Case Study: My Site Grew While One Article Lost 74% of Impressions

A growing website can still have a declining page

I build Local Waifu, a desktop AI companion application, and GSC Magic, a tool for querying search data through AI assistants. I was checking Local Waifu’s Google Search Console performance using GSC Magic when an encouraging site-wide result caught my attention: organic Google clicks had increased across two consecutive, equal-length periods.

That number was real. But it was not the whole story.

Further down the page-level comparison, an article about an offline alternative to Janitor AI had lost most of its Google search visibility. The rest of the site was doing better, so a dashboard that displayed only the total would have obscured the decline.

This case study describes what the recorded numbers show, how I would investigate them and which conclusions I deliberately do not draw from such a small sample. It is not a claim that GSC Magic caused the growth. It did not. The product helped surface a contrast in existing Search Console data.

The two 28-day comparisons

The measurements below came from two consecutive 28-day reporting periods. The original comparison did not record their exact calendar dates in the notes used for this article. I therefore won’t invent a date range, publication event, Google update or intervention to make the story sound neater.

The site gained 174 clicks, but the affected page lost eight. Those statements are completely compatible. Several other pages can grow more than a single page declines, even if the shrinking page deserves attention.

The article’s impressions dropped much more sharply than the site’s overall click trend suggested. Impressions measure how often the page was shown in eligible Google search results, not visits or sales. The 353-impression difference is the clearest signal that this particular URL appeared much less often.

Notice the scale: 12 clicks becoming 4 is a small count. One or two extra visits would materially change the percentage. I would not describe this as a site-wide SEO emergency or present the percentage as a precise estimate of future performance.

Why the site-wide average sent the wrong operational signal

A site-wide total answers: “How many clicks did Google send to this property?” It does not answer: “Which content became more or less visible?” A mixed portfolio of product, documentation, comparison and informational pages can move in opposite directions.

Consider a founder evaluating marketing by the weekly chart alone. If the line rises, the natural instinct is to publish more of the same or move on to the next project. But an older comparison page might stop reaching relevant searches, a key integration guide might start ranking worse, or a page might attract more impressions while receiving fewer clicks.

None of those cases can be diagnosed from one property-level number. You need at least a page-level breakdown, and often a page-by-query breakdown.

Google’s own Performance report documentation explains that aggregation by property and aggregation by page can differ. Treat page totals, property totals and query-filtered totals as distinct views rather than numbers to subtract indiscriminately.

The correct operational response to rising total traffic is therefore not “everything is fine.” It is “which pages drove the change, and did any important pages move in the other direction?”

Step 1: reproduce the page decline with complete dates

Before touching the article, I would reproduce the comparison directly in Search Console. Select the same property, search type and two complete 28-day windows. Avoid a fresh partial day in one window or a different search type in the other.

Next, filter to the exact URL of the declining article. Check that URL variants are not splitting the result: trailing slashes, query strings, language versions, redirects and canonicals can complicate a comparison.

Then record clicks, impressions, CTR and average position for each period. We know the first two metrics from the original notes, but I have not supplied verified CTR or position for these particular windows. Filling those cells from memory would make the analysis less trustworthy.

A reproducible report should record the property, dates, search type, page URL, filters and any API limitations alongside the numbers. That also makes a future follow-up meaningful.

If you’re using an AI assistant connected to Search Console, the first question can be:

Compare this exact page’s Google clicks and impressions for two consecutive, complete 28-day periods. Return both totals and their absolute and percentage changes. Do not infer a cause yet.

This prompt forces the assistant to separate a calculation from an explanation.

Step 2: inspect the queries rather than guessing about rankings

Next, I would group the affected URL’s performance by search query. I would look for terms responsible for the largest impression and click differences, and check whether the page used to rank for a broad mix of terms but now appears for only a narrower subset.

Several mechanisms could create a similar pattern. Search demand for the topic might have fallen. Another page could now answer the query better. Google might show a different result type. The article could have lost position for important terms. A competing URL on the same site might receive the visibility instead. Or Search Console’s anonymized and truncated query rows might leave part of the story hidden.

Those are hypotheses, not findings from the three numbers we have.

Google documents how anonymized queries and table limitations can make row totals differ from chart totals. If the filtered query list sums to fewer impressions than the page’s reported total, don’t invent a missing keyword.

A useful AI follow-up is:

For this page, show the top queries with the largest impression losses between the same two periods. Separate changes in position, CTR and query impressions. Mark any missing or insufficient data explicitly.

This helps decide whether to investigate topical demand, snippets or search visibility first.

Step 3: distinguish seasonality, position and technical problems

A decline in impressions is not automatically a technical SEO problem. Before modifying content, I would perform three independent checks.

Seasonality and demand. Compare a longer window or a year-over-year period where data is available. Search demand for consumer AI applications and companion products can change quickly. Even a perfectly indexed page can receive fewer impressions when people stop searching a term.

Search position and CTR. If impression opportunities are stable but rankings decline, inspect competing results and whether the page matches intent. If position is stable but CTR worsens, title wording or SERP presentation may deserve attention. These two patterns lead to different work.

Indexing and canonicalization. Check the URL’s reported indexing state, canonical and any accidental noindex, redirect or server issue. URL Inspection can help, but it does not prove that a page should rank for a particular query.

I would also look at country and device splits. A shift from desktop to mobile impressions, or traffic moving between regions, can change the apparent average even without one universal ranking change.

For a repeatable process, use the SEO traffic drop checklist and the content decay guide. Both give more structure than asking the model to “fix my SEO.”

Step 4: decide whether the content actually needs an update

The natural response to a declining article is to rewrite it. That can be the wrong response.

An offline alternative comparison can lose demand because the competing product changed, because readers now search for different features, or because older claims stopped matching the market. Updating the article is sensible only after checking its accuracy and the queries that used to find it.

My review checklist would cover:

• Is the alternative still relevant to the original user’s need?
• Are download instructions, supported systems, pricing and product capabilities accurate?
• Does the introduction answer the query without forcing the reader through unrelated context?
• Does the page disclose limitations and give a fair comparison?
• Are links, screenshots and dates still useful?
• Does an existing page on the same site already answer the same question more completely?

If the content is accurate and the decline follows falling demand, rewriting it simply to satisfy an SEO schedule may create work without results. If it is outdated or mismatched to current intent, a targeted update is much more defensible.

What this evidence cannot establish

The figures show a temporal difference, not a causal chain. They do not establish that Google changed its algorithm for this page, that competitors stole the traffic, that an AI Overview replaced the result, that the page was deindexed or that any GSC Magic feature changed the outcome.

Nor does the site-wide increase prove that the broader content strategy worked because of one particular intervention. We would need additional evidence: exact date windows, page and query breakdowns, relevant changes to the website and ideally a follow-up period.

Percentages are useful labels but can be misleading on low denominators. A decline from 12 clicks to 4 deserves inspection; it should not be described with the same confidence as a sustained decline from 12,000 to 4,000.

If a future report shows a recovery, I would still separate what changed from why it changed. A revised title followed by more traffic is not automatically proof that the title caused the increase.

How I used an AI connector without treating it as the answer

GSC Magic’s role here is straightforward: connect a permitted Search Console property to an AI assistant, return measured page and query data, and help formulate the next investigation. It can make comparisons easier to request and repeat.

It cannot reveal Google queries that Google withholds, know a search algorithm’s private reasoning, or replace judgment about content quality. Its responses are only as useful as the date ranges, permissions and data supplied.

For someone wanting to reproduce this type of page-level review, the AI SEO audit guide covers the broader workflow. The AI SEO audit landing page shows what a connected audit might look like using clearly labeled sample data, and GSC Magic’s pricing explains the available site and history limits.

You can also reproduce the essential comparison in Google’s own Search Console interface without buying any tool. The value of an integration is convenience, repeatability and access through your chosen workflow—not ownership of the data.

The repeatable lesson for a small SaaS website

I now regard a monthly SEO review as two separate questions:

First, did the overall site grow or decline? That determines whether the acquisition channel is changing in aggregate.

Second, which commercially important pages moved against the overall trend? That tells me where a page-level investigation might be worthwhile. A landing page that loses qualified impressions can matter more to the business than a low-intent article gaining the same number.

A small founder-led site doesn’t need a giant weekly reporting presentation. It needs a short list of pages, the supporting metrics and a decision for each: monitor, investigate, improve or leave alone.

That’s what this Local Waifu example taught me. The property grew, one article lost visibility, and the right next move was to look more closely—not to claim a crisis or celebrate a tool.

Frequently asked questions

Can Google clicks rise while one article loses impressions?

Yes. Property totals combine many URLs. Other pages can gain clicks while an individual page becomes visible less often. Inspect both property-wide and page-filtered performance before interpreting growth.

Did GSC Magic increase Local Waifu’s search traffic?

No such claim is supported here. GSC Magic was used to inspect the data; it was not shown to have caused any ranking or traffic changes.

Does a 74% impression decline prove a Google penalty?

No. Search demand, rankings, query mix, indexing and other factors can affect impressions. This comparison does not identify which mechanism explains the decline.

Why not fix the page immediately?

The counts are small, and without query, position and indexing evidence a rewrite could target the wrong cause. The more efficient approach is to verify the cause of concern before choosing an intervention.

What I will monitor next

The next review should record the affected page’s impressions, clicks, CTR and position over another full comparable period, then track the relevant query segments. If I make a content change, I will record exactly what changed and when, while recognizing that correlation still will not establish causation.

The broader habit is simple: when the site-wide chart looks healthy, check the pages that disagree. For related workflows, see Search Console content decay analysis, a structured AI SEO audit or the GSC Magic connector documentation.