Cluster Analysis in Google Ads: How We Find Hidden Waste in Healthcare Accounts

By Raman Soni, Head of Performance Marketing at ICG.

A keyword report shows you how individual search terms perform. A cluster analysis shows you how themes of search intent perform, and that distinction reveals patterns that keyword-level reporting systematically hides.

At ICG, every healthcare Google Ads account we manage gets a weekly cluster summary. This post explains what it is, how to read it, and what it tells us that standard keyword reports miss.

What a keyword cluster is

A cluster is a group of search terms that share a common semantic theme. Instead of tracking "psychologist in Delhi" and "psychologist near me" and "therapist Gurgaon" as separate keywords, we track them as the "Psychologist" cluster. Instead of "best dermatologist" and "top skin doctor" and "specialist doctor", we track the "Best Doctor" cluster.

The clustering logic for each account is built from the actual search terms triggering ads, not from assumed keyword lists. This means clusters reflect how patients actually search, not how the campaign was initially structured.

How to read the cluster summary

Reading the table in the screenshot:

"Best Doctor" cluster (+81.4% impressions, +111.8% clicks): This is a positive signal. Patient searches for doctor-specific terms are growing sharply, either because the campaign is now showing for more relevant queries, or because demand for this intent is rising. This cluster deserves more budget and dedicated ad copy.

"Hindi" cluster (+130.4% impressions, +200% clicks): "Pagal" and "Paagal", Hindi colloquial terms for mental health conditions. A 130% impression spike on this cluster indicates that a patient segment is entering the funnel through vernacular searches. This is not a niche experiment. It is a signal to build Hindi-language ad copy and landing pages.

"Psychologist" cluster (-3.3% impressions, -0.9% clicks): The core cluster is flat to slightly declining. If this cluster carries the majority of budget, declining impressions could mean: competitor increased bids, Quality Score dropped, or impression share is being lost to budget constraints.

"Near" cluster (+10.9% impressions, +14.8% clicks): Location-intent queries growing. This means either more patients are discovering the service, or the campaign is now showing for more near-me queries, which may or may not be desirable depending on the account's CPA target by tier.

What cluster analysis reveals that keyword reports miss

1. Trend direction before it shows in conversions. A 130% impression spike in the Hindi cluster will show in conversion data 2-3 weeks later, by which time competitors may already have captured the demand. Cluster analysis flags the signal early.

2. Budget concentration risk. If 60% of clicks are in the "Near" cluster and the account's CPA target is aggressive, the account is over-indexed on location intent (moderate converter) and under-indexed on treatment intent (high converter).

3. Campaign structure gaps. A cluster with rising impressions and low clicks means the ad copy is not relevant to the theme. The cluster analysis tells you where to write new ads, not where to adjust bids.

4. Negative keyword priorities. Clusters with high impression growth and near-zero clicks (like "Research/How-to" queries) are consuming match budget. These are the first clusters to address in a negative keyword session.

How to build your own cluster summary

1. Export search term report for 60 days (or two comparable periods)

angryturtle/07-top-keywords.png" alt="Angryturtle Top Keywords — every keyword the listing appears for, with impressions, rank, and click share" loading="lazy" decoding="async" style="width:100%;height:auto;display:block;">

2. For each search term, assign a cluster label based on the dominant theme keyword

3. Aggregate: sum impressions and clicks per cluster per period

4. Calculate delta %

5. Sort by impressions delta descending. The clusters at the top are where the account is moving

The process takes 2-3 hours for a mid-size account. At ICG, it is automated through Agency OS and runs weekly for every account.