…That Will Save You (lots of) Money and (even more) Time!
The topic of negative keywords is one of those things you learn in SEM 101. It’s certainly one of the first things we discuss with our new hires, and it’s something they claim to understand rather well.
So, with such a topic you might expect consensus amongst PPC vets, but that certainly hasn’t been my experience (at all).
If I were to sum up the generally accepted approach, it’d be, simply put, “Favor exact match negatives.” Of the folks I hear from who have been in PPC for a while, about half take this approach. I tend to (not surprisingly) disagree.
To better define my logic on negatives, we need to start with the definitions of each match type. Surprisingly, a fairly large percentage of people who have been doing search for a fair amount of time don’t have a grasp of this definition, as they assume negative match types are the same as positive ones.
Here’s what Google says (they tend to use examples instead of hard definitions):
Negative Match Types
Exact: [free trial]
Adding this as a negative keyword would prevent your ads from showing on the search query free trial only
Example: free trial
Phrase: “free trial”
If you were to add free trial as a negative keyword to your account, the system would prevent any search query containing the phrase free trial from triggering your ads
Example: free trial lawyer
Broad: free trial
Adding free trial as a negative keyword to your account would prevent your ads from showing on any search queries containing the terms free and trial
Example: trial outcome defendant free
Typically, broad match negatives are the main point of confusion, as this match type is fundamentally different for negative and positive keywords. The misconception is that broad match negatives will expand to plurals, misspellings, semantic variations, etc., much in the way the positive broad match does. SEMs who believe this tend to avoid broad match negatives like the plague, since they also believe broad match negatives block far more than they actually do. If, however, we understand the definition, broad match negatives shouldn’t be feared.
So, understanding the definition of the match types put us quite a bit closer to an efficient (defined as “Achieving maximum productivity with minimum wasted effort or expense”) negatives approach.
General Match type Distribution
What is also lacking in the community is an understanding that negatives (in quantity, primarily) depend on how you use positives. To overly generalize, I’d create the following guidelines (note: the table assumes selection of relevant positive keywords):
Next, let’s discuss typical use cases I see for each negative match type. The first stage of my approach: the “Reactive Negatives Approach.”
Reactive Negatives Approach (Stage 1)
In the first stage of our negatives approach, we are looking at the search query report to reactively make adjustments.
After finding a bad query, I’d choose match type as follows:
Typically I only use exact match in accounts with a heavy positive broad match emphasis (which most well-established accounts should still be using). The typical use case is actually in the instance where you have a positive keyword of perhaps 3 tokens, and are matched to a query of fewer tokens. An example of this would be if you bought the keyword “power of attorney” in broad match and Google matched you to the query ‘power’ or ‘attorney.’ I see this rather often, so as a result my exact match negatives tend to be just one or two tokens long, and usually have high search volume. There is another instance where you might use an exact match, but it’s a bit more rare. This would be the instances where adding additional tokens actually changes the meaning of your keyword. To steal from our founder, David Rodnitzky, a great example is if you sell ‘night stands’ and want to negative [one night stand].
This is what I most commonly use in my accounts. It requires analysis as opposed to recognition (as in the case of exact match negatives). I tend to use phrase match when there is an irrelevant token (or more than one) contained within a query. For example, if I were a divorce lawyer and saw the query ‘heidi klum divorce,’ I’d make ‘heidi klum’ a negative phrase match. To be even more aggressive, I might consider making ‘heidi’ a negative as well as ‘klum’ and ‘klums.’ This is aggressive, but if you’ve never had a conversion containing these tokens, and you can’t think of instance where a query containing these tokens would be relevant to your business, then you should try this approach (as opposed to just making a negative of [Heidi klum divorce]). Otherwise you’re just waiting around to make misspellings and variations a negative the next time around. Not only does that type of reactivity cost you money, but it also takes extra time since you have to continue to be very diligent with negative scrubs each and every time.
The same can be said for all of those unique 1 click 1 impression queries you find in the SQR. I’ve found most people make these exact match negatives, but in all likelihood no one will ever search this query again, so it’s an effort that’s just for show! Instead, I’d trim this down to a phrase match negative too, which prevents us from being matched to a much larger set of bad queries. For example if I were matched to ‘fake divorce papers to make your wife upset’ instead of one exact match negative, I’d choose ‘fake’ and ‘upset,’ since not only will this prevent me from being matched to that strange query in future (if, by minor miracle, someone searched for it again) but also all other irrelevant queries that contain those tokens.
After understanding the definition of each negative match type, and using the ‘query trimming’ approach I just discussed, there isn’t much room left for broad match negatives. This match type is really only useful when your irrelevant tokens are separated by articles. For example, if you found the query ‘work at home’, and the irrelevant portions of the query were ‘work’ and ‘home,’ but only if they appeared in a query together, you might make ‘work home’ a negative broad match. Again, I think this is a pretty specific use case.
If you prefer, for whatever reason, to use broad match instead of phrase match negatives, it’s certainly valid to do so in most cases. If you’re very proactive with identifying negative tokens, phrase and broad match negatives are the same thing (i.e. a one-token phrase match is, by definition, the same of a one-token broad match negative).
After going through this type of process with your search query report (which, by the way, really doesn’t take much longer than a traditional search query scrub that relies on negative exact match), you’ve completed your reactive scrub. The next step is the proactive approach.
Proactive Approach (Phase 2)
As mentioned in the ‘phrase match’ negatives use case, it’s necessary to extend to plurals for your negatives. If you’d like, you can also extend to common misspellings. So, if you have ‘letter’ as a negative phrase match, you should also make ‘letters,’ ‘leters,’ and ‘leter’ negatives.
Additionally, however, you might want to consider an approach that extends to the semantic meanings of the negatives you’ve already found. For example, if you’re a B2B marketer who sells enterprise ‘live event streaming’ services, you might be matched to something like ‘free live soccer streaming,’ which is clearly B2C traffic. In addition to making ‘free’ and ‘soccer’ negatives, why not make ‘basketball,’ ‘baseball,’ and other sport names negatives? If you’ve seen that Google is matching you to the incorrect traffic, the idea is to not wait around for more of it.
You may also apply this approach pre-launch, which many people advocate, though I feel this can turn into a never-ending situation of hypotheticals, so I’d limit yourself to a certain time limit for an activity like this. Don’t wait too long to do your negatives scrub after launching a new account, though. As soon as query data becomes available, jump right in!
Summing It All Up
At the end of the day, I’m claiming that finding negatives is based primarily on semantics, not metrics (though, this is certainly an oversimplification!). That means that I don’t use ad group-level negatives (except to control query mappings) because what’s semantically irrelevant in one ad group is probably irrelevant in all ad groups.
I’ve been using this approach to negatives for a long time. Now, I have to admit a time or two I’ve been too aggressive with my negatives, and I’ve cost our client some conversions by blocking potentially relevant queries I hadn’t considered. That being said, we always check our complete list of negatives (which is usually only a few hundred) once per month or so to make sure we didn’t make a misstep. Those occasional missteps, and maybe a strange person who would have converted on a seemingly irrelevant query, are not nearly as impactful as waiting around for Google to charge us with bad clicks that we would’ve known in advance to be bad 99.9% of the time! With this approach, I can honestly say I’ve saved our clients thousands and thousands of dollars this year alone, and I’ve saved myself countless hours scrubbing search query reports.