SMX Notes – Personalized Search: Fear or Not?
Despite the session’s title, the Q&A quickly devolved into less of a discussion of “Fear or Not” and more into “Why did you do this to us, Google?!” (Followed closely by the response, “You’re not normal. Er, a regular user.”)
Danny Sullivan, Editor-in-Chief, Search Engine Land
The basic definition of personalized search is results are reordered based on what’s deemed to meet your personal preferences. Pages may move up or down or even out of the top 10 (usually 1 page leaves the top 10).
- iGoogle (personal homepage) content
- Google bookmarks
- Search history (clicks)—a noisy signal to be used on a large scale, but it’s not noisy for just one person
- Web history (visits)
- Inflates your opinion of Google: I’m #1 for my name! My stuff is tops! I rule.
Google is pushing web history.
And then there’s Yahoo. Their closest competitor in personalization features is MyWeb, which is about two years old. You can save pages to MyWeb. Old features included adding notes and blocking sites (Wikipedia be gone!). Yahoo collected information and could have used it. But they don’t.
Ask’s MyStuff has the ability to save info in folders. (He then gave an example of how to save something.) It’s a fundamental, rudimentary take—let’s reshape our results on what you’re clicking on.
Summary: Google is the only major search engine that’s doing it. Yahoo and Ask are harvesting some information but haven’t put it to use yet. Google is likely to be even more aggressive with personalization over time. Their reasons: it reduces spam (potentially) and delivers better results (potentially).
Gord Hotchkiss, President and CEO, Enquiro
(Total side note: Gord was the first “famous person” I encountered at SMX, but I wasn’t brave enough to actually speak to him. I was too busy having a panic attack!)
His presentation’s on what SEO might look like with personalization world, and also what black hat might look like.
SEO pre-personalization revolved around keywords & algorithms (links, keywords, etc).
SEO personalization revolves around users: search history, web history, current tasks, social patterns. If personalization is in place, social search can be an offshoot of personalization. But it’s very difficult for a marketer to look at an individual users.
We think we’re gonna look at buckets of behavior that work around themes: themes around products or things. Long tail optimization becomes really interesting because it’s more personalized—and more important.
Universal results & personalization: personalization can drive a much more confident universal SERP. Gord loves that understanding user behavior will be vital to success. Knowing what people are looking for will be vital. User-centered development will take hold.
Black hat techniques in this area: your results depend on your past history. For optimizers, there is a green field in which they can play: emerging spaces that are new topics without web histories. Create “buzz sites” around new, budding content as an SEO tactic.
Moving up the funnel.
For SEOs, a lot of our optimization has been about moving up the conversion funnel. Personalization moves SEOs up the funnel. We’re more worried about developing stickiness now. Build out of content and functionality. Sites will become research bases. Content aggregations, comparison wizards, mashups will all be popular. Black hats: widgets and gadgets to push visits to site (not inherently black hat, but could be swung that way).
For every theme, we’ll see circles of importance. Handfuls of sites with specific content will emerge for each theme. These will be the first places where people will go when searching for things. These sites get the bulk of early funnel traffic. These will be inundated with offers of RSS content, etc. Black hat: scraping/aggregating content to look like an authority.
User intelligence will become really important. Click-stream-based user tools, engines will introduce tools on the paid side as well. Social bookmarking sites will be hot. Yahoo has del.icio.us; Google has . . . ? Personalization is the scalable way to do social search. There will be more use of profiles in SEO. Black hat: spyware/spam to track clicks. (Hijack?)
Michael Gray, Owner, Atlas Web Service
Hooray for personalized search. This is a really exciting time for unethical SEO tactics. Google guidelines say not to install special software to manipulate SERPs—now Google personalized search will do it for you. Tell your clients to sign in on all their computers and visit their sites while signed in: “Hey, we’re #2!”
There’s already a fear of uncertainty: different data centers and geotargeting already create SERPs with varying degrees of differences. Personalized SERPs add more uncertainty to this issue.
This forces your customers to become Google addicts, using all (and only) the Google services.
How is Google dumbing down the user? By teaching users they can use less sophisticated queries (“public library” instead of “boston public library”), they are actually teaching people to think less and trust Google more. People are then less likely to use other search engines.
How to fix some of the problems
- Stop hiding the fact that people are logged in
- Be clearer on the SERPs when a result is there because of personalized search and not a normal result. If they’re that much better then why not highlight them?
- Make it easier for people to turn off personalized search. Right now it’s so difficult.
- If you’re going to try to spam personalized search, social media is the best way to do it. That’s your best bet. They’ll bookmark or link to your site. If there are lots and lots of people bookmarking your sites, that’s probably a less noisy and more reliable signal that they’ll take into account.
Tim Mayer, VP of Product Management, Yahoo! Search
One of the big trends in search is understanding the user query better. There are several different ways of doing that. Personalization = understanding the user better. Also under this is query categorization and better understanding query intent.
The subtlety we’ve missed is in user connections.
Different techniques of search personalization
- Session-based. Looking at what a user is specifically using for. Disambiguate the query based on the clicks and query. The challenge is figuring out where sessions begin/end. When do you shift topics/themes?
- Interest-based personalization. Understand the interests of the user based on their own declared preferences or user behavior inside or outside of the search content. Challenge: users sometimes do searches outside their areas of interest/normal behavior (like client work!)
Impact of personalization on search results
Queries should get shorter. Current average is ~2.7 words. For example: library vs. boston library (quite spooky that he and Gray Wolf would think of almost identical examples).
More of the top ten should be relevant to the user assuming a strict intent is extrapolated from the query.
Impact on SEO
Since they’re doing a better job of matching the results that show up to queries, it behooves SEOs to create more relevant content. Give the search engine enough content per page to help it determine the topicality of that page.
Yahoo’s approach: social
Discovery: Yahoo’s users’ activity in social products (Flickr, Answers, del.icio.us upcoming) can help them discover content that is interesting to them and their community.
Recovery: social search applications can help you save, store and “refind” info that you have found and is important to you.
30-35% of searches are for opinion-based queries: best restaurant, cool lamps. If SEOs come in, these can yield borderline spam sometimes. (He shows an example of a top 10 result for “cool lamps” with, well, not cool lamps. Very normal lamps.)
Socially-influenced results: his friend tagged some cool lamps, he gets served with them. Her reputation is on the line with that link. There’s a social incentive for people to help and tag things appropriately.
Matt Cutts, Software Engineer, Google
(SMX’s bios list his job title as “Software Engineer Guru.” Matt says he only agreed to speak if he didn’t have to use PowerPoint.)
The inherent idea behind personalization is better results. This is powerful because different people think about different things in different ways. Jane‘s definition of “kiwi” is different than mine. (Here he gives a Monty Python example: remember the one scene where that one guy can see a knight waaaaay off in the distance riding toward him. He’s far away, still far away, still far away—aah! All the sudden he’s attacking you! Aah!)
Personalization should not be a surprise. It’s been coming for a long time (Danny mentioned how we’ve seen this predicted for years). If we know even a little bit of information because you’re willing to opt in, then we can give you better SERPs.
PageRank by itself takes links and says how important those links are and puts those votes in ranks and runs over and over again.
Everyone has different singular value compositions, etc. You get different PageRanks for pages for sports people and tech people and cat people. Then you can blend that. Personalization is a dense version of PageRank across the entire web.
Personally I think it’s handy. You can find old searches and find what clicks actually helped you.
Is personalization the death of SEO? No, nothing will be. There will always be a need for people to present themselves well and there’s always gonna be people who don’t have time to do it themselves. SEO won’t be the same, but it’s not such a huge change that you won’t recognize it.
For all those who think it’s too hard to turn off personalized search: add this parameter to any search &pws=0 (thanks to Ionut Alex Chitu for that one).
Personalizing news improved the click-through rate 40-50%. Greg Linden (of Findory) said that that seems too low. In his experiments, they were able to get 200% better click through. That’s cool. There’s obviously a lot of opportunity here.
There’s a problem of always being logged in: mom’s looking for braeburns, dad’s looking for iPods (ie types of apples)
Matt: we break it into sessions. The results would probably still be better with personalized search, even if they were muddied by multiple users.
Tim: Safe search. Filter on/off
Matt: Don’t argue against data. If people hand you data, don’t not use it. Worst case, they’ll just go back to the baseline if personalization isn’t any better or is totally muddled.
I’m gonna call Matt out: &pws=0 still collects data (ie in your web history)
Matt: You can delete, turn off completely or suspend your web history.
Tim: There’s another clear way to opt out and that’s just not to use Google.
Matt (taking the jab well): Or different profiles.
Click-through rates for paid search results: will their clicks influence their organic results?
Matt: Not now. Web history is trying to provide a complete picture. He thinks it doesn’t influence organic results.
We’ve got a lot of great designers at Yahoo, so why is the Yahoo “Save to My Web” button so ugly?
Tim: If you don’t like the my web one, you might like del.icio.us better.
Danny (teasing): Is that a hint? Are you phasing it out? Oh, sorry, Matt—I would have used PicasaWeb & Google Bookmarks but . . . I don’t.
Tim: They’re not as good.
Do you incorporate any kind of Google Coop subscriptions in personalized search?
Matt: That’s a good question [which of course really means “I have to think for a second” or “that’s a good question—for you to ask someone else”], honestly I’m not 100% sure. I can imagine that happening. I wouldn’t be surprised to see that over time. If it improves the quality of results, we want to do it.
Supplemental pages & personalized search: is it possible to use personalized search to find supplemental pages that are hard to find in main index?
Matt: You could, but it would be a little tricky because you’re trying to bootstrap. We call it “huffing your own signal” [does it kill brain cells?]. If a page never gets hit, it’s not as helpful.
Follow up: I ask because every time I get into an argument with a peer, I go to your blog to quote you, and a lot of the pages on your blog are supplemental.
Poll: How many are afraid of personalized search? A few. How many opted out? Many. Some were like “okay.”
Danny: What if Ask were doing it? Would it be icky if Ask were doing it?
We hear a lot about the importance of diversity in search results, listing pages from lots of different sites. Might personalized search hinder that?
Matt: The simplest example of that is the current 2 results from each site (called host/site crowding). I think we were the first major search engine to do this. Right after we brought it out, a VC emailed me raving about it. Without it, you’d have all kinds of crap from Geocities or Tripod. You can also think about genres. We need information, blogs, reviews—a good thing all together. It can be complimentary to personal results. Within this genre, what’s the best site for their interests? There’s still a wall for that. By and large, there’s opposition but they can work together.
del.icio.us has a diverse audience that’s actively tagging. Is it influencing SERPs?
Tim: We use signals from those pages to increase diversity. Rather than just getting signals from you, we’re getting signals from other users who like similar things.
Follow up: Yahoo owns a lot of thematic verticals. Are you going to use that data in SERPs?
Tim: There’s an opportunity for that, but it’s important to give the users what they want & the best results.
How deeply does Google try to guess at or identify the demographic of the user? (gender profiling)
Matt: In my experience, no. You don’t have to provide demographic information to register. Geographic information can make a big difference. Historically, Google hasn’t tried to collect or guess demographic data. I think I saw an MSN paper on trying to guess the age/gender of surfer.
Tim: To clarify that, since Google started as search, and Yahoo started as a media company, we collect different kinds of info. Hotjobs for example required that information to register. They may be useful to search.
Gord: This is marking a new generation for search on marking new signals on query-based intent. We don’t want the organic results and sponsored ads to be too far out of sync: expect the sponsored answer soon. There’s totally arbitrary host crowding. Personalization impacts familiar territory. But they’ll leverage this however they can to improve search. There’s a lot of room to develop there. It’ll fuel a lot of innovations behind the interface and within the interface itself. If you’re worried about opting in out, you’re an anomaly. We have to think about regular users.
Matt: About 10% of queries are misspelled. If you could guess grade level, age, etc. from those mistakes, if you can find out what kind of a searcher somebody is, you can return better SERPs.
Have you done any eyetracking studies relative to personalized versus nonpersonalized SERPs?
Gord: It’s funny you’ve mentioned that, I was just thinking that. It’s tough because personalization is so individualized, that it’s very tough to mock up. It would be fascinating. If you combine personalization with those results.
Is &pws=0 a one off parameter? What happened to “turn off personalized search”?
Matt: There’s a funny story. We’re trying to simplify a mental model. Very few people actually care. When we have parameters at Google they can be sticky or not. I think it’s not sticky. If you’re checking a bunch of keywords, I’d just log out.
I’m upset that there’s no transparency with personalized search. A second issue: I’m lazy user. Most people in search know spell checks work so well that you don’t care about spelling. I know the speller works.
Matt: We think overall it’s better. If you’re not signed in, it’s not personalized.
Gord: On a user basis, how many of you people think personalization will bring better SERPs? A few. How many excited to be able to target ads more effectively? Some. People won’t opt in to get better ads. People will opt in to better SERPs. When you look at opting in on a user basis, think on those lines. They’ll catch up like Marissa [Meyer of Google has] said.
Danny: They’ve long been saying, “We’re gonna personalize everything we can.”