Skip to content
Analytics

Google impressions decreased by 72%, although positions improved. What has really changed?

28 August 2026·8 min
Google impressions decreased by 72%, although positions improved. What has really changed?

The first reaction to such a red graph is almost automatic: "Google has imposed a penalty", "the page has been removed from the index" or "recent changes have harmed SEO". In our case, none of these hypotheses fit the full picture.

In Google Search Console WebHeaven we compared two equal, five-day periods covering the same days of the week. Views dropped from 2033 to 569, i.e. by 72%. At the same time, the average position improved from 37,63 to 27,83.

This is not proof that anything works better. It is also not evidence of punishment. This is a signal that the viewing chart alone is not sufficient for diagnosis.

The data looked bad - until we put it together

We compared the period 15–19 August with the period 22–26 August 2026. Both windows cover Saturday, Sunday, Monday, Tuesday and Wednesday, thus limiting the impact of differences between the days of the week.

Indicator 15–19 August 22–26 August Change
Views 2033 569 −72,0%
Clicks 3 0 −3
Average position 37,63 27,83 improved by 9,80 space

With just three clicks, you cannot draw far-reaching conclusions about conversion or traffic quality. However, the number of impressions was high enough to ask the right question: how could visibility drop when the average position improved?

Average position in Search Console does not describe the entire page

Average position is calculated for impressions that actually occurred. If your site stops showing up for hundreds of broad, low-match queries from 40–70, those low rankings disappear from the average.

There are narrower and better-tailored queries left, in which the page ranks, for example, 8–25. The effect may look paradoxical:

  • views are dropping dramatically;
  • average position increases;
  • click rate not improving;
  • most important pages can remain indexed correctly.

Therefore, "average position increased" does not automatically mean that SEO is winning. Likewise, "impressions have dropped" does not automatically mean that your site has been penalized.

We described more about reading data in the guide how to measure SEO and its impact on queries and sales.

What we checked before any big change

Instead of rebuilding the page right away, we split the diagnosis into four layers.

1. Pages that lost the most views

The decline was not evenly distributed. Broad local pages of cities, which previously appeared in a large number of queries but generated almost no clicks, lost the most views.

This is an important difference. Losing 1000 impressions with a CTR close to zero is a different problem than losing 1000 impressions from phrases that previously regularly delivered customers.

2. Indexing status and canonical addresses

Top declining pages continued to pass indexing checks. Google was able to download them, the robots.txt file didn't block them, and the canonical addresses matched.

If the problem were a technical failure, we would expect a broader pattern: download failures, noindex, invalid canonicals, redirects or a large group of addresses removed from the index.

This pattern did not appear. This avoided annoying rollbacks, mass URL reporting, and another rebuild without proof.

If visibility decreases after migration or deployment, start over technical control for migration and redirection, and not from adding further texts.

3. Home page and highest value pages

The home page did not show a similar breakdown. Her view count was stable and her average position improved. Two important local landing pages were also not removed from the index.

This is another argument against the general domain penalty thesis. When the problem mainly concerns a wide group of pages with a similar pattern, you must first examine their intention, uniqueness and real value - not the entire domain at once.

4. Queries, not just URLs

One page can appear for dozens of different intentions. Some of them are valuable, some are random, and some result from Google temporarily testing the results.

Therefore, when decreasing, we analyze separately:

  • queries that lost views;
  • queries still generating clicks;
  • new queries with better matching;
  • local phrases mixing the home page with city landing pages;
  • devices and countries if the traffic profile may have changed.

Without this, it's easy to "fix" the page for traffic that never had a chance to turn into a query.

Is AI at Google behind the decline?

The honest answer is that we have no proof that AI caused this particular decline.

Google is expanding AI Overviews and AI Mode, and the search experience is actually changing. In May 2026 Google reported that queries in AI Mode are becoming more complex and conversational, and the system uses the technique query fan-outto search for information in several related areas in parallel.

This may change the distribution of views between pages and content types. However, you must not attribute every weekly change to AI. Seasonality, demand shifts, performance testing, competition, content freshness, user location, and simply shifting the query mix also come into play.

Importantly, Google still uses SEO fundamentals as the foundation for visibility in generative features. The official recommendations do not require a magical "GEO", a special file for AI, or page multiplication. These include:

  • content useful and difficult to replace with a generic summary;
  • own experience, data or point of view;
  • clear page structure;
  • correct indexing;
  • good photos and videos;
  • consistency of information about the company and offer.

Google explicitly discourages treatment llms.txt or special tags as a shortcut to visibility in AI Search. This is good news: the most important work still happens on the site and in the quality of the information, not in another "hack".

What we did instead of panic rebuilding

First, we determined which elements are stable and which require observation. We didn't roll back the entire deployment and ask Google to re-index all the addresses.

We limited actions to what had concrete evidence:

  1. we have confirmed indexing and canonicalization of the most important pages;
  2. we have separated the inheritance by addresses and queries;
  3. we left the home page without a sudden change during an ongoing measurement;
  4. corrected two incorrect links in the local draft of the future version of the home page;
  5. for the recreated service page, which Google still remembered as 404, we performed a test of the published version and a single submission to the queue;
  6. we have saved the reference point for the next comparison.

It's less flashy than the "big SEO fix," but safer. With low-click data, the greatest risk is changing content, architecture, linking and metadata at the same time. Later it is impossible to determine what helped and what harmed.

Visibility drop diagnosis plan in 60 minutes

If the Search Console chart suddenly drops, follow this sequence.

First 10 minutes: compare equal periods

Compare the same days of the week and ensure that the newer data is complete. Don't compare the full week with the last three days.

Write down four numbers:

  • clicks;
  • views;
  • CTR;
  • average position.

Next 15 minutes: find the parties responsible for the change

Sort addresses by view difference. Check whether the decline concerns the entire domain, one directory, template, language or several specific landing pages.

Next 15 minutes: go to queries

For pages with the highest decline, check for lost phrases. Separate sales enquiries from very broad, no-click impressions.

Next 10 minutes: check the technique

For the most important URLs, confirm:

  • HTTP response 200;
  • missing noindex;
  • valid canonical;
  • access for Googlebot;
  • presence in sitemap;
  • last download and indexing status in GSC.

Last 10 minutes: select one hypothesis for test

Don't change everything. Choose one area: content intent, internal linking, incorrect canonical, weak title, missing content or technical download error.

Record the change date and next measurement date.

How to create content that stands a chance in Search and Discover

There is no "enable Google Discover" button. Google automatically selects content based on users' interests, and meeting the requirements does not guarantee impressions.

However, you can increase the chance by combining timeliness, credibility and good visuals. Official Google recommendations for Discover indicates, among other things, large, high-quality images with a width of at least 1200 px, representative og:image and setting max-image-preview:large.

Even more important is the material itself. The content should give the reader something that they won't find in another automated list:

  • own data or experience;
  • clearly described measurement method;
  • fair separation of fact from hypothesis;
  • practical, actionable process;
  • current primary sources;
  • interesting title, but consistent with the content.

This article does not promise that we know one cause for all declines. It shows something more useful: how not to confuse improvement in average position with improvement in overall visibility, and how not to react too abruptly.

When a decline requires immediate response

Not every drop is an alarm, but some signals warrant quick inspection:

  • home page or key services disappear from the index;
  • increasing number of 404 or 5xx errors;
  • Google chooses different canonical than site owner;
  • deployed noindex or robots.txt lock;
  • sitemap contains redirects or non-existent addresses;
  • simultaneously, impressions, clicks and positions of the most important phrases drop;
  • loss of visibility begins right after a migration, domain change, or large deployment.

In such a situation, a technical diagnosis is needed, not waiting for the "algorithm to return". It will be helpful too SEO audit checklist before starting SEO.

The most important conclusion: first diagnosis, then movement

The graph can look dramatic and yet not indicate a domain-wide penalty or outage. In our five-day comparison, impressions dropped by 72%, but average position improved by almost 10 places. Only the analysis of pages, queries and indexing showed that the problem was mainly concentrated in wide local impressions with low click potential.

This still requires observation and work on content quality. However, it does not require a panic reset of the website.

If your Search Console shows a similar decline, we can start with a clearly scoped website audit for PLN 990 net. We review the data, indexing and key URLs, then provide a prioritised action list.

Check scope of website audit WebHeaven

Editorial sources

  1. Google Search Central, Optimizing your website for generative AI features on Google Search, update 2026.
  2. Google Search Central, Get on Discover, current guidelines for content quality and large images.
  3. Google, How AI Mode is changing the way people search, 19 May 2026.
  4. Google's own data Search Console WebHeaven, comparison of 15–19 August and 22–26 August 2026.

Related services WebHeaven

This guide links to specific services that help you translate what you learn from the article into implementation.

View GA4 implementation Learn about the scope of GA4 implementation and form, phone and purchase measurement testing. We will set events and data limits for your website.
WH

Author: WebHeaven

Article prepared by the WebHeaven team based on work on company websites, WooCommerce stores, SEO audits and campaigns aimed at obtaining queries from Google.

Consult your website with WebHeaven

WebHeaven guides

All services · All English guides