If the campaign allocates the money, the ad group allocates the meaning: it is the place where specific searches meet the ad written for them. A good group raises relevance (a third of Quality Score), CTR and conversion rate all at once. A badly built group drags down every ad in it, no matter how good they are.
The question that defines a group
Every keyword in a group shares the same ads. So the only question that matters when grouping is:
Do the person searching for this and the person searching for that expect to read the same ad?
If yes, they go together. If no, they are two groups. "Dental implants price" and "dental implants urgent" are the same service, but the person searching on price wants to see a figure or a finance option in the ad, and the person with an emergency wants to see "same day" and a phone number. Two groups.
How many keywords per group
Years ago the recommendation was single keyword ad groups (SKAGs), to control the ad down to the millimetre. Today, with intent-based match types and responsive ads, that structure fragments your data and adds nothing. The practical benchmark in 2026:
| Size | Verdict |
|---|---|
| 1-3 keywords | Too fine-grained except for very special terms (brand, flagship product) |
| 5-20 keywords | The healthy range: one intent and its natural variants |
| 30-50 | Review it: there are probably two intents mixed together |
| More than 50 | Junk drawer: no ad can be relevant to all of them |
And within the group, 2-3 responsive ads at most: enough for Google to test combinations, not so many that the impressions get diluted.
The universal mistake: the junk drawer
Almost every inherited account has a group called "General", "Misc" or named after the campaign, holding 80 keywords of every kind and one generic ad. The symptoms: low Quality Score on relevance, weak CTR and a conversion rate that does not match the rest. The cure is not writing a better ad (impossible: there is no good ad for 80 intents), it is splitting the group.
A three-step method for designing groups
Step 1 · List the intents, not the words
Before you open the keyword tool, write down what the people searching for you want: to compare prices, to solve an emergency, to inform themselves, to buy a specific brand, to find something nearby... Each intent is a candidate group.
Step 2 · Assign variants to each intent
Now, yes: for each intent, the ways of saying it. «dental implant price», «how much does an implant cost», «cheap dental implants» → group Implants · Price. If a variant does not fit any intent, it is either a new group or a negative.
Step 3 · Write each group's ad before you create anything
If you cannot write an ad that answers every keyword in the group, the group is badly split. This step is the coherence test: cheap to run, and it saves you months of mediocre Quality Score.
Example: from one drawer to four groups
An inherited group called "Plumbing" (62 keywords, average QS 4). After applying the method:
| New group | Intent | Example ad headline |
|---|---|---|
| Plumber · Emergencies | Fix it NOW | "24h Emergency Plumber · At Your Door in 1 Hour" |
| Plumber · Quotes | Compare prices | "Free Plumbing Quote · No Obligation" |
| Drain unblocking | A specific problem | "Drain Jetting Service · Written Guarantee" |
| Water heaters and boilers | Appliance install/repair | "Boiler Repairs · All Makes and Models" |
Same budget, essentially the same keywords: the average Quality Score went from 4 to 7 in six weeks and the CPA dropped by a third, without touching the bids. It is the effect we saw in the auction lesson: relevance = cheaper clicks.
💡 Ninja trick: regrouping is tedious work in large accounts (hundreds of keywords, thousands of search terms). Our QS Analyzer spots the groups with low relevance, and the RSA Optimizer, in its «Restructure groups» tab, proposes with AI how to split them and, if you accept with a checkbox, creates the new groups and pauses the original. The three-step method, automated.
What you should remember
- One group = one intent that deserves the same ad.
- Healthy size: 5-20 keywords and 2-3 ads.
- The junk drawer is not fixed with better ads: it is split.
- Method: intents → variants → write the ad before creating the group (the coherence test).