Search Console Block Flattening in AI Overviews Explained

24 September 2026 · The Cresia team

Does Search Console's position figure tell you how visible you are in AI Overviews? Not on its own. Search Engine Journal reports that Search Console uses block flattening for AI Overviews, which can make weak visibility look like a top ranking, so the safer way to judge AI search performance is visits and conversions.

Key takeaways

Search Console position can flatter your AI Overview presence, because the block's placement is not the same as your link's prominence inside it.
Treat position as a label for where a block sat, not as proof that anyone saw or clicked your citation.
Judge AI search by visits, engaged sessions and conversions on the pages that get cited, segmented from classic organic.
Get cited by writing self-contained passages that answer one question fully, and keep your facts consistent across your site.
Don't build AEO or GEO dashboards around a single position number, and don't chase every prompt variation.

What block flattening means for your Search Console data

A classic results page is a ranked list. Position 3 means something: there are two results above you, and a searcher scrolling down meets you third. That meaning is what most SEO reporting is built on.

An AI Overview isn't a list in that sense. It's one block, usually at the top of the page, holding a generated answer and a handful of source links. Search Engine Journal's report describes Search Console handling this by flattening the block: the position belongs to the block as a whole, rather than to your link's standing within it.

The practical consequence is easy to state. If the block sits at the top of the page, a link inside it can be reported as a top position even when that link is one small source among several, tucked into a carousel or a collapsed list. The number describes where the block was. It says much less about how easy your link was to notice.

Exactly how position is defined for your property, and how AI features are folded into it, is something to confirm in Google's Search Console documentation rather than assume. Definitions like this get revised, and the Search Central blog is where those revisions show up.

The headline is that position, the number many teams have quoted for a decade, has lost some of its meaning for a growing slice of queries. Nothing is broken. The metric is doing what it was designed to do. It was just designed for a page that looked different.

Why a top position can hide weak AI Overview visibility

Here's an illustration, not a case study. A personal-finance publisher tracks a set of 200 question-style queries, such as how much deposit a first-time buyer needs. Average position for the set improves over a quarter. The team reports it as a win.

Clicks to those pages don't move. Some fall. Nobody can explain the gap, because the dashboard says the pages rank better than ever.

The likely explanation is that many of those queries now trigger an AI Overview, the publisher's pages are cited somewhere inside it, and Search Console records the block's high position. The searcher reads the generated answer, perhaps glances at the source icons, and leaves. The page was technically present. It wasn't visited.

That's the pattern to watch for: average position improving while clicks and conversions stay flat or drop. Before block flattening became a topic, you might have blamed seasonality or a competitor. Now there's a more specific suspect.

There's a second-order problem. Once position stops discriminating between a prominent citation and a marginal one, it stops being useful for prioritising work. A page reported at position 1.2 might be a real winner or might be a passive footnote. You can't tell from the number, so you can't decide whether it deserves investment or whether it's already done.

This is also why reports about the growth of AI Overviews matter for measurement, not just for strategy. Search Engine Land has covered AI Overviews appearing on a large share of US desktop searches, and Search Engine Roundtable has reported on AI Overviews carrying far more external links. More coverage plus more links means more pages inside blocks, and more rows in your report where position is doing something other than what you think.

How AI answer engines choose and cite sources

Measurement only makes sense if you have a working model of what's being measured. Here's the broad picture, which holds across Google's AI Overviews, AI Mode, and the standalone answer engines, though each differs in detail.

An answer engine takes a question and often expands it into several narrower ones. It retrieves candidate pages for each, pulls out passages that address them, and writes a synthesis. Citations get attached to the passages that supported the claims. The unit that matters is the passage, not the page.

That has a few consequences:

A page can be cited for one paragraph. The rest of it may never be read by the engine.
Self-contained passages travel well. A paragraph that states the question, gives the answer, and names the entities involved can be lifted without losing meaning. A paragraph that says 'as mentioned above' can't.
Consistency counts. If your pricing, definitions or dates disagree across your own pages, an engine has reason to prefer a source that doesn't contradict itself.
Crawlability is table stakes. A page that can't be fetched, rendered or parsed can't be cited, however good it is.

A citation is also not a click. Engines show sources for verification and attribution, and people mostly read the answer. Search Engine Land has covered a study reporting that Google's AI Mode cuts clicks; treat that as direction, not a number to plug into your forecast. Your own analytics will tell you what your audience does.

One more thing worth saying plainly: selection is not fully predictable. Outputs vary by prompt wording, location, time and the engine's own updates. Anyone selling a guaranteed recipe for being cited is overstating what's known. What you can do is raise the odds and measure the results.

What to change in your AEO and GEO practice

The news doesn't ask for a new strategy. It asks you to stop leaning on one metric and to tighten a few habits. Here's a sequence that works for a small team.

Split your reporting. Separate queries and pages that trigger AI answers from those that don't. Even a rough split, built from a tracked keyword set and manual spot checks, beats a blended average.
Stop reporting average position for question-style queries. Report clicks, click-through rate and landing-page sessions instead. Keep position for classic, list-style results where it still means what it used to.
Rewrite your top pages for passages. Take your ten most important informational pages. For each key question they answer, make sure one short section answers it completely, in plain words, with the entity named.
Fix contradictions. Audit definitions, prices, dates and specifications across the site. Pick one canonical statement and make every page match it.
Add a brand-and-citation check. For a fixed set of prompts, record whether you're mentioned, whether you're linked, and what the answer says about you. Repeat monthly.
Tie cited pages to outcomes. For every page you know is cited, track what happens after the visit: sign-ups, demo requests, purchases, or whatever your conversion is.

Step 6 is the one teams skip, and it's the one that answers the question leadership will ask: is any of this worth the effort?

On the content side, resist the urge to turn every article into a wall of question headings. Use the natural phrasing of your reader's question as the heading where it fits, answer it directly, and then add the depth a human reader wants: a worked example, a caveat, a decision they'll have to make. Engines cite the direct answer. Humans stay for the rest.

How to measure AI search by outcomes

If position is unreliable inside AI blocks, what replaces it? A short set of signals, each with a clear owner. Here's a measurement plan you can adapt.

Question	Signal	Where it lives	Owner	Cadence
Are we shown in AI answers?	Mention and citation rate across a fixed prompt set	Manual log or an AI-search visibility tool	SEO lead	Monthly
Do people click through?	Clicks and click-through rate on question-style queries	Search Console	SEO lead	Weekly
Do they arrive and stay?	Sessions and engaged sessions on cited pages	Web analytics	Analytics lead	Weekly
Does it pay off?	Conversions and assisted conversions from those pages	Web analytics and CRM	Growth lead	Monthly
Is the tracking trustworthy?	Event and parameter coverage on landing pages	Tracking specification	Analytics lead	Quarterly

The last row deserves attention. Outcome measurement is only as good as the events behind it. If conversions on your informational pages are tracked inconsistently, the whole plan collapses into guesswork. A written tracking specification that names every event, parameter and owner is the unglamorous foundation, and governance tooling such as OmniSpec exists for keeping that specification honest as sites change.

For the visibility side, where you need to see how your brand appears across AI answers rather than in a Search Console export, the MediaPilot page covers media and AI-search visibility. Whatever tool you choose, check its documentation for how it samples prompts and engines, because sampling choices change what you can conclude.

A few practical notes on making the numbers hold up:

Annotate the date your AI-triggering queries changed. Without it, a quarter-over-quarter comparison will mislead.
Compare like with like. Don't set this quarter's cited pages against last quarter's whole site.
Watch for the divergence. Position up, clicks flat, conversions flat: that pattern is your prompt to investigate, not to celebrate.
Accept some noise. Small prompt sets wobble from month to month. Look for sustained direction across several months.

What to ignore, and what not to do

Not every reaction to this news is useful. Some are actively harmful.

Don't discard Search Console. Clicks, impressions, queries and pages are still valuable, and classic results still drive a large amount of traffic for most sites. The issue is one column of one report, and only for queries where an AI block appears.

Don't panic-rewrite the site. One measurement quirk isn't a reason to restructure everything. Fix your most important pages first and let the data tell you whether the effort mattered.

Don't treat citation as the goal. A citation with no visit and no brand recall is a vanity metric with better branding. Some queries will always be answered in full on the results page, and being cited there may still build familiarity. But that's a hypothesis to test, not a result to report.

Don't chase every prompt variation. You can't monitor every way someone might phrase a question to an assistant. Pick a fixed set that mirrors your real customer questions, keep it stable, and resist expanding it every time someone has a new idea.

Don't build markup tricks for engines. Structured data helps machines understand a page, and that's a fine reason to use it. It's not a switch that forces citation. Anyone promising otherwise is guessing.

Don't copy a competitor's dashboard. Their goals, funnel and audience differ from yours. A dashboard built to impress a board is a different object from one built to make decisions.

There's a temptation, when a metric turns out to be misleading, to swing to distrust of all metrics. That's the wrong lesson. The right one is narrower: choose measures that sit close to what you actually want, and be suspicious of any number that can improve while the business doesn't.

Where this is likely to go

Expect the reporting to change. Google has reworked Search Console before, and AI features are new enough that the definitions may be refined, split out or annotated. If a dedicated view for AI-generated results appears, it will be worth adopting. Until then, the sensible position is to hold your position figures loosely and your outcome figures firmly.

The teams best placed for whatever comes next are the ones who've already stopped treating rank as the finish line. They know which pages are cited, what those visits are worth, and whether their tracking would survive scrutiny. That's operational work, the kind that sits between SEO, analytics and content, and it's the reason AI search is turning into a marketing operations problem as much as a search one.

If you're an analytics lead, the immediate job is small: pull your question-style queries, compare position against clicks over the last two quarters, and see whether the gap has opened. If it has, you have your answer, and your next report should lead with outcomes.

Frequently asked questions

What is block flattening in Search Console?

According to Search Engine Journal, it's how Search Console treats an AI Overview: the block is handled as a single unit, so the position attaches to the block rather than to each link's prominence inside it. The result is that a modest appearance within the block can be reported like a top ranking. Check Google's Search Console documentation for the current definition of position.

Should I stop using average position for AEO and GEO reporting?

Stop using it for question-style queries where an AI Overview is likely to appear. Keep it for classic list-style results, where it still describes rank. Whatever you report, put clicks, sessions and conversions next to it so a flattering position can't stand alone.

How do I know if my pages are cited in AI Overviews?

Build a fixed set of prompts that reflect real customer questions, then check them on a regular schedule and log whether you're mentioned and linked. You can do this by hand or with a visibility tool. Confirm in the tool's documentation how it samples, because that shapes how far you can trust the result.

Is being cited in an AI Overview worth anything if nobody clicks?

Possibly, but you have to test it rather than assume it. A citation can build familiarity, and some visitors do click through, but reports of falling clicks in AI results mean you shouldn't count on traffic. Track branded search, direct visits and conversions on cited pages, and judge the value from those.

Sources

https://www.searchenginejournal.com/search-console-uses-block-flattening-for-aios-forget-position-focus-on-outcomes/589582/
https://www.seroundtable.com/google-ai-overviews-more-external-links-42135.html
https://developers.google.com/search/blog