When I went to college, I studied Artificial Intelligence before there was any such thing as Google AdSense (or, in fact, Google). I know, it’s hard to believe.

So I spent a couple of years learning about robots, intelligent systems, evolution and natural language processing. Little robots will go to Mars, they told us. They’ll be Fast, Cheap and Out of Control. I thought it was strange at the time that no one seemed concerned with more immediate practical applications.

Shortly before I graduated, I met an ex-boyfriend of my cousin’s, who worked in marketing. “They’re just starting to use neural networks where I work now,” he told me. They used them to target direct mail advertising. So there it was, the first real life example of AI I’d personally encountered, and it was used for spam. I really should have known. By which I mean, I really should have known, and started Google. Then I’d be laughing.

Cybot by Macaruba

Adsense: not that kind of robot

Ten years later, after having the kids and so on, I went into freelance web development. Many of my former classmates do in fact do interesting and groundbreaking work involving artificial intelligence. They’ve invented guard robots, simulated the climate and ecosystems, programmed artificial lifeforms for videogames and invented Web 2.0 software applications. Wow, I wish I’d done that stuff. But when I look around on the common-or-garden Internet, I mostly see artificial intelligence being used to target contextual PPC (pay-per-click) advertising.

And that, I am often told, is the way to effortless millions. So I thought I’d check it out.

Having tried some experiments with affiliate marketing, pay per click advertising seemed like the next logical thing to try out. The basic setup is this: you sign up for Google AdSense or a similar contextual PPC advertising program, use their online advert design process to create snippets of Javascript code, and then paste this code unaltered into the appropriate part of your web page. Google’s Adsense robot visits, processes the other text on your web page and/or website and selects relevant adverts to display in the colours and layout you have chosen. Then the Adsense code loads when your web page does. You’re not allowed to click on your own adverts, but if someone else does you make a small amount of money (often very small). So millions of people click on the ads and you make an effortless fortune. Right? Well, so far I’ve found the system works apart from the last part.

The first thing I was interested in was how the adverts were chosen to be relevant. Would it depend more on the content of the page, or the website as a whole? I also didn’t want Google ads for my competitors appearing on my home page, although I did see the appeal of taking some of their money. So I tested out the ad selection process by setting up my first Google ads on a mix of obscure and slightly odd pages to see what they came up with.

Check out how the ads appear on:

• my privacy policy
  
  
  This is a screenshot, not an ad:
  
  The text before and after the adsense block mentions privacy several times, but the adverts that appear are for general web design and backup security. A quick search on Google for ‘privacy site:uk’ gives me no PPC ads next to the search results, so that would be why the ads weren’t so relevant to privacy: no one was advertising with the ‘privacy’ keyword in the UK.

• my ethical policy
  This is a screenshot, not an ad:
  

  Not bad – these contextual adverts do all relate to business ethics and professional codes of conduct. However, it is obviously not a simple case of pattern matching, linking words in the text to words in the adverts, as my page does not mention the word ‘whistleblower’, for example. Previously on this page I’ve seen ads for sex offender registers, which are really not mentioned in my ethical policy! These ads probably came from keywords in the policy about my web hosting not allowing porn, and using the film rating standards as a guideline.

  Surely if Google used some kind of vast semantic network to link related words together, wouldn’t it be very slow? If that’s how they identify keywords for a webpage, perhaps they would store the results rather than do it every time the page loads. Perhaps they combine two approaches, identifying and storing keywords for the website as a whole (hence the delay in seeing ads when you first put the Adsense code on a website) and then checking individual pages for keywords when they load. Most probably, this is all on record somewhere.

  In fact, when pay-per-click advertisers bid for keywords they write the adverts separately from that. So in theory, they could bid for a keyword and write a completely unrelated advert, although there probably wouldn’t be much point!

• my copyright / credits page
  This is a screenshot, not an ad:
  
  Again, these ads are nicely relevant – perhaps there’s some lucrative legal work in the copyright field.

• my legal document templates
  This is a screenshot, not a real ad:
  
  This one surprised me, because there’s plenty of PPC advertising on Google for legal document services. So why did my web design business site get a whopping graphic leaderboard ad about building websites in minutes? This is the last sort of thing I want my customers to be told when I charge by the hour! Obviously this possibility is a major drawback in putting adsense on a business site. To some extent, competitors ads can be blocked out, and it’s also possible to block image ads if you want to, but it’s still likely that some contextual ads will be for competing services.

  Reloading the page a few times gives me a text alternative to the banner ads, which explains the situation better:

  This is a screenshot, not an ad:
  
  Looking at my page about legal contract templates, and the ads about building websites with website templates, it’s obvious that the Google Adsense code has connected the word ‘template’ to the general web design theme of the site, and taken it out of context, giving a higher weighting to ‘template’ in determining the advertising keywords, rather than the ‘legal’ and ‘contract’ keywords.

  It would be interesting to know how the adsense code manages these weightings (and probably very lucrative too!). Thinking back to how I wrestled with neural networks in college, it would be so ironic if Google had just gone and made one of my website.

Next I tried:

• my ‘Under Construction’ page, with free photos of diggers.
  This page gets some nice ads for sustainable buildings, green architecture and design engineers, possible picking up on the ‘ethical policy’ link in the footer, and the ‘design’ and ‘beach’ keywords used in the site name.

  Surprisingly, this page turned out very well, and for a while it was a popular resource for free photos of diggers! It even got some Adsense clicks – but why? Well, for one reason, I think although the button to click for more photos doesn’t reload the adverts, perhaps it encourages people into interacting more with the page in general. So once they’ve clicked something a few times, they look for something else to click. There’s also the consideration that the rest of my website is unrelated to diggers, so the easiest place for visitors to follow their interest in construction is out through the adverts: interaction design (accidentally) in action!

Then I thought of some types of work I was too booked up for at the time:

• Content Management Systems, which gets general web design ads, and
• Database Websites, which gets some nice looking ads for business database software and professional database development tools.

While keeping an eye on the Google forums and looking at how the Google AdSense code is used on the net, I discovered several gimmicky pages along the lines of ‘Find Your Hobbit Name’ that seemed designed to keep people reloading and seeing new ads they might click on. Reloading the whole page over and over again seems a little dodgy. So I wrote this page, similar to the free photos page, where different parts of the page can be changed by clicking buttons. It took a while to be indexed, but it was popular in February and gave a little boost to the adsense this year:

• Random Valentine’s Day Poems

How romantic.

I even tried it on a ‘Lorem ipsum’ page, before realising that would go against AdSense policies by placing them on a page with ‘content primarily in an unsupported language’. For the record, they were mostly about beach resorts, because of the domain name, and temp agencies, probably because of the ‘tempor’ word fragment.

Here’s a sample of lorem ipsum, for anyone who hasn’t seen it before:

“Lorem ipsum dolor sit amet, consectetur adipisicing elit, sed do eiusmod tempor incididunt ut labore et dolore magna aliqua. Ut enim ad minim veniam, quis nostrud exercitation ullamco laboris nisi ut aliquip ex ea commodo consequat. Duis aute irure dolor in reprehenderit in voluptate velit esse cillum dolore eu fugiat nulla pariatur. Excepteur sint occaecat cupidatat non proident, sunt in culpa qui officia deserunt mollit anim id est laborum. “

I guess I can see their point.

Finally, I signed up for Gmail (or Google Mail as it’s called in the UK because someone else already had ‘Gmail’ there), and I discovered the Adsense code was reading my mail, and putting relevant ads around it. What’s more, my Google inbox had a Page Rank of 7 (Huh? Who’s linking to it??) I wondered if this was all a creepy invasion of privacy, and then realised that Google would forget what my emails said almost as soon as I would, and I decided not to mind.

Today I read a post by Stephen Wildstrom on Business Week’s TechBeat blog that really sums up Google Adsense:

“I came across a curious item about a man whose family had placed his ashes inside the case of an old Sun Microsystems SPARCstation. Sure enough, beneath the item was an ad for Shine On Brightly cremation urns, plus offers for discount cremation, a San Francisco Bay ash-scattering service, and a company that will place your cremated remains in an artificial reef off Miami…”

Retro Robot by Sasan

Robots in the shed. Yup.

With all those Suns and SPARCs and artificial things, it looks like it should be science fiction, but in fact it really isn’t. It’s an urn full of somebody’s ashes, for people who needed somewhere to put them. Futuristic technology met real life and created something unglamourously useful.

So after all the exciting sci-fi promises of artificial intelligence, it could be that Google Adsense is its most widespread application. It watches what you read and write, and it can make you pennies! It may be a bit disillusioning, but I guess the trick is to make those pennies on a big scale, and build the robots in your shed.