Watch Your Match Types

Google AdWords Broad Match has recently got a whole lot broader. While this may be good for Google’s top line, it isn’t as good for users or advertisers.

In Summer of 2005 Google changed the definition of “broad match” to include “expanded match” technology. We’ve never been huge fans of “expanded match”, but it appears to us that over the last month or two the definition of expanded match has been blown out even wider.

Prior to summer of ‘05 “broad match” essentially meant that Google could serve an ad if the user typed in all the words in the ad phrase in any order and with any other words included. So, if I had an ad running for “acme widgets” my ad would be served if the user typed in “yellow widgets by Acme”, or “acme wodget-style widgets”, or “i hate acme widgets”, or “widgets acme” etc.

When you look through user search logs you realize pretty quickly that even advertisers who are as obsessive-compulsive about term creation as we are are can’t anticipate all the possible permutations of word order and add-ons that might come along with a standard search term. The long tail is very long — a huge fraction of searches are unique. In the logs you find entire questions like “Where’s the best place to buy an Acme widget for my daughter’s birthday”. Broad match helped capture these long-tail query crumbs.

We did some research a couple years ago (prior to expanded match) comparing actual user searches to the ads that they clicked on. This internal study clearly showed that on broad matched ads, the conversion rates of those who actually typed in the exact ad phrase were considerably higher than the rates for those who came in on broader matching algorithms.

However, something like 80% of the traffic on these broad matched ads was from exact search matches, so the overall quality of the traffic was a bit worse, but with a wide array of well-considered negative associations we could make most terms work efficiently on broad match and get the benefit of wider reach.

When expanded match came out we noticed a number of annoying “benefits” of the new system.

One was that if you searched for one retailer’s brand name, often Google would serve ads for their competitors as well even though the competitor wasn’t advertising on that trademark name. In their wisdom, Google had ferreted out that people looking for Acme, might also want to see Apex’s collection of widgets. There were some helpful elements in expanded match in that capturing all the pluralization combinations, word tenses, etc became less important. But for diligent agencies this wasn’t very helpful; we’re going to do all those combinations anyway for the sake of the ad copy.

We restricted our use of broad match because of this, but still used in appropriate circumstances. With the changes over the last few months, we’re finding many fewer situations where broad match makes sense at all.

Over the last few months, we’ve noticed that the broad “matches” Google makes have become extreme. As just one example of many, Google will at times match specific SKU searches against generic category ads, or national retail store ads, even when we’re running a much more specific SKU ad (eg matched to the specific phrase, specific landing page, and specific copy) with a high max CPC.

I think Google’s revenue maximizing algorithms have discovered that the combination of “extended match” and quality-score based auctions means they can pretty much serve any ad you have on any search related to your category. The result? Higher cpcs, less traffic because of the less targeted copy, and lower conversion rates because the landing pages are wrong: a perfect storm for advertisers, degraded results for users, but more short-term revenue for Google.

Smart advertisers need the breadth of the “old-style” broad match (without extended match) to cover the permutations of search phrases. Without this, advertisers are forced to explode their term lists for word ordering, which isn’t good for agencies or for Google. N words can be arranged N! ways. Permuting 2 word phrases doubles the number of two word phrases. Permuting 3 word phrases increases the size of the phrase list six-fold. And permuting 4 word phrases generates a 24-fold increase.

Google needs to maintain their focus on ad relevance. Poorly targeted results train users to avoid sponsored links, and that hurts Google and advertisers alike.

Users, advertisers, and Google itself would be better served by bringing back the option for the old-style broad watch.

– George