Internal Search Optimization (iSEO): Creating Promoted Search ResultsImproving Search UX

Editor's Notes

1. Linda Caplinger I’ve been working in search-related positions for the last 15 years. Search engines Infoseek and Yahoo! Where I worked on the director and internal search programs. Consulting - most notable project was for the MicroSoft Bing Toolbox launch. Software Companies: McAfee, Symantec – both Anti-Virus companies, and then SAP’s Global Search Marketing Lead in NYC (mostly SEO, Internal Search and Paid) And now….NVIDIA where I do a bit of everything search. Types of search products that I’m worked on and are most familiar with: Google Search Appliance (GSA) or Enterprise search Google Custom Search Google Site Search
2. Internal Search Optimization, or iSEO as I like to call it. The process of optimizing your site’s internal search results To help users find information Via a query As fast as possible And To take them to the most relevant results. When you think about it, it’s sort of a combination of SEO, paid search, and UX skills all wrapped up in one and this is what I’m going to dive into today.
3. John Sherman just spoke to you about WHY you want to have search on your site and I’m going to try to tell you a few neat things you can do with it to add value. How many of you are using a “search” feature on your website? To reiterate and build on the WHY: Your navigation might be bad. Many companies organize their site navigation according to how their BUs are structured rather than how a user might look for content or products. For instance, can we all agree on what a small business is as compared to a medium sized business? Related to having a confusing nav structure: Your content or Information Architecture (IA) might ALSO be organized in a confusing manner. Sometimes the language that we use on our site is “inside out”. We use what I call “Marketing-ese” rather than the customer language. We like to say “notebook” when our customers are really querying for “laptops”. Sometimes the organic landing page just fails to deliver. The meta description says one thing but the page content is thin or not deep enough. Sometimes users just don’t trust your site and they head straight to the search box. These folks are “impatient”.
4. Content: What content do you want to be crawled and displayed? You might have multiple subdomains or websites you want the user to find. Or forums. With Google custom search, you can even point your engine to crawl a site you don’t even own. Results: Once the user gets to your landing page, then what? Can they sort or filter? Can they narrow results? And then there’s the entire User Experience: How do you make it easy for the customer to find what they want?
5. How many of you are already doing SEO and SEM? What is key here is that if you are already doing SEO and SEM, you have the skills you need to optimize your search results. Of course you might need a bit of technical help INSTALLING your search engine (and that’s what your IT team is for), but the rest is fairly straightforward.
6. You have a default way that your search results appear. Those can be customized for look, feel… to have images (or not) to show the URL (or not) Part of customizing the content you deliver to the searcher is what are called “promoted results”, or if you are using Google Search Appliance, “Key Matches”. This is an area that’s often overlooked by whoever controls onsite search. Google Search Appliance (GSA) – often called Enterprise Search – server or blade that crawls your content (internal or external site). Crawls YOUR results. Google Custom Search – a view of the “organic results” (with respect to search collections that you point it at). There are a lot of players in this space but we’ll just go into Google Search today. I’m going to focus on creating promoted search results for Google “ search products.
7. . Benefits of customizing your promoted search results So, we know a few reasons why users use the search box and that provides a natural starting point for why you want to use it. But there are also a few other reasons: Configuring your search results page lets you control what the user sees to a large degree. You can point visitors to sections of your site (or other sites!) that aren’t being crawled by your search engine. In creating custom links you can drop customers directly into the sales funnel. You can bubble up what you feel are the best or most relevant (or highest AOV) items. In the end, it just makes for happier site visitors.
8. The added benefit, is that you get immediate gratification. Results are quick. SEO takes time. Paid search is nearly immediate.
9. These are the 7 steps that I use for creating Promoted Results or KeyMatches: Get your onsite search data Scrub it Cluster or group together like-terms Pull out the low-hanging-fruit Write & optimize your ad copy (like you would for paid search) Integrate the most relevant (and hopefully the highest converting) URL to point the user at Add tracking parameters if you want (you can integrate Google URL UTM codes, Digital River CIDs, etc) And launch. One at a time now.
10. Collect data on the terms that people are searching for on your site. You’re probably using Google Analytics to track internal search. If you’re using Microsoft Fast or other tools, there will be data you can grab. If you want to iron out seasonal variations and anomalies, get lots of data. Maybe a year’s worth. Expect your data set to be HUGE. If you want to create campaigns for recent behaviour, maybe just grab a month. Export it all to a spreadsheet. What you need are the full query string AND query instances. At Symantec, we started with the top 50,000 queries. That’s a lot of data. Trust me, there’s going to be a great deal of duplication in your data set so you’ll need to scrub your data.
11. Cleaning data is the most fun and frustrating part of this exercise. Many internal search engines will ignore stop words. Also, thinking like paid search, the default search type in GSA and Google Custom Search is similar to a broad match type. Google calls the promoted query that you are going to build a KEYMATCH. If you enter the keymatch “Tegra” into your engine, it’ll return results for “Tegra processor”, “Tegra Tab” and any other query unless you tell it not to. So, a customized result for “Tegra” in your engine, might cover many many more queries than just “Tegra” alone. So clean out all of the stop words, characters, and so forth. Convert all your data to lowercase. tegra lowercase and TEGRA uppercase will pull up the same search result. Yes? Right. And we don’t want to waste time creating two search results for Tegra. Once clean, sort your data A through Z. Remove all of the duplicates. Give your data a sanity check. Did you remove all of the stopwords and random characters? Leave the misspellings in there. You want those. You might need to do this a few times to get to a list that you can work with. You are trying to scrub the data so that you are left with truly unique queries in the ed. It’s also interesting to see, much your data set was reduced by this process. This helps you understand how scalable this will be if you need to localize this to other regions or sites.
12. The next two steps sort of work together. They’re not necessarily done in a linear fashion. First: keep in mind that GSA and Google Custom Search default to broad match in that a single term is wildcarded and will cover a great number of queries that contain that term. Look at the Tegra example. I see 4 queries that all contain Tegra. I can create a “Tegra” keymatch and ad copy in GSA that will cover all query permutations that contain Tegra. Remember: you want to: Write once, and use over and over. Also: think “broad match” when you’re starting out.
13. Keep an eye out for queries that you know are associated to high AOVs, high CVR, and so forth. If you have a very high value query, something that’s important to your company, you can actually create what’s called an ExactMatch keymatch where you have one KeyMatch that shouldn’t pull in other permutations. (Doesn’t always work as it should, but for the most part it does). Also keep an eye out for queries that are more lead gen focused, or might cover different areas of the conversion funnel. You may find something happen that we call “bleed”, where one result bleeds into the result of another. For example: What if you have one keymatch entry for Tegra queries And another for “mobile processors” And then someone does a query for “Tegra Mobile Processors”? You might see two ads show up at the same time. You can solve this with the ExactMatch function.
14. Now this step is not a lot different than writing ad copy for Adwords. Make sure the query term is in the title and the description. For GSA, the query will show up bolded by default in the search results. Your developer may have turned this off. If they have, revert it back. Depending on your engine installation, your URL may or may not show up. If it’s long and convoluted, you can use a shortener or a vanity. Try to include a CTA. And of course, images are nearly always compelling and can boost your click through rates. Also worth mentioning for those of you in large organizations is that Google Search Engines will show 3 (or 5?) promoted search results. You may find, over time, that as people in your organization see how well this works particular business units (or Bus) may pressure you to let them own all of the promoted results. Not a great user experience. You may even find yourself in the position of having to create a policy for your company that given three results, one is for B2C, one for B2B, and one for support related results. Sometimes creating such a policy keeps competing interests from polluting your search results. Also, when the user’s query is ambiguous, and you don’t know precisely what they are looking for, try to show a good mixture of promoted search results. Spread the love over your different organizations.
15. When you’re thinking of where to point users, try to grab the most relevant landing page possible. Think of Google’s quality score when you’re creating your ad copy and destination URL. Not only does your ad copy need to contain the query term, your landing page needs to resonate with the user that, “Aha! This is it!” If it doesn’t resonate, they’ll bounce, they’ll search again, and eventually your promoted search results will not be seen as being trustworthy. For generic organic queries, Google likes to show you Wikipedia results. It just doesn’t know what you want. For a very generic query where you aren’t sure where to take the user, you might want to provide a mix of results, a “What is a..” page, a “glossary” page or other area that is high level. Akin to the previous slide, for some companies this might mean providing three results: one small business, one medium, one enterprise. Or some similar division. And again, if your program takes off and you are at a large company, you may need to create a process for “who owns which keymatch” (or do they all have to share)?
16. I like to add conversion tracking in my Promoted Search results so I can prove value. If you’re not directing users to pages where they’re going to convert, or if your web dev resources are scanty, you can always add in a hash and then a parameter so that you can easily parse the data out of your weblogs for page hits.
17. Once you’ve gone through the previous steps, you’re ready to push out your search results. You’ll want to test variations of ad copy, images, and destination URLS. Like paid search, continually optimize. You’ll also want to analyze which queries do NOT result in ANY search results for your user. In these cases you’ll probably want to create a customized landing page that reflects their query back to them, maybe they misspelled something, and offer to help them find what they’re looking for. Lastly, you’ll want to compare your list of KeyMatches in your search engine to the actual query logs and determine what queries are NOT addressed by your KeyMatch settings. I do this in Access. I had a developer friend create a script for me that I run in Access that takes the two lists of data (queries and volumes) versus (KeyMatches), scrubs the query data, and then churns out a list of which queries are still not KeyMatched. Essentially, it isolates the queries I still need to create KeyMatches for. It’s handy.
18. Very meager data. Not that exciting. Integrate with Google Analytics and view over on the GA site.
19. So, did it work? To reiterate what you want to track: Bounce rates Repeat queries Page consumption (may or may not be a good thing) Conversions (whatever that might be to you)
20. What are the KPIs you want to track for internal search?
21. Scour logs for searches with 0 results – what is the user looking for? Does a keymatch solve this? Ensure your empty results page includes suggestions for what to do next (sitemap, related searches, etc). Update this frequently, as your site will change over time. Continually look for searches with repeated search attempts (either the same query or something very similar) that happen in same session – create KeyMatch results Go over your top 200 or 300 results (80/20) and ensure results are tight and that the results are what you’d expect.
22. There are a few extra slides in the appendix for other things you can do to improve your site’s internal search results. Please feel free to email me directly for any questions.
23. SAP did a great job on meta tagging their content for pivot capabilities.