Search advertising, specifically Google, Yahoo!, and Bing, is successful because it allows marketers to reach consumers who have expressed direct interest in their product. This extends to targeting options on the Google Display Network as well, since, by and large, GDN is one of the few major ad-serving platforms that have turnkey semantic targeting available.
For the last year-and-a-half or so, however, Google has identified two big problems with ad targeting options on GDN. For one, they’ve realized they’ve got a lot of ‘remnant’ inventory, which they had been selling for extremely low CPMs (think video gaming websites, myspace, dictionary.com, etc.). The solution here was remarketing, and audience-level buying, which allowed high-CPM bidders to serve ads on traditionally low-CPM placements.
The latter of these two, audience-level buying, also helped address the second problem, which was satisfying the needs of more traditional display advertisers. Think along the lines of a Groupon, OneKingsLane, or Netflix – companies that didn’t have ‘hand raisers’ like the traditional direct-response company. These companies are new mousetraps, and that’s big business, so Google answered with less granular, more scalable targeting options like the aforementioned audience buying, as well as topics and demographic ad buys. These media buying options also spoke to big-brand advertisers, who previously were burdened by the limited scale of GDN when it only had semantic targeting options.
So, to completely over-simplify, direct response companies love semantic targeting, and ‘new mousetrap’ and brand advertising companies prefer the scale of audience- and topic-level buying, especially when give the opportunity to layer on demographic info (something on which Google is working especially hard).
The problem lies when Google decides to blur the lines between two advertising approaches.
On the surface, it appears that Google has done a great job separating the targeting methods for these two type of advertisers, but over time, I’ve seen that, in fact, there are several instances when an advertiser thinks he/she is doing a semantic buy but is in fact relegated to a purchase not based on the content of the placement. I’ve found three main situations where this behavior occurs.
If you’ve ever bought a keywords like ‘Atlanta medical schools’ in the GDN and found that your placement report had domains and urls that only had content related to ‘Atlanta’ and nothing to do ‘schools,’ let alone ‘medical schools,’ you’ve been a victim of keyword parsing in GDN. This is when Google determines there isn’t enough volume in a particular ad group, and does the advertiser a ‘favor’ by deciding to ‘cut’ keywords at the token level and split them into multiple semantic categories. So, in the above example, you could be matched to pages about ‘Atlanta,’ ‘Medical,’ or ‘Schools.’ If you speak to any display specialist at Google, he/she will note this as a common occurrence, though I’ve never found a record of it any of Google’s literature.
In search, you’ll see this occur in a less explicit way with broad match keywords.
How to Avoid Keyword Parsing
In the search network, you can avoid keyword parsing by using more restrictive match types, such as modified broad match. Additionally, consider negative exact match usage. For example, if you have a 3-token keyword being matched to irrelevant 1- or 2-token queries, consider making the shorter version exact match negative.
The bigger issue is in GDN. To limit keyword parsing in GDN, avoid using geo-modified keywords (or other modifiers that aren’t relevant as stand-alone tokens). Also, create ad groups with a clearer semantic theme by adding in a few more keywords. Lastly, and most importantly, make use of negative topics, categories, and domains. Typically, a topic that could very easily be made negative by most direct response advertisers is the ‘games’ topic. However, avoid audience-level negatives, as excluding an audience like ‘games’ will exclude a lot of great potential customers – because, let’s face it, we’re probably all a part of the ‘games’ audience!
Automatic Targeting Option Extensions
When advertisers create keyword-targeted campaigns in GDN, I’d imagine it’s typically because those advertisers only want to serve ads on content pages related to their keyword’s semantic theme. Unfortunately, Google doesn’t quite agree. Here’s a note from one of my better Google reps:
All Google Display Network campaigns which use contextual targeting are
given the opportunity to appear on as wide a range of relevant pages in
the network as possible. In addition to the factors which we more commonly
use to match these ads to the page (e.g. page content, language, link
structure, page structure, URL), we take into account additional
contextual signals, such as the content of a page which a user recently
browsed. The aggregate of all of these signals, and the performance and
bid(s) of each of the ads, ultimately determine which ads appear to a user
for a given impression.
What this means is that Google can extend semantic targeting to ‘audience’-level targeting at will. I don’t understand this at all; if advertisers wanted to do audience-level buys, they surely would. The result of what I’ve coined as ‘audience-targeting extensions’ are placements on domains that have absolutely nothing to do with the keywords in the ad group! I’ve found this occur more commonly, for whatever reason, with image ad formats. Additionally, this is much more of a problem when you’re in a vertical with high CPM or CPC bids. If you’re in one of these verticals, this is something you should really pay attention to. I’ve seen up to 80% of spend coming as a result of an audience-level extension (I’ve seen this twice in fact, both times in the legal vertical. I also see this in EDU verticals to a lesser extent). The results of the campaigns were, predictably, extremely poor.
In search, automatic audience extension exists as ‘session-based’ ad serving. The story is the same here: the advertiser was automatically extended to show ads to people based on their prior behavior, not based on what their current interest has been expressed as.
How to Avoid Automatic Targeting Option Extensions
This one is tough in search, though it’s a pretty ‘small’ problem in most instances. Unfortunately, there is no real lever I’m aware of to control this, and I’ve been relegated to waiting for an ‘opt out’ session-based ad option from Google.
In GDN, the options are much the same as with keyword parsing. The main one, again, being topic-level exclusion.
Semantic reaches can also be coined as ‘broad match gone wild’ – essentially, it’s when a relevant keyword gets matched to an irrelevant query. In fact, semantic reaches were once such a problem that most advertisers avoided broad match completely. With the advent of modified broad match, however, semantic reaches aren’t in the spotlight nearly as much.
As Google’s search algo understands semantics better, another problem has arisen. Typically, advertisers bid very high on their brand keywords to ensure position 1.0. Additionally, they are highly concerned with capturing 100% of their brand traffic, and hence run broad match. In this scenario, both adCenter and AdWords will match brand terms to relevant non-brand queries, the result being incredibly high CPCs for queries that weren’t intended to be bid that high. I’ve shown this behavior to some of my clients who are new to PPC, and they are shocked! Google’s algo is really advanced in this area, so if you’re ‘Netflix,’ Google will likely match your brand to things like ‘movie rentals’ and other similar queries. If advertisers aren’t checking their search query report, this problem often goes unnoticed, because the true brand queries cover up the poor performance of the non-brand queries, and the campaign and keyword level CPA (or ROAS) appears just fine.
How to Avoid Semantic Reaches
Avoiding semantic reaches is relatively easy. For brand queries, I’d advise bidding high, but still capping them at reasonable max CPCs. So, say you’re paying $0.30 for a brand click (position 1). In that scenario, bidding $1 would provide the same advantages of bidding $100, so I’d choose the former. Additionally, migrating away from broad match to modified broad match is a very safe approach. In this scenario, you’d likely capture 100% of the traffic you intend to capture, without the risk of capturing some traffic you didn’t intend to capture.
I don’t think there’s much dispute for direct-response advertisers that GDN is the best available display network, in general, for creating on-click (as opposed to view-through) sales. However, as Google starts to cater more and more towards brand advertisers, my worry regards the attempts to blend the two approaches. As an advertiser, I’ve never found any benefit from Google extending my reach beyond what I have explicitly intended to target. Though, having the option to explicitly layer the options on top of each other makes perfect sense, which is an area in which Google has done well.
On the other side of the coin, I’ve asked (informally) several AdSense publishers what their thoughts have been about semantic extensions, remarketing, and audience buying. For the most part, they are in the dark about the changes Google has made of the last year or two, but they have noticed consistent decreases in their CPM (with the exception of game websites and other previously extremely low CPM sites), and ‘less relevant’ ad serving. Hence, putting two and two together, a lot of my clients have connected directly with publisher and cut out Google (and its 32% fee) altogether. Back in the early days, these direct buys were the main way to buy inventory. Google changed that, and the DSPs added to the fire. I wouldn’t go so far as to saying the tides are rolling back to those original 1-to-1 relationships, but I’m seeing more of these relationships occur, and I wonder when/if Google will take notice.
The answer to me is simple; keep semantic targeting based on semantics, and only serve ads in the manner advertisers explicitly intend to serve them.
Opinions expressed in the article are those of the guest author and not necessarily Certified Knowledge. If you would like to write for Certified Knowledge, please let us know.