Contextualization, Localization & Searchable Attributes All Contribute to Changed Search Rankings
The question on everyone's mind these days is "What is going on with Etsy search?" On any given day, your guess might be as good as anyone else's, since it seems like they are changing elements weekly or even daily. Well, I am about to tell you all the messy details we know so far, but beware - this is not short. So grab a coffee (or a survival pack) and hunker down for a long read about Etsy's latest algorithm changes.
I also need to warn you: if you are reading this because your goal is to get all of your items on page 1 for generic searches, you are going to be disappointed. Not only because that will be impossible for everyone eventually (I can't say that in regards to your shop right now), but also because the rapid pace of changes right now means that even if you "crack the code" tonight, your methods could be obsolete tomorrow. Things are fluctuating that much! And that is what Etsy is aiming for. The same listings coming up on page 1 day after day appears to be a thing of the past.
[I will do my best to update this post with new developments as things change over the holiday shopping season. All updates will be marked as such.]
"Context Specific Search Ranking" - Narrowing Down Searcher Intent
While many of us could see that search results looked different, we didn't have confirmation of a key change until Etsy's 3rd Quarter Earnings conference call on November 6. CEO Josh Silverman stated the following:
In September, we introduced a major search improvement called the context specific search ranking, or CSR. CSR uses query and user-level information to rank results in real time. As a first step, we are leveraging transaction data to fuel our machine-learning technology to create more relevant search results. To give you a better sense of the difference, Slide 5 includes some before-and-after images to illustrate how this technology can deliver a better search experience for our buyers.
But what exactly does that techno-babble mean, you ask? The idea is actually pretty simple: Etsy wants to show searchers items that match their search intent as closely as possible. Take the example given, shown at the top of this blog post. When a buyer entered "laundry basket" into the search bar, they used to get a lot of planner stickers, as well as actual laundry baskets. Those sticker listings had "laundry basket" in their titles and tags, and are popular items, so the basic search rules ranked them well. But the results weren't what the buyer was really looking for; they likely would use the words "planner stickers" or something similar if that is what they truly wanted. Most buyers who entered "laundry basket" are probably looking for something to store their dirty clothes in; their "search intent" is actual laundry baskets. But the search algorithm can't understand these things the way a human would.
All of us who do much searching on Etsy can think of other good examples: if you search "dress", you don't want a pin with a woman wearing a dress, and "earrings" doesn't usually mean you are looking for supplies to make earrings. So fixing this problem sounds great, right? The second screenshot shows much better search results for these buyers. The problem is, how do you get the algorithm to understand this, for millions of different searches, and even for different languages? Enter the machine learning technology from Blackbird Technologies, which Etsy bought in 2016. Often called "artificial intelligence" (AI), this type of program absorbs large amounts of data, then makes connections between words, ideas and situations that might not be obvious to the program at first, or to humans at all. When used in search, an AI ideally "learns" from how well searchers respond to sets of results, so it can show better results the next time. In some ways, it is continually testing and refining what the best rankings should be. (Yes, you read that correctly - the AI will never be finished changing things. I am sorry to be the one to break the news to you.)
The AI started this current project using only some of the information available, namely the sales made on these items, likely made directly from these searches. If people who search "laundry basket" are more likely to buy laundry baskets than planning stickers or art work, then the AI learns that it will get better sales if it shows more laundry baskets for that search. The more sales patterns that are observed, the more distinctions that can be made, tailoring the rankings with other criteria as well. Silverman made it clear this was just the starting point for CSR, and that sales are up because of it.
So what other information can the AI use? The laundry basket example was straightforward to the average human being, but what if a word or phrase could mean two or more different things, depending on the context? A classic example is sports team names, and that is how I stumbled upon another apparent example of CSR several weeks before Etsy announced this program: Etsy is delivering a different set of results to European Union than to the rest of the world, and you can see distinct preferences in searches where the words may have more than one meaning. They are either testing or permanently allowing the ranking scores to develop differently in the EU from the previous algorithmic ranking. This is not regionalized search, which is controlled by the "ship to" filter. (I have more on search regionalization later in this blog post, but want to be clear: what I am showing you in the next few screenshots is completely different than that.) [UPDATE March 2018 - Etsy appears to have ended this practice for the moment, but seems to have rolled the ranking change into their normal regionalized search.]
The two screenshots below are of the search "blue jays", one with country settings (at the bottom of the page) on United States and one with United Kingdom. Note that I changed the "ship to" filter to the US on the UK version to avoid Etsy regionalizing the second search and showing me mostly UK items.
See how the search done with the UK setting shows more actual birds, while the US search shows mostly items from the Toronto Blue Jays professional baseball team? Sure, both searches show some of both, but the far more of the top-ranking items outside of the EU are baseball-related. The distinction continues as you scroll further down page 1 of each search. Note that in both cases, you see a mix of different countries' items, so this is not about showing the searcher things from their own country. In fact, you can change the countries and get the same results, as long as one country is in the EU and one isn't. So Canada and China will produce the same result (as long as the ships to filter is on the same country), as will France and Germany. (European countries that are not part of the European Union are grouped with the rest of the world.)
Not convinced? Another good example is the search "football". At the time I am writing this, the first few rows of the US search show American football results, while the UK search shows mostly what Americans call soccer. (Remember, by "US" & "UK" searches, I mean with the account set on those countries, but with the ship to filter set to a non-regionalized country.)
So, how are they doing this? I am not entirely sure what data they are sampling to create these distinctions. It could be through the sales data alone, as with the "laundry basket" example, or it could be that search is only using clicks, sales and favourites from the EU to decide the EU ranking, excluding non-EU information that shapes the quality score in the rest of the world. My best guess is the latter, but I could be wrong. Whichever answer is correct, this still appears to be a test or at best a limited rollout, as it doesn't make much sense to have China, Australia, Finland and the US in one group, and the entire EU in the other. I expect that if they are happy with the results, they will break this down much further, perhaps separating it out by each country at least in part. [UPDATE: this example has now changed; see discussion in my latest post]
So those are two existing examples of context specific search ranking. Personalization (results tailored to what you search for, click on, favourite and purchase) could be another, although at the moment I am seeing very little of this happening, compared to the other recent changes. Several people have been insisting in the Etsy forum that CSR is about personalization, but really, you can have CSR without considering an individual shopper's search and purchase history at all, as the above examples show. Since many buyers land on Etsy for the very first time every day, you cannot base search improvements on personalization alone, and Etsy knows that. So they aren't.
The End of Diversity - "Clumping" Makes a Comeback
When more than one item from a shop shows up on the same search page, Etsy forum posters have nicknamed it "clumping". Since Etsy made the relevancy search the default in 2011, they have always tried to limit the number of items from any one shop on the top pages of the search, known as the "diversity" rule. While this rule has had many variations over the years, since late 2014 only very small searches showed more than one item from each shop on the first few pages. The Etsy search help files explained diversity as follows:
Shop diversity - Buyers on Etsy are looking to explore a marketplace of unique items. In order to meet this expectation, Etsy’s search algorithm works to show results from a variety of shops, when available.
In August, sellers began noticing that the first pages of major searches had begun displaying multiple items from the same shop again. Etsy admin confirmed it was a test, but then in late October, the above definition disappeared from the help files. Staff have since told some sellers via email that sometimes "clumping" will show up now, although they have not made an official public statement.
So, it seems clumping is here to stay for the moment. I know that some of you are hoping it is still a test, but all signs point to the decision being made to keep it. In most cases, it isn't as damaging to non-clumped shops as it might appear, although I realize people feel it is unfair. Other recent changes are more likely to have had a larger effect on your rankings. I know some people hate this, but I can't see it completely tanking anyone's rankings, and I can't see Etsy getting rid of it entirely unless it starts making less money for Etsy.
Localized/Regionalized search - the European Union is now one region
As mentioned above, search localization or regionalization means that users with their "ship to" filters set to certain countries will see items that ship from those countries rank higher in large, generic searches. In smaller, more specific searches, it has less impact, and I am seeing that continue lately. So, aiming for more specific searches, as I always recommend anyway, is your best approach to combatting any loss of views from regionalized countries. You can test what people in regionalized countries see by changing your country settings and "ship to" filter to that country.
Within the past few months, Etsy regionalized the entire European Union. Yet again, Etsy made no official announcement about this change - they simply rewrote the help files to add the whole EU, replacing France, Germany, the UK and the Netherlands, which were previously explicitly mentioned. While I detected Etsy testing regionalization in Italy and Spain, among others, earlier this year, this update was a fairly large shift all at once, adding several new countries. (Note that European countries that have not joined the EU are not regionalized.)
Regionalization now works slightly different in the EU than it works in Canada and Australia, the only other two countries included. Let's use Germany as an example. A person in Germany will see German items rank fairly high in really large general searches, but will also see items from elsewhere in the EU near the top. Shops from Canada, the US, New Zealand etc. will start appearing lower down. In contrast, when my country settings are on Canada, I see Canadian items at the top, followed by a mix of everyone else. Someone in the US, which is not a regionalized country, will see a mix of all countries from the very first row.
Being in a regionalized country does not mean that buyers in other countries will not see your items. If their country is regionalized, shoppers will see more items from that country at the top of larger searches (not the smaller ones), but remember, most online searches are niche searches. If you optimize properly for many niche searches, you can be found in all countries you ship to, including those that are regionalized.
New Filters - Sales, Free Shipping, and Fast Processing
I wrote about the new search filter that limits results to items on sale back in August; it has now become permanent. Since then Etsy has added a filter to show all items that have shipping included (AKA free shipping), and have recently started testing two more filters which limit search results to items with very short processing times. It is impossible to say how many shoppers are using these options, as Etsy has not added the necessary reference values for sellers to track clicks in Google Analytics.
Historically, customers do not use filters very often on Etsy, so these may not be getting much of a workout at the moment. I am not seeing any of those filters on the app, but they do appear using mobile browsers (once you click on the filter button), as well as on desktop computers. During the site-promoted holiday sale, however, Etsy has moved the "on sale" filter all the way up to the top of the list, above the main category links, and has highlighted it in orange. (By the way, you can track clicks from Etsy's Cyber Week sales page in Google Analytics, as they include the reference value "cyber_category" - just search for it under Behaviour-->Site Content-->All Pages.)
Some people must be using them, or Etsy would not keep these filters. That means that you could be excluded from some searches at some times, if you have longer processing times or don't offer free shipping or Etsy discounts. While some sellers are worried that this means successful shops will need to change their business practices to show up in more searches, in some ways these options can help us avoid the buyers we didn't really want, like the ones who are last-minute shopping and might be angry if their order does not arrive in one business day as it does with Amazon Prime. As with any change to Etsy search, it will likely take a while for the effects to become more evident. If your sales have suddenly dropped in the past few months, do not assume this is the only cause, but presume it could be a factor.
Many Attributes Now Searchable; a few filters have been added
When Etsy introduced listing attributes back at the beginning of the year, we were told they would eventually be added to the search as both searchable terms (just like tags and titles) and as filters. In general, the Etsy-provided colour attributes have been searchable for several months now, along with the holiday and occasion options. We're still waiting for them to finish, but more and more of the other terms are (finally!) searchable. Some were added within the past few weeks while I have been writing this blog post, so be aware that more examples could be included or removed without any notice from Etsy.
I am not seeing any obvious logic as to which attributes can be found through Etsy search and which cannot. Some may still be added in the future, of course, and some might only be used as filters, eventually. Unfortunately, since Etsy has removed many of them from the overview section on the listing, I cannot even tell which attributes other sellers have used, so for the most part I can only test my own. For example, for the "Recipient" attribute on jewellery items, "Men", "Boys", "Girls" and "Women" are searchable terms, but "Unisex Adult" is not, either as a phrase or as individual words.
Some other jewellery attributes that are searchable at this time: ring size, earring location, chain style, gemstone, the new materials, and pendant shape. Examples of jewellery attributes that are not yet searchable (and may never be, I suppose): bracelet length and width, bracelet closure, earring closure, necklace length, and pendant size. Remember, however, that the AI could be using these to categorize items in the future, although I am not seeing any evidence of that now. They were also originally intended as search filters, which may still happen in the future.
Many items have different attribute options than jewellery, and even if I could test them all, the list would be far too long to include here. If you want to know whether an attribute element is included in Etsy search currently, test it, by searching very specifically for a listing where you know you haven't included that word or phrase in the title or tags. (Descriptions are still not searchable within Etsy.)
All of this means that some sellers now have more keywords that buyers can use to find them. While attribute words do not get much weight, they can be a crucial addition for those niche (a.k.a. long tail) search phrases you want to be found under. For example, I rarely add the word "women" to my jewellery titles and tags, but I am now getting a few more search hits that include it, since I have selected the recipient attributes where applicable. I cannot stress enough that this can be changing the search competition and rankings overall, especially for sellers who rarely added these types of words before and are now reaping the benefits of keyword diversification. I know a lot of sellers fee that attributes do not apply to their items (and in some cases, they are correct), but if your sales are dropping lately, you need to look at this as one possible cause.
Filters: most attributes have not yet been added as filters, which are found on the left of the search page on desktop computers, and through a button at the top of search results on mobile browsers and on the Etsy app. (Examples of the colour filters are included in the "blue jays" screenshots above.) Active filters include the colour attributes, and a sometimes lengthy list of options for various housewares from mugs to rugs. For many items listed under the "Home and Living" category, shoppers have the option of narrowing down their results by occasion and holiday, as well as a diverse set of choices including materials, patterns, size, shape and recipients. For example, items such as glassware have several different attribute filters visible in the search results, including "material". (Remember, that includes both vintage and handmade listings.) Objects such as rugs and decorative pillows have search filters for “design”, which involves a lengthy list of most common patterns, as well as some uncommon ones.
Strangely, though, you don't always need to have attributes applied to your listing to show up when shoppers narrow down the results through filters. Back when I could still see attributes in the listing overview area, I found many products in filtered searches that didn't use attributes, but showed up due to their titles and tags. (Lately, I am seeing less of that, however.) I am not sure what to think about this. Are sellers not using enough attributes, so that Etsy is forced to make guesses so that there are enough items in the final results? Or is this one of the ways they are training the AI? It's difficult to say. As one seller pointed out, some of the options available for buyers aren't actual listing attributes, so in at least some cases right now, you won't need to use attributes to show up in filtered searches. Don't assume that will continue, however.
Filters provide a way for customers to sort through huge volumes of items easily, without entering additional search terms. Set up properly and displayed well, they could be a partial solution for search bloat, especially in larger, more generic searches. I don't expect Etsy to give up on this idea, so sellers may need to make peace with it if they want to be seen through search. If attributes apply to your item and you didn't use them, you could be losing views and sales.
Other search News - tests, unproven and disproven theories, and what still works
There is only one important test I want to cover that wasn't already mentioned above, and then I want to deal with some of the smaller changes, and a few of the largest theories (or rumours) about how the algorithm is currently working. I promise to make this as brief as possible!
The colors are there to group options by 'theme' to make it easier for our buyers to visually choose the theme and then read the options. So for example: if you search for "ring"; the first few options in orange are occasion-specific, the next ones in yellow/light orange are related to the material/type.
At the moment, it seems to be visible in the United States only, but that is the biggest group of Etsy shoppers, so it could be having an impact in larger, more generic searches. (It doesn't appear in small searches.) These appear to be generated from popular keyword queries, but there may be other sources as well. I wouldn't necessarily try to tailor your items to fit the current options, but this test is definitely useful for keyword research! These are searches Etsy thinks people want to look at, which likely means they are not only popular, but also have a good conversion rate, i.e., sellers makes sales from these searches.
Etsy is now explaining some of its buyer tests on this page, but you will notice that they rarely mention algorithm tests there, at least not yet. Guided search isn't included, but the various sale and shipping filters are. Beware that you won't always learn enough by consulting that information.
Seller Theories: A lot of these are based on observation only, not testing, so many are either not true or may have a different cause than the one shop owners initially come up with. I will try to tackle the major ones I have read recently.
- Item price is not a direct factor in search. I took several of my listings, and either increased or decreased the price, and did not see any significant changes in ranking over several days. With some listings, I first lowered the price, then raised it above the original number several days later, just to check if there was a high or low pricing threshold, but I was unable to make my listings move based on price. Larger searches where the average price has dropped noticeably over this year may be due to the context specific search ranking (CSR) change discussed above, where the AI is using sales data to figure out searcher intent. Lower priced items may sell more often, skewing this part of the algorithm for the time being. That said, I am not seeing this in a lot of searches, so it may be item-specific.
- Using all 10 photos is not a direct factor in search at the moment. This rumour started almost as soon as Etsy gave us 10 listing photos, but it really picked up steam after a listing critique thread by Etsy staff in which most critiques mentioned using all 10 photos. I tested this by adding and even subtracting photos from some of my listings, and it had no effect on my search ranking. However, I do expect that using more than just a few photos leads to increased sales (although not increased search ranking), and I also suspect Etsy may want us to use more photos to help with any image recognition search technology they may be developing. Blackbird (the AI company Etsy bought last year) specialized in image recognition programs, and eBay has already added an image recognition program to its app. I would be surprised if Etsy wasn't at least thinking about it. The more photos you have, the more likely it is that an algorithm can accurately match your item to a potential customer's photo.
- New listings do seem to be appearing around page 5 of larger searches (for shops that are used to ranking well), and on the last page of smaller searches. I will admit to not having tested this at all, since I haven't had any time to list new in the past few months, and I never used to check where my new listings turned up even when I did, so I have no idea what my shop should expect. Most top sellers report that their items move up after a few days, so it is apparently not a permanent ranking. Still, if you were one of those people who expected a new product to go to page 1, at least briefly, you should probably stop basing your search strategy on that for the moment.
Search Changes: Besides the diversity rule, there are a few elements of the algorithm that have changed a bit, but they have not been removed entirely. These are both things that I expect to keep changing in the next few months, and may disappear in the long term.
- Having the exact phrase in both title and tags still improves your ranking in most situations, but this is in flux, especially in searches where very few items qualify. For example, at this time, I am the only seller with "Turquoise Blue Abalone Earrings" in a title, but out of 165 results for that search, I am well down page 1, behind some items that don't even have most of the words in the title. (You can tell how many other listings have the exact phrase by searching it in quotes.) It used to be that those earrings would be in the top row. However, I have tested removing exact phrases in larger searches that have more optimized items, and that absolutely does drop my search ranking every time. While there are probably some exceptions to the rule right now, I would be more worried about future changes in this area, as I think they are coming. [UPDATE (June 14): exact phrase matching is still working for large searches; I have more search updates here.]
- Recency still affects ranking, but it is more complicated and less effective than just a few months ago. If you use a renewal strategy to get better search standings, you may want to reevaluate your approach. I have noticed, however, that it still works quite well in categories, for items that already had a good quality score.
What you need to watch out for - results without all of the search terms
I am saving this for last because I don't know if it is a test or a change, but it is important enough for its own section: it is now possible to rank in a search without having one of the words in your title, tags or attributes. That was never possible before this year. There aren't many results like this so far, but I expect them to grow over time.
Please understand that I am not talking about some of the forum reports you may have seen recently, which mostly seem to involve words from attributes being included, or the periodic reports that involve stemming or other misunderstandings of how search works. (By stemming, I mean when someone searches a longer form of a word and the listing only has the root, or "stem" - that has worked for years. E.g., someone searches for "beaded" and gets listings with "bead".) It is also not caused by the word being in the description (I checked). Etsy does not currently use the description in search, and I haven't identified any examples of them testing it, either.
As far as I know, Etsy has not commented on this situation, so all I can do is speculate. It is likely part of the AI testing, or could even be using the old "similar items" algorithm, although the examples I have found did not look like the latter to me. Right now, I am not seeing this very often, and when I do, the listings are always at the back of the search. It's unlikely to have a large impact on anyone's views or sales at this low level, unlike Google's contextual search and Hummingbird algorithm, which can easily identify synonyms and other related elements of a page, and will sometimes show you a website that doesn't have the exact words as the number one result.
But if Etsy gets good results from this, expect the examples to increase; expect Etsy search to become more like Google's. If it works (and that is still a big "if" at the moment), it would be great for buyers. Right now if I search for "paua" and a seller only has "abalone shell" in the title and tags, I will not find their item. That's bad for a buyer, bad for the seller, and bad for Etsy, who might not make any money on the search. This is a real example, by the way. I have run into this exact issue many times while searching Etsy for paua and abalone destash (paua is a type of abalone), and have had to learn all of the alternative words people might use for it. I know that some shop owners will complain that this gives an unfair advantage to people who didn't do their homework on descriptive phrases, but heck, even after doing extensive keyword research for my creations, I sometimes run into valid search terms after the fact. So, this has the potential to help everyone, but it is important to remember, Etsy is most interested in what helps buyers first.
The items that include all of the searched words are still shown first, so using all of the synonyms and related phrases still does give your listing a large advantage. Plus, you can't show up for synonyms or related terms if you haven't used the right core terms in the first place. For the AI to understand context, it needs enough data to chew on, and titles like "paua bracelet, blue bracelet, silver bracelet, paua jewelry, blue jewelry, silver jewelry" simply don't provide it (that's only 5 different words!), and, conversely, a word salad of 140 characters with as many style keywords as possible will tend to overwhelm the algorithm, and dilute the power of the core phrases you do have. A title such as "paua bracelet, blue abalone shell jewelry, sterling silver chain and clasp" will get you a lot further, where Etsy is going.
Future of search: Voice & Image Search will need to work sooner rather than later
The above test (or change) is one of the main reasons I believe exact phrase matching is on the way out, or will at least be far less important in the future. The increasing use of voice search, and the advent of image recognition for search purposes, are two other reasons. According to a recent US study, "...one in 5 customers (19%) have made a voice purchase through Amazon Echo or another digital home assistant, and another third (33%) plan to do so in the next year" - and that doesn't even include all of the people searching on phones and tablets. Voice searches tend to be wordier, making it less likely that any one page will have the exact phrase the shopper entered. And, needless to say, image matching avoids language altogether. [UPDATE: see more on Etsy image recognition testing here.]
Having a search engine that only recognizes exact phrases cannot continue; Etsy knows this. And that is why getting people to start thinking about search differently is one of my biggest goals right now.
HOW THE HECK DO I GET FOUND NOW?
I know you, and I know what you are asking. I've seen it in the Etsy forum. "Search is all different now - SEO doesn't matter - how do I get to page 1?"
Here is what you need to know:
- SEO means optimizing for search, and the old rules still matter for Etsy search at the moment, even if they aren't the only rules any more, and even when they may get different weight than they used to. So yes, SEO matters, and keywords are still part of that.
- It's time for you to stop worrying about ranking on page 1 for generic terms, and start listening to me about aiming for more niche searches.
- More than ever, the other factors will weigh heavily in buyer decision making, and they involve how you market your business - photos, your points of difference from the competition, general branding, and understanding customer psychology are going to drive your clicks and sales from search. What worked last year may already not be good enough.
- Seriously, you need to believe #2, before it is too late.
Right now, you could choose spend a lot of time playing with your titles and tags and attributes, and sometimes you might get your items to rank better in large searches. I have tested this with some success. However, I am finding that the effect of most "tricks" fade quite quickly now, and the rapid rate of change means you cannot count on any rankings to continue at all. Plus, regionalization, contextual search and personalization all mean that most people may not see the same things as you do anyway. It's better to spend your time employing diverse, high-quality and relevant keywords on each listing, then stop worrying about search for a while. Leave listings a few months to see how they perform, then do more keyword research as necessary. Put your extra efforts into other ways of driving revenue.
As always, I strongly advise people not to rely on Etsy search as their only source of sales from Etsy, and not to rely on Etsy for your only source of business income. (But Etsy search is by far the easiest and cheapest way to get traffic on Etsy, so there isn't much point in having an Etsy shop unless you pay at least some attention to search.)
It appears that Etsy isn't going to announce most big search changes any more. They are apparently just going to change the help files without pointing it out, or release the details to investors, but either way, they aren't letting us in on these secrets first. That means it is up to sellers to keep up with these other sources of information about Etsy search. The next few months are going to be very busy, so I will try to keep upcoming posts short and sweet, instead of doing these huge multi-topic scroll-fests. (I hear a huge sigh of relief from everyone!) Please share this around so others do know what has happened up until now.
For future Etsy search news, you can follow my blog by email here, as well as my Twitter account, and my Google + page for SEO and ecommerce news, including Etsy. In addition, any of my SEO shop customers can sign up for my email list on all search changes; current buyers will receive a fully updated copy of my ebooks when it is complete next year. Finally, you can get more information on broader ecommerce and search trends such as voice search and image recognition by following this Etsy team thread. (I do all of the reading so you don't have to!)
Updated June 14, 2018