There is a growing consensus in the digital advertising world that fraudulent impressions and non-viewable impressions are inseparable. 

The thinking goes something like this: neither had the opportunity to be seen by a human being, so when it comes to transacting on viewable impressions, fraudulent inventory should simply be treated as a subset of non-viewable inventory.  Non-viewable inventory, in turn, is a subset of wasted delivery—unsafe, low-quality, or any other content the advertiser wants to avoid.  Indeed, when it comes time to pay, media buyers would rather the publishers wrap all undesirable impressions into one and only charge them for what the advertiser asked for.    

While some players are already taking this approach, today’s complex ecosystem also demands patience and flexibility as the industry transitions to a new currency.  Of course, fraud and viewability overlap, just as any indicators of traffic quality.  But they are not inseparable: 

  • A fraudulent impression is an ad impression that a human never could have seen, even if circumstances had been different.  In an ideal world, tolerance for fraudulent impressions should be zero, but disagreement on definitions, different measurement methodologies, and moving targets means that demanding 0% fraud is like chasing shadows. 
  • A non-viewable impression, on the other hand, is an ad impression that the user never saw, but could have seen if site construction (e.g. responsive design?) and/or user behavior (e.g. did they stay on the page long enough?) had been different.  Expectations of 100% viewability across the board are also presently unreasonable, which is why the IAB split the difference and recently proposed a 70% target. 
  • A sub-standard impression, such as one to a page with unsafe content, is an ad impression that a human may have seen, but that the marketer would rather they hadn’t.  Though the industry is getting better at pre-bid targeting, ad blocking, and verification, buyers should be more lax on these subjective classifications than fraud or viewability when holding publishers accountable for flawless delivery.

The nuances behind each of these negative impression types are important to understand, especially when players with budgets and bottom lines start playing the blame game: ultimately, identifying responsibility for all the different ways a campaign fell short of expectations will help everyone determine who picks up the tab, when make-goods are necessary and for how much, and who needs to take action to remedy the situation.  To begin this process, it might help to start with a few questions:

  • Who or what generated the bad impression?
  • On what kind of site was the bad impression generated?
  • How was the site constructed?

Who or what generated the impression?

A fraudulent ad impression can either be generated by a bot or by a human.

Bots are sophisticated programs, usually hosted on unsuspecting users’ computers, which can perform various activities on the internet. They can do almost anything a human can:  they can click on links, generate web page traffic, and can even be segmented and targeted like any other user.   Groups of bots hosted on many users’ computers are called botnets.  Most of the time, this traffic is driven to fake websites in order to generate ad impressions (more on ghost sites below). However, sometimes bot traffic is driven to legitimate sites as well.  The main reasons why this might happen are two-fold:

  • One way for a bot to avoid detection is to mimic the behavior of a human user by visiting many legitimate sites, in addition to the website(s) they are targeting.
  • Certain publishers, hoping to better monetize their inventory, occasionally purchase traffic from third parties or affiliate networks.  And occasionally this traffic ends up coming from bots (regardless of whether the publisher, the reseller, or both actually intend for any of this augmented traffic to come from bots, is hard to know).    

While it is clear that any bot-generated impression is fraudulent, proving beyond doubt that it was a bot, let alone determining who is to blame, can be a challenging process.

There are also many instances in which a human-generated impression might also be fraudulent.  One example is ad stacking, the practice of stacking multiple ads on top of one another, with only the top ad visible to the viewer.  In these cases, although only one ad is visible, the impression counts for each served ad—even the hidden ads underneath the “stack.”  While most of the time ad stacking is purposeful, occasionally it can also be the result of improper implementation (such as multiple SDKs in a mobile app that don’t communicate properly and each try to deliver an ad to the same spot).  Another example is pixel stuffing, where a 1×1 pixel is placed on one site that loads up an entirely different site.  Impressions to this site, though technically generated by human activity, are fraudulent because no user can possibly have seen the content of the website “stuffed” into a 1×1 pixel.  While this method of fraud can be used to simulate false ad impressions, it’s also often used in affiliate marketing scams, where the hidden site cookies the visitor. The hidden site then gets to share the credit for any conversion or purchase on the site the viewer is actually visiting.  

Since these human-generated instances of fraud all occur at the site-level, it is the publisher’s full responsibility to rectify.

On what kind of site was the impression generated?

Another way to think about fraud is to consider the kind of site where the impression was generated.  As discussed above, legitimate sites almost always see some portion of bot traffic.  But some sites, called ghost sites, are built specifically to defraud advertisers: after creating them and filling them with dummy content, their owners make them available through ad networks or exchanges that participate in real-time bidding.  Then, they hire botnets to go to the site, which in turn generates ad impressions that enter the auction environment, which are then purchased by advertisers.  This is the most widely reported type of fraud, but it’s also the most clear-cut: the owners of such sites bear full responsibility in this case.

How was the site constructed?

Viewability, by definition, assesses whether 50% of an ad’s pixels were in view for at least 1 second (display) or 2 seconds (video).  When an ad is reported as not viewable, however, it could be for one of two reasons:

  1. The impression was fraudulent
  2. The user experience on the site was not conducive to ad viewability

If the impression was fraudulent, it would be worthwhile to understand what kind of fraud it was—ghost sites, bot traffic, ad stacking, pixel stuffing, or any other type not explored here.  Whether the fraudulent impression was technically counted as viewable or not viewable, the ramifications of fraud are very different than those of non-fraudulent, non-viewable impressions.  In this case, viewability has mostly to do with how the site is constructed.  To boost viewability across their inventory, publishers can create more engaging content and experiment with responsive page layouts, better ad placement, or ad formats more conducive to being viewed.  However, non-viewable impressions also have something to do with the inherently capricious behavior of users.  No matter what publishers do, there will always be some level of non-viewability: users who scroll too fast, users who open and close multiple tabs, etc.  It’s very important not to conflate the difference between fraudulent, non-viewable impressions and non-fraudulent, non-viewable impressions because doing so can lead to big misunderstandings about who is to blame or how to remediate.  

So how do we as an industry come together and find an approach that works for everyone?  The answer is flexibility.  Technology vendors have a duty to provide as much information as possible, in as transparent and digestible a way as possible.  In turn, users of that technology (both buyers and sellers), can collaborate to bundle that data in whatever form most suitable for that transaction, and establish an acceptable margin of error across different parameters.  Some parties will agree on a catch-all approach to ad blocking or make-goods.  Others will only sign off on a more piecemeal, shared-responsibility approach.  Either way, with wildly different expectations, ever-changing definitions, and new technological challenges cropping up all the time, one thing is clear: there’s no one size that fits all.

Zach Schapira