ve been in more than one if you are a regular Shutterstock user.  If a change results in more downloads or improvements in other key metrics, it will usually find its way into the production experience.  If not, we’ll move on and try something else.  You can think of the process a little like evolution – we are constantly getting slightly better through a process of trial and error in very small amounts. In general, we will only ship an algorithm change if it results in more downloads, and thus more income for our contributors. 


Best,

Scott
VP of Content
Shutterstock