The problem of video search has been waiting in the wings for a number of years now. True believers like Blinkx and Truveo (now part of AOL) were patiently experimenting with ways of indexing and tagging video assets long before the broadband penetration rates and usage curves supported it. During the last year, at YouTube, the dam has broken. People are starting to look for video in the same way they hunt for text.
Now we expect (or hope) that plugging “cat playing piano” into a query box will render that unctuous and ubiquitous clip. This is not just an issue for consumers searching for ephemera, either. As trade publications, libraries, event organizers, and others start filling their online coffers with video of events, news, how-to's, historic footage, and so on, the standard technology behind the search box reveals its limitations.
Try doing a search on Google Video for “last night’s David Letterman monologue” and you will get a weird assortment of Charlie Rose clips (because Rose’s company partnered with Google and has posted scores of videos to Google Video) and Stephen Colbert’s infamous White House Correspondents’ Dinner routine. Letterman himself doesn’t show up until the fourth or fifth result, and then it is in the form of his infamous interview with Madonna. Part of the problem with Google is that its video search is not really video search; it is aggregation. Widespread spidering of video assets throughout the web is only beginning to happen, and Google does not perform a web-wide search of this material so much as it directs people to its own Google Video and YouTube trove. Yahoo! and AOL do have web-side video search, but even some of the best cheerleaders for the future of this platform admit that video search is at the same point text search was in the mid ’90s, when we started moving from Yahoo! directories to Alta Vista’s deep indexing approach. Video content is poorly tagged (if at all) and ferreting the relevant material out from the other material on a page (especially in an age of Ajax and embedded Flash players) is a daunting task.
Finding the right way to tag and categorize video for more accurate search results is not just a technical issue. It has real economic implications for any company serving this media. Creating a video ecosystem for users that is as fluid and reliable as the current text-centric web is integral to creating and selling the ad inventory needed to underwrite the content. For instance, a hot topic among media buyers is how video search is a very different kind of activity from standard keyword text query. Most of the video search engineers talk about users “browsing” rather than “searching” video. We may use a keyword to locate ourselves in vaguely related clips but then we browse laterally across topics (financial news) or drill into specific providers (CNBC, Bloomberg, etc.). In other words, digital video search allows us to channel surf in a more vertical way. This different discovery mode creates enormous challenges for building results pages and building business models. Just look at the video results at companies like Pixsy.com, AOL Video, and Blinkx to see the range of experimentation that is going on around simple results presentation: static and video thumbnails, clip previews, and channel surfing opportunities, just to name a few.
The direct marketing approach that fueled text search just doesn’t work here. People don’t look for video necessarily because they are in the market for certain goods. Even if text ads in search results become the norm, the engines may have to become more like brand marketers and sell their inventory against their audience demographics and psychographics rather than keywords. And if we move to dynamically inserted pre-roll advertising, then how do publishers or aggregators target against that? When and if Google finally institutes some kind of pre-roll video advertising into YouTube, what exactly is the ad that you should properly serve into the piano-playing cat clips? Kitty litter? Yamaha? I have spoken to some agency executives who now discuss the need to identify and target the emotional states that bring a person to specific videos. How do you tag and index for moods?
The rapid acceleration of user and advertiser interest in online video exposes just how text-centric the internet is and always was. We are years away from solid search solutions for video, let alone business models to support them. Video is not just another content type being poured into the same digital containers as text. It promises to change the way we think about content delivery and distribution, ad formats, page design, and even our audiences.