Healthcare Case Study: Impressions Up 21%, Clicks Flat, CTR Down

The situation

The client is a home care services provider operating across multiple US states, which places it firmly in YMYL territory. Health and caregiving decisions are judged against the highest standard in Google’s quality rater guidelines, and the site’s structure depended heavily on state-level landing pages.

The brief was a familiar one: traffic is down, we think an algorithm update hit us, fix it. The first job in any engagement is to check whether the premise is true.

What the data actually showed

Metric	Previous 3 months	Last 3 months	Measurement (what it means)
Clicks	199,000	198,000	Essentially flat, not a traffic loss
Impressions	9.16 million	11.1 million	Up 21 percent, so visibility grew
Average CTR	2.2 percent	1.8 percent	Down 18 percent, the real problem
Average position	22.5	23.7	Slightly worse, but not a collapse

Read those four lines together and the story is clear. Google was surfacing the site for two million more searches a quarter and those extra impressions were converting to nothing. Average position sitting at 23 tells you why: the new visibility was arriving on page two and three, where clicks are scarce.

This is a pattern I now see constantly, and it is covered in more depth in impressions stable, clicks falling. A site in this position that starts deleting content or rewriting everything is solving a problem it does not have.

The diagnosis

Three causes were in play at once, which is what made this engagement longer than a single-cause recovery.

• State landing pages carried most of the loss. Why: The site had one page per state, and they shared most of their text with each other. Near-duplicate location pages are the single most consistent casualty pattern in recent core updates, because the differences between them were the state name rather than anything a resident of that state actually needed.
• Thin content on the pages that were losing ground. Why: Analysis of the losing URLs showed a consistent gap against whoever now outranked them. The competing pages were not dramatically better written. They were more complete, and on a query where a reader has a real decision to make, completeness wins.
• A share of content read as machine-written without review. Why: Not because AI assistance is a violation in itself, but because the output had gone up unedited. Under Google’s scaled content abuse policy, what matters is whether pages were produced mainly to rank rather than to help, and unreviewed output almost always reads that way.

Alongside those, the audit found on-page problems that were suppressing performance without causing the drop: missing heading structure on several templates, outbound links that should have been nofollowed, and a lack of supporting media on long pages. Indexing and backlink strength were both healthy, which was useful information because it eliminated two expensive lines of investigation immediately.

Why YMYL made this harder

A home care site is judged on trust before it is judged on anything else. Google’s rater guidelines state that trust is the most important member of the E-E-A-T family, and that a page lacking it rates low regardless of how expert or authoritative it otherwise appears.

In practice that means a state landing page for a caregiving service cannot be a template with a place name inserted. Someone deciding how to arrange care for a parent needs the specifics that apply to them: the eligibility rules in their state, the programme names, who is accountable for the information. Those specifics are also, conveniently, what makes a page genuinely different from the forty-nine others on the same site. The YMYL recovery guide covers the wider standard.

The roadmap

Delivered in priority order, so the highest-impact work would be live and crawled before the next core update rather than half of everything being done.

• Rebuild the state pages with genuinely local substance. Why: State-specific eligibility, programme detail and local proof, researched per state rather than generated from a template. This was the largest single piece of work and the one most likely to move the result.
• Expand the thin losers to match the depth of what outranked them. Why: Benchmarked against the specific competing pages rather than against a word count target, because the goal is completeness on that question, not length.
• Stop unreviewed publishing and edit what had already gone out. Why: Fixing the archive without fixing the production process means the ratio returns within months. This was the change the client had to own.
• Add E-E-A-T signals throughout. Why: Named authors with relevant credentials, sourced claims, and visible accountability on pages that advise on care decisions.
• Clear the on-page and technical findings. Why: Heading structure, link attributes and supporting media. None of these caused the drop, but each one caps how well the content work can perform.

What this case teaches

Three things worth carrying to any similar situation.

Check the premise before accepting the brief. This client believed it had lost traffic. It had lost earning efficiency while gaining visibility, which is a different problem with a different fix. Had the engagement started from the stated brief, the first month would have gone into recovering rankings that were never lost.

Rising impressions are not automatically good news. Two million extra impressions a quarter that convert to nothing means Google is testing you in positions you cannot earn from. It is a signal that the pages are considered relevant but not yet convincing.

Location pages need local substance or they become a liability. The template approach that built traffic for a decade is now the most reliable way to lose it, because near-duplicate pages are exactly what recent updates have been narrowing. If the only difference between two of your pages is a place name, they are one page.

Timelines for work of this scope, including why a site-wide content project usually waits on a core update to be reassessed, are set out in how long recovery takes.

Figures in this case study are taken directly from the client’s Google Search Console during the March 2025 audit. The client is not named and identifying details are withheld under the confidentiality terms set out in the editorial policy. Diagnostic findings are reproduced from the audit report delivered at the time.