Between February and June 2026, a healthcare SaaS site's Google impressions fell from 1.15 million a month to 535,000. Cut in half in five months. That chart, on its own, has ended agency relationships and launched emergency content audits.
The correct diagnosis, reached over three monthly report cycles and confirmed by July's data, was: nothing is wrong. And the site's search performance was actually improving the entire time.
This is a case study in reading the right metrics, in resisting the most clickable interpretation of a scary chart, and in a small piece of discipline - writing down what would prove you wrong before the next data arrives - that kept a client relationship calm while half the impressions disappeared.
§ The numbers that caused the alarm
From Google Search Console, monthly, for the site in question:
| Month (2026) | Clicks | Impressions | CTR |
|---|---|---|---|
| February | 27,800 | 1,146,000 | 2.4% |
| March | 24,300 | 998,000 | 2.4% |
| April | 18,800 | 759,000 | 2.5% |
| May | 21,100 | 603,000 | 3.5% |
| June | 23,700 | 535,000 | 4.4% |
| July | 24,400 | 609,000 | 4.0% |
Look only at the impressions column and you see a site in freefall. Look only at February-to-April clicks and the story seems confirmed: down a third. If the monthly report had shipped in April, panic would have been a defensible conclusion.
It would also have been wrong, and the columns to the right are why.
§ What the full table actually says
Clicks bottomed in April and then climbed for three consecutive months, back to within noise of February's level. Click-through rate nearly doubled across the period, from 2.4% to over 4%. And in July the impressions themselves reversed, rising 14% month over month. August, at the time of writing, is pacing consistent with July.
Put together: Google showed the site to half as many searchers and the site got the same number of visits. That is not decline. That is a pruning of impressions that were never going to become clicks - appearances deep in the results for queries where the site ranked on page three or four, counted as impressions, clicked by nobody.
The mechanism matters. When a site loses far-page rankings for long-tail queries, its impression count collapses while its actual traffic barely moves, because impressions from position 35 were decorative to begin with. What remains is the inventory that works. The efficiency doubling is the tell: you cannot double CTR by getting worse.
One trap inside the same dataset deserves its own warning: average position moved from 8 to 15 over these months, which looks like catastrophic slippage. It is a mix-shift artifact. Average position is an average over whatever basket of impressions Google happened to measure that month, and this site's basket just changed by half a million impressions. When the composition moves that much, the average describes the new basket, not a change in rankings - and the proof is sitting in the other columns. If real rankings had slid from 8 to 15, clicks could not have held and CTR could not have doubled; pages do not get twice as clickable while dropping seven positions. I refused to headline that number in any client report without the mix explanation attached, and I would give the same advice to anyone: clicks and CTR are the honest health measures here.
§ The discipline that kept everyone calm
The useful part of this story is not the happy ending. It is what happened in the months before the ending was known.
When the decline first showed up, I gave the client a named hypothesis instead of a mood: this looks like benign long-tail trimming, not value loss. And I named, in writing, what would falsify it: if the decline started reaching into clicks and into the queries that drive signups, the benign reading was dead and we would escalate.
Then each monthly pull tested it. Clicks held - hypothesis survives. CTR rising - hypothesis strengthened. July impressions reversing - hypothesis confirmed, case closed, in exactly the direction predicted.
That structure - hypothesis plus falsifier, stated before the next data arrives - is the cheapest insurance in consulting. It converts "trust me, it's fine" into "here is what would change my mind, and we are watching for it together." When the reassuring reading turns out correct, you have a record showing it was reasoned, not lucky. And if the falsifier had fired instead, the escalation would have been immediate and pre-agreed rather than defensive and late.
§ How to run this diagnosis on your own scary chart
- Pull the monthly trend, not the 28-day comparison. Search Console's default views make every wiggle look like an event. Month-by-month over six or more months is where shape becomes visible.
- Read clicks before impressions. Impressions measure how often Google chose to show you; clicks measure whether the visibility you kept is the kind that matters.
- Watch CTR as the efficiency signal. Falling impressions with rising CTR is trimming. Falling impressions with flat or falling CTR is the bad version - that combination deserves the alarm.
- Do not quote average position across a composition change. If impressions moved massively, the average is describing a different basket of queries than last month. Compare positions per query or per page instead, on the queries you actually care about.
- Write the falsifier down. Whatever your reassuring interpretation is, name the data that would kill it, and check for that data on a schedule.
Numbers dated as of August 2026, from Google Search Console monthly data for the site in question, rounded; the site is anonymized as a matter of client confidentiality.
A halved impression count cost this client nothing except the few hours it took to diagnose honestly. The expensive version of this story - the emergency rewrite of content that was working, the migration launched to fix a decline that was not real - happens somewhere every week, usually because someone read one column of a six-column table. The chart that looks like a cliff is sometimes a haircut. The columns to the right will tell you which.