Search Console Uses Block Flattening For AIOs, Forget Position & Focus On Outcomes via @sejournal, @TaylorDanRW

Declining clicks from organic search have been a shift in the industry for the past couple of years, and we’ve faced increasing pressure from clients and stakeholders to provide explanations for this, or to show that “traffic” has basically shifted to AI.

This has led to a panic in the industry, as we try to find a new normal to educate stakeholders on what matters and what doesn’t. Alongside this, AI tool vendors have been quick to jump into the space, forcing a narrative that their tool is the answer to AI visibility when, in reality, it just suits their product and not what is really needed.

On top of this, Google has now openly admitted that its Search Console reporting cannot keep up with how AI search actually works.

The discussion started when Google Search Advocate John Mueller replied to a thread on Reddit, explaining that tracking traditional rankings for AI features is “hard to do in a way that makes it useful.” Mueller went on to say that Search Console still treats entire AI-generated overviews as a single block.

In doing so, he confirmed what many marketers have noticed for a while, namely that trying to squeeze AI answers onto an old 1-to-10 ranking scale no longer makes sense.

This is a shock to the system, as we’re still working on stakeholder education to move away from what has been a reliable, stable, easily reportable, and forecastable value (potential traffic) metric.

Why ‘Average’ Position Gives The Wrong Impression

Even before AI arrived, Search Console’s ranking numbers could be misleading because they rely on a simple mathematical average. I’ve always had a problem with this metric in the form it is presented.

An average works well when numbers cluster together near the middle, but search results do not work like that. They change based on where someone is searching from, whether they are on a phone or a laptop, and what kind of result Google decides to show at that exact moment.

When you blend all those different outcomes into a single average, you end up with a number that rarely matches what real people actually saw. This gets even more confusing when you can change your views in Google Search Console for the same terms and pages, and achieve different average position numbers based on the cardinality and thresholds applied.

Your page might show up at position 1 whenever an AI Overview appears, but drop down to position 19 in normal results when it does not. If both things happen equally often, Search Console reports a tidy Average Position of 10.

A neat average hides these extremes, giving website owners a false sense of consistency.

See also: How Do I Know If AI Overviews Are Taking Clicks From My Site And What Can I Do About It? – Ask An SEO

What ‘Block Flattening’ Means In Practice

This problem gets worse because of how Search Console records AI Overviews, using an approach known as block flattening. Instead of checking where an individual link sits inside an AI summary, Google just looks at the overview box as a whole and gives it one overall rank. Because these AI boxes usually appear above the standard organic listings, the whole box is counted as position 1, and every single link inside it gets given that top position in your reports.

This creates a serious mismatch between the data and the real world. An eye-catching link card shown right under the first sentence gets recorded as position 1, but a plain link tucked inside a “Show More” dropdown also gets marked as position 1.

To Google’s reporting database, both placements are identical, even though a searcher is far more likely to click the first one and completely overlook the second. By grouping those two different placements together, the data suggests you have top-tier visibility when a user might never have noticed your site at all.

See also: AI Visibility Measurement: What To Track & What To Ignore

How Impression Rules Skew The Numbers

This flattening issue runs straight into Google’s standard rules for counting impressions, which were written long before search results became interactive. Under the normal rules, an impression is logged the moment a search result is loaded onto the page, regardless of whether the person ever scrolls down far enough to see it.

When an AI Overview loads, all of its default links count as impressions right away, which can easily fill your reports with views from people who only read the first sentence and left.

The opposite happens when a link sits behind an interactive button.

Search Console makes an exception for anything that needs a click to open, so links hidden behind a “Show More” toggle do not count as an impression until someone deliberately clicks to expand the section.

This leaves us with an odd contradiction where unread links in the main text get full credit for views they never really had, while useful links hidden behind a toggle look like they do not exist until someone takes the time to click on them.

See also: Google Put AI Visibility Inside The SEO Tool On Purpose

What This Means For Real Traffic

When you put these reporting quirks together, they make standard traffic metrics much harder to trust. Articles across Search Engine Journal show just how much AI answers are changing user behavior. Looking at data where AI Overview CTR fell by 61%, research from Seer Interactive showed that click-through rates for cited websites dropped noticeably, largely because Search Console recorded so many automatic impressions that the overall math was skewed.

At the same time, people simply have less reason to click through to an external site. A field study covered by SEJ found that having an AI Overview at the top of the page cut clicks to standard organic results by around 38%, because people got what they needed from the summary and moved on.

This pattern is particularly clear for informational searches, with research showing that AI Overviews appear on 21% of searches and show up far more often when someone asks a question. In those cases, Google is answering the query directly rather than pointing people to another website, breaking the old link between having a top ranking and getting real visitors.

Moving Beyond The 1-To-10 Ranking Scale

During his comments on Reddit, John Mueller mentioned that Google has tried to explain these rules in its help documents, while asking the SEO community for practical ideas on how position could be measured when search pages no longer look like a simple list.

Traditional search results historically were a neat list. Every SERP entry had an obvious number, but modern search results change constantly.

A single search might show three link cards on a mobile phone, show five on a computer screen, hide several more behind a button, and change the wording entirely if you run the search again a few minutes later.

Trying to force that kind of layout into a single column labeled “Average Position” gives a false sense of accuracy without offering any real insight.

How To Measure What Matters Now

The reality is that the old ways of measuring performance cannot simply be patched up to solve this, and the tsunami of AI visibility tools can’t provide any deeper or further insights into this problem.

If you want a clearer picture of how your content is doing in search today, it helps to step back from position numbers and focus on what users are actually doing.

Rather than worrying about whether your site has an average position of 1.2 in the AI report, which usually just means your link was included somewhere in the AI Overview box, it makes more sense to treat AI visibility as a simple yes-or-no question of whether your brand is being cited at all.

From there, the focus needs to move towards outcomes that cannot be distorted by how an impression is logged.

Actual site visits, genuine inquiries, and conversions are much more reliable indicators of performance than search engine numbers. Where you do need to know exactly how a page looks to searchers, checking the layout directly or using automated tools that inspect the live screen will tell you far more than Google’s flattened data.

The list of 10 blue links served us well for 25 years, but holding onto that model now will only cloud your understanding of your audience.

More Resources:

Featured Image: Summit Art Creations/Shutterstock