Quick Answer
Inherited Search campaigns often park 100 keywords in a single ad group. Some lists run 500 to 1,500. The story you hear is that leftover terms train Smart Bidding. They do not. Budget leaks onto $5-a-month keywords until your best terms never collect enough clicks to judge. Group by intent, keep about 10 keywords in the ad group, and pause the dead weight.
The junk drawer nobody is allowed to pause
You inherit the Google Ads account. Or a director built the Search campaigns and treats the keyword list like a monument. Open one ad group and it looks like a dump. A hundred keywords. Sometimes five times that. I have opened groups with more than a thousand leftover rows, every match type stacked on every old idea.
The defense is always the same. Those terms still get impressions. Pause them and you "lose coverage." Dead keywords supposedly feed the algorithm. In accounts like this, roughly 20% of spend sits on keywords that burn maybe $5 a month. You cannot get a conversion off $5. You cannot get a trustworthy CTR or Quality Score off $5 either. The row just stays enabled because cutting it feels like shrinking the account.
I used to argue about this in weekly reviews. The person who built the list remembers when some of those terms "worked." They did, years ago, under a different match type and a fatter daily budget. The list never got pruned. It only grew.
Leftover keywords are not training data
Smart Bidding scores an auction from conversions and from queries that actually get traffic. A keyword that never spends enough to convert is not teaching the model. It is fighting your good terms for the same budget. Cheap, low intent queries often win that fight on broad match or a loose phrase match. High intent keywords wait. Search terms get noisy. The RSA has to cover too many intents. Ad relevance drops. Expected CTR follows.
SKAGs made sense when exact match meant exact. You put one keyword in an ad group and wrote copy just for it. That trick is mostly dead. The junk drawer is the other bad extreme. Every leftover term in one bucket, on the myth that more keywords give the algorithm more to learn. Theming got skipped.
A themed ad group is a small set of keywords that share one intent. They can share one RSA and one landing page without the copy hedging. Match type supports that theme. Negatives support that theme. Neither one replaces it. You cannot negative your way out of 400 unrelated rows.
People mix this up with the too-many-campaigns problem. That one splits conversions into tiny campaign buckets until Smart Bidding has nothing to chew on. This is the other mess. One bloated ad group. Every leftover keyword sharing the same ads. The signal is loud and you still cannot read it.
How to theme the ad group without going back to SKAGs
Step 1: Export the list and sort by intent
Download every keyword in the bloated ad group. Ignore match type on this pass. Exact, phrase, and broad of the same idea are the same idea. Sort the rows by what the searcher wants. Buy something. Compare options. Find a location. Apply for a job. Read an article.
You will get a few real clusters. You will also get a long pile that belongs to none of them. That pile is leftover spend that never amounts to anything.
Step 2: Build new ad groups around one intent, about 10 keywords each
Each cluster gets its own ad group. Keep the set small enough that one RSA can talk to every keyword. The landing page has to keep the same promise. If you have to hedge the ad copy so it fits, the group is still a drawer.
Ten is a working target. Eight can be right. Fifteen can be right if the intent is truly shared. A thousand is a filing error. If the keywords cannot share ad copy, they cannot share an ad group.
Step 3: Pause the dead weight in one batch
Pull cost for the last 30 days. Keywords sitting at a few dollars, no conversions, no realistic volume, pause them. Do it together so you can watch impression share on what you kept.
Someone will worry about lost impressions. Those impressions were leftover coverage. They were not a read. You want clicks stacking on the themed set. Then CTR, conversion rate, and the search terms report start meaning something.
Step 4: Let match types and negatives protect the theme
After the split, stop cloning the same keyword into exact, phrase, and broad inside one group out of habit. Put exact and phrase on the core terms. If you keep broad match, give it a group that can absorb extra query noise. Mine negatives from the search terms report on a schedule.
Negatives keep stray queries out. They do not fix a 400-keyword group. Theme first. Then negative.
Step 5: Rewrite the RSA for the 10, not the pile
Write headlines for the intent in that ad group. Point the final URL at the matching landing page. Give it two weeks. Promote converting search terms into the theme. Pause anything that still dribbles spend.
Now the clicks sit on a small themed set. You can finally tell which queries earn their keep.
If the inherited mess runs past one ad group, I walk through the full rebuild in Signal Rich Structures. It is a $49 course on Search campaign structure so Smart Bidding gets a clean signal. I show how to theme the groups so leftover keywords stop posing as training data.
FAQ
How many keywords should one ad group hold?
There is no official cap that saves you. What matters is whether every keyword can share the same intent and the same RSA. I aim near 10. If you cannot write one ad that fits the whole set, split the group.
Will pausing $5-a-month keywords hurt Smart Bidding?
No. Smart Bidding needs conversions and enough clicks on related queries to see a pattern. A keyword that spends $5 a month supplies neither. Pausing it frees budget for the themed terms that can actually produce a read.
Are SKAGs still worth using?
Almost never for a normal Search campaign. SKAGs were a control trick from the exact match era. Theming does the job SKAGs were supposed to do. It does not split conversion data across a hundred tiny ad groups.
What if leadership will not let me pause the leftover terms?
Show cost next to conversions for 30 days. Add how much impression share the leftover terms actually bought. Frame it as moving spend onto keywords that can collect data. If they still refuse, park the junk in its own low budget ad group. Then it cannot keep diluting the theme.