The Trouble With Ad Targeting -- and a Solution

By: Steve Outing As I reported a few weeks ago, one of the hot trends at news sites is contextual targeting -- putting ads in front of online readers based on their behavior (what type of content they tend to look at on a site) and/or demographic characteristics (collected by sites that require user registration).

But as good as that sounds, there's one aspect of contextual content targeting that's troublesome. Consider this possible scenario:

A site using contextual targeting places ads for a travel agency on all site pages that a particular demographic group sees. Sounds great until, alas, the user pulls up the story about terrorists trying to hit passenger planes with shoulder-fired missiles -- and sees the accompanying ad for a special deal on flights to Athens.

Oops!

The trouble with the latest whiz-bang automated ad targeting programs is that they place ads on content without really analyzing the content closely to see if there might be a reason not to serve an ad with a particular page. Because they are automatic, and typically not closely monitored by humans, ad-editorial conflicts can appear in embarrassing ways.

An Historic Problem

This ad conflict is not new, of course. Every newspaper copy and layout editor is taught to watch for the airline ad showing up on the same print page as the plane-crash story, or the ad for a soft drink appearing on the same page as the recent story about a tainted batch of the product being pulled off store shelves.

Whereas print newspapers tend to shy away from placing specific ads adjacent to specific stories, it's a whole new world on the Web. The new online mantra is to position contextually relevant ads on appropriate content pages, and/or on all pages that a specific user sees based on what Web sites know about that individual -- the theory being that such targeting means more payback to advertisers because their messages are put in front of a highly receptive audience. It's a money thing.

Given that inexorable trend, the publishing and advertising industries need to figure out how to monitor and halt potentially embarrassing ad-editorial conflicts.

Where do we look? To the smart folks at Google, for one.

Google Leads the Way

The publishing industry has been excited about Google's AdWords initiative, a program that places appropriate text ads from advertisers on Google search results. AdWords has been wildly successful, attracting more than 100,000 participating and paying advertisers. It's attractive to businesses because it puts ads on pages where people are seeking something specific -- so sales results have been impressive compared to more conventional forms of online advertising.

After the success of AdWords, Google expanded the concept to place these same text ads on contextually relevant content pages. With a new program called AdSense, it invited major Web sites and publishers to carry Google text ads (from that pool of 100,000-plus advertisers) and placed them on appropriate content pages on these partner sites, sharing the revenue. Later, AdSense was opened to smaller Web sites (which must get Google's approval; acceptance into the program isn't automatic).

[Editor's note: E&P Online participates in Google's AdSense program; the ads can be seen below the left-hand side of each page on the site, below the navigation menu.]

As the AdSense program evolved, Google began applying its considerable brainpower to the notion of tweaking contextual content placements. It has now developed a system that prevents Google AdWords from appearing alongside content that might be inappropriate. It's entirely algorithm-driven, with no direct human involvement -- except to tweak the algorithms when it's discovered that the system is fouling up.

Don't Buy an ATV!

Here's a real-life example of Google's ad-conflict filter in action. Recently, the Albuquerque (N.M.) Journal began testing AdSense ads on its Web sites, including placing them on a section with stories about the dangers of all-terrain vehicles (ATVs), in the left column. (The Google ads are no longer there.) Assistant Managing Editor Donn Friedman says that almost immediately, Google's AdSense program began inserting ads offering ATVs for sale. But an hour later, they were gone, replaced by generic "house ads."

What happened? Google Project Manager Susan Wojcicki explains that the AdSense algorithms analyzed the content of those pages and recognized that the content was negative in its coverage -- and determined after a short while that advertisers would not be served by having their messages displayed on those ABQJournal.com pages. You may find it remarkable that Google figured that out -- and acted on its determination -- based entirely on the logic contained in its AdSense algorithms. No human hands touched this transaction.

Wojcicki explains that "advertisers like us because we send them very qualified leads," but inappropriately matched ads and content are not good for business. Indeed, the search-industry rumor mill suggests that advertisers are less happy with contextual ads on content than they are with contextual ads on search results. (AdWords advertisers have the option of being excluded from content placements and sticking purely to appearing with search results.) Ergo, Google had to set out to improve contextual content ad placement.

Google officials wouldn't get too specific on how the AdSense algorithms work, but Wojcicki says the system looks for about 10 basic concepts as it analyzes a specific page -- then makes a determination about whether it's appropriate to serve contextual ads on the page. (The technology behind this is in large part from Applied Semantics, a company recently acquired by Google that developed paid-advertising link technology and specializes in natural language processing -- understanding concepts in blocks of text, not just identifying keywords to try to determine meaning.)

Some content issues are obvious: the plane crash news, the product recall, etc. all trigger the no-ads-served mode. (In what may seem odd to some Web publishers, if there are to be no ads served due to a bad-context situation, sometimes Google will serve up white space with the message, "This white space brought to you by this Web site and Google.")

Other situations are in the gray area, and for now, Wojcicki says, Google is being conservative. Obituaries, for instance, currently trigger no-ads-served. But that's an area, she says, that's up for reconsideration. An obit on a local newspaper Web site would be an appropriate place for ads from local florists, funeral homes, and greeting-card companies. Google also won't serve ads to adult or sex-related content.

Even the plane-crash type of ad conflict could be considered as being in that gray area. Perhaps ads for flight-insurance companies could be served to those pages, for example. For now, Googles opts for no ads served.

Faith in Machines

The general theory behind Google's approach to ad-context filtering is that it be almost entirely automated. Wojcicki explains that there are so many potential ad-content conflicts possible that it is impractical to have an army of editors constantly watching for them. Instead, the Google approach is to tweak the AdSense algorithms whenever a new bad-match arises.

Some conflicts, of course, are specific to individual publishers, so there must be an override mechanism in the process. For example, if a publisher has a contract with Budweiser that prohibits it from carrying ads from any other beer company, then the publisher has the option of making sure no beer ads appear via AdSense. If a publisher sees an ad served by AdSense, objects to it and wants it down, the publisher can call or e-mail Google representatives to get that accomplished. (Larger publishers that deal directly with Google on the AdSense program have a specific Google rep to contact; smaller AdSense publishers of course will get lesser priority service.)

Google competitor Overture, which also offers a contextual program, argues that Google's over-reliance on technology will prove to be unreliable. Quoted in a recent New York Times article, Overture Senior Vice President and General Manager Bill Demas said: "Technology will never do the job of a human in making sure this ad makes sense in this news story," so his company has a team of human editors who make sure Overture's system is appropriately inserting ads into relevant content. (Overture recently inked a deal with Knight Ridder, which means that the newspaper chain's Web sites will include contextuals ads -- on separate pages -- from both Google and Overture.)

Media Vendors Not There Yet

While Google has tackled the ad-content conflict head on, contextual-advertising vendors serving the media business have not yet developed similar features. And Google, for now, at least, does not have plans to license its ad filtering systems to other companies -- though that's a possibility in the future.

New York-based Tacoda Systems works with a number of major media companies, offering technology to match up advertising with the characteristics and behavior of people viewing media sites.

As President and CEO Dave Morgan explains, "we focus on the person, not the page." By that he means that his company's technology is capable of identifying characteristics and behavior of specific users, then serving up ads that are relevant to that individual. As to specific content of a page where ads are inserted, Tacoda doesn't know what's on that page because it doesn't currently analyze the content. It's up to the publisher to make sure that auto-loaded contextual ads aren't going to content pages where conflicts might occur.

Morgan suggests that for now, publishers think about tagging ad-inappropriate stories such that they'll be exempt from automatic contextual-ad placement. For example, tag the passenger-plane story as "terrorism" and not "travel," to avoid conflicts.

Tacoda likely will add Google-like filtering to its business line down the road, Morgan says, but it's also up to publishers to let companies like his know what's on a page.

A competitor to Tacoda, Washington-based Revenue Science (formerly known as digiMine), which is providing contextual ad-placement services for Dow Jones and CBS Marketwatch, is taking a similar approach. Senior Vice President and General Manager of media Omar Tawakol says his company, too, focuses on serving ads to the individual, not to the page as Google does. In a slight variation of Tacoda's model, Revenue Science "is completely focused on making advertising relevant by using technology to ascertain people's interests from their behavior," he says.

Tawakol says Revenue Science has begun to work with a hybrid model to address the ad-editorial conflict issue: "We will show an ad to the right person in only certain places. We already do a form of this today where a group of people can be targeted in only certain content categories. This goes a long way toward mitigating the risk of uncertain ad-to-context association. In addition, wherever a place is defined by a set of words (and not a category) we will also provide the capability to avoid certain 'problem' words and concepts. This however, is based on heuristics and it will be a secondary aspect of our offering."

Language 'Too Complex'

Ad-content conflicts have long been a problem for print media; now they're a bigger potential minefield for digital media. Will Google show the way with an algorithm-driven filtering system that watches for embarrassing conflicts and stops them automatically? Google's AdSense system is so new that the jury is still out on its effectiveness.

As Colorado-based search-industry analyst Chris Sherman suggests, it's a huge challenge to think that an entirely automated system can catch all the ad-content snafus that are bound to come up. "Human language is too complex for that," he says, and unanticipated problems are always going to pop up.

Nevertheless, Google has made a good start on tackling a problem that has vexed publishers for decades.

---

Other recent columns

Blogs Have a Place on News Web Sites, July 31
Attract the College Set With Design, Interaction, July 16
How the Web Can Restore Journalism's Credibility, June 25
Targeting Technology Should Boost Online Ads, June 11
Newspapers' Taste Standards Get Loose Online, May 28
Profiting From E-mail In an Ocean of Spam, May 14
Previous columns

---

Stop the Presses! by e-mail

To be notified by e-mail when a new "Stop the Presses!" column has been published, sign up here.