Every site has problems. The question is which ones to fix first — and most commercial SEO dashboards, while useful in other ways, are of little help in answering that question.
Core Web Vitals — Google's three metrics for real-world user experience — measure how fast your main content loads (Largest Contentful Paint, or LCP), how quickly the page responds when someone taps or clicks (Interaction to Next Paint, or INP), and how much the layout jumps around during loading (Cumulative Layout Shift, or CLS). Google considers them a ranking signal: sites that fail these metrics can be penalized in search results, while sites that pass them meet what Google considers a baseline for usability. The thresholds are specific: LCP should be under 2.5 seconds, INP under 200 milliseconds, and CLS under 0.1.
These numbers feed into the performance score you see in tools like Google PageSpeed Insights — the big number at the top, scored out of 100, that most site owners look at first. But here's the thing: that performance score is a lab simulation. Core Web Vitals, by contrast, are measured from real users via the Chrome User Experience Report. The two don't always tell the same story. And when they diverge, most people pay attention to the wrong one.
A Shopify-based land sales company recently approached &listen... Their SEO audit score was a decent 85 out of 100, but their PageSpeed performance score was only 52. And buried in their Core Web Vitals report was a Cumulative Layout Shift score of 0.8 — more than three times over Google's "poor" threshold of 0.25.
Ignoring the SEO score (for now), and noting the low performance score, the team went straight for the catastrophic CLS.
The reasoning behind this call applies to almost any site trying to figure out where to put its time and money.
What causes a high CLS score on Shopify?
Three things were producing the layout shift, and all three culprits were baked into the Shopify site's theme liquid. This meant that every single page was affected.
A page builder plugin was running a layout "optimizer" that fired after the page had already started rendering. Instead of preventing instability, it was actually causing the CLS since it was rearranging elements after visitors could already see them.
The sticky navigation header was also causing problems, recalculating its own height on each load. It would briefly appear at one size, then snap to another, pushing all the content below it downward.
Additionally, the hero section used a YouTube iframe with no predefined dimensions. Initially, the browser allocated zero space for it, but then suddenly expanded the container when the content arrived — shoving everything else down the page.
None of these were visible in the site's SEO score. They didn't affect title tags or meta descriptions or heading structure. But there was an unexplained plunge in organic traffic and the CLS score meant the site was borderline unusable on first load.
Why fixing CLS matters more than improving your SEO score
An SEO score of 85 means the basics are in place. Pushing it to 90 or 95 or 100 would involve adjusting meta descriptions and declaring canonical tags. That work is obviously important too, but in this case it wasn't the first priority.
A CLS of 0.8, on the other hand, means the foundation of the site is unstable. And a slippery foundation doesn't just sit there passively: its negative effects cascade in both predictable and unpredictable ways.
It hurts your Google ranking. Core Web Vitals are a confirmed ranking signal. Google's own John Mueller has downplayed their weight — calling them "not giant factors" — but that assessment assumes you're in the normal range. At 0.8, you're not in a tiebreaker situation. You're in active-penalty territory. And because the causes here were theme-level, every page on the site was being assessed as "poor."
It drives visitors away. A page that jumps nearly a full screen height while loading is a page people leave. Higher bounce rates, lower time on site, fewer conversions. Google measures all of this. These behavioural signals feed back into ranking, compounding the direct CWV penalty.
It may block AI crawlers. This is the part that gets less attention but matters increasingly. AI crawlers — the bots that feed content into ChatGPT, Perplexity, Google's AI Overviews — operate on tight timeouts, typically 1 to 5 seconds. Research from Discovered Labs found that roughly 18% of pages larger than 1MB were abandoned mid-crawl by AI bots. A site whose infrastructure causes excessive layout recalculation and DOM thrashing doesn't just frustrate visitors. It can frustrate the systems deciding whether your content enters the AI index at all.
The problem goes deeper than that. Rather than simply crawling pages, AI systems extract structured representations of a site's architecture. That process assumes a stable layout. But when the DOM shifts during rendering, extraction becomes unreliable: headings may detach from paragraphs, sections may appear incomplete, and content may load too late to be captured at all.
This results not only in slower performance, but in a degraded AI signal — a page that cannot be consistently interpreted. This is the difference between AI visibility and audibility. A page can exist, load, and even rank well, and still fail to resolve into something an AI system can reliably use.
It shrinks your AI search candidate pool. AI search tools draw their source candidates primarily from existing search indices. If your site has been pushed down in organic rankings through a combination of CWV penalties, poor engagement signals, and crawler difficulties, you're not in the pool that AI systems select from. You can't be cited by a system that never sees you.
An SEO score of 85 triggers none of these cascades. A CLS of 0.8 triggers all of them.
Do Core Web Vitals affect AI search results?
This is a fair question, and the honest answer is: not directly, but the indirect effects are real at the extremes.
Google's documentation on AI features says flatly that there are "no additional requirements to appear in AI Overviews." Core Web Vitals aren't mentioned. Danny Sullivan has said "good SEO is good GEO" without ever singling out page performance. No AI search platform — Google, OpenAI, Microsoft, Perplexity — has confirmed CWV as a direct factor in deciding which sources to cite in generative answers.
The largest empirical study on the question, published on Search Engine Land in January 2026, analysed over 107,000 pages appearing in Google AI Overviews. Dan Taylor of SALT.agency found only weak correlations between CWV and AI visibility — LCP at -0.12 to -0.18, CLS at -0.05 to -0.09. His conclusion was blunt: AI systems choose sources based on content clarity, alignment with established knowledge, and user intent — not loading speed.
But — and this is the important nuance — those weak correlations were driven almost entirely by extreme outliers. Pages so broken that their performance failures cascaded into engagement signals and ranking position. A CLS of 0.8 is exactly that kind of outlier.
So the picture is: if your CWV is in the acceptable range, improving it further won't move the needle for AI search. If your CWV is catastrophically bad, fixing it removes a genuine barrier — not because AI systems care about your layout shift score, but because the downstream damage puts you outside the pool of content they draw from.
What ranking factors matter most for AI search visibility?
Since we're on the subject, it's worth noting what the research says does drive AI citation. The hierarchy looks quite different from what most SEO dashboards emphasize.
Brand authority is the strongest signal. An Ahrefs study of 75,000 brands found that web mentions of a brand correlated at 0.67 with AI Overview visibility — roughly three times stronger than backlinks.
Content quality and originality matter substantially. The Princeton/Georgia Tech paper on Generative Engine Optimisation found that adding statistics to content boosted AI visibility by 22–41%, and adding quotations from credible sources by 28–37%.
Content structure matters too. CXL research found that 55% of AI Overview citations come from the top 30% of a page's content. Pages with clean organisation and schema markup earn significantly more AI citations than equivalent content without it.
Freshness is surprisingly important. Ahrefs found AI-cited content is 25.7% fresher than typical search results. On ChatGPT specifically, 76.4% of the most-cited pages had been updated within 30 days.
CWV doesn't appear in this hierarchy. It sits below it, as plumbing. Essential when broken. Invisible when working.
How to prioritize SEO and GEO: a simple framework
The mistake most businesses make isn't doing the wrong optimisation. It's doing the right optimisation in the wrong order. Here's how we think about sequencing:
Our client's SEO score of 85 belonged in the third category. Their performance score of 52 was a symptom of the first. Their CLS of 0.8 was the cause. The dashboard made the SEO score feel urgent and the CLS feel obscure. The reality was the opposite.
How we fixed the CLS without changing the content
The fixes were targeted and technical. We neutralized the page builder's post-render optimizer through CSS overrides in the theme's base stylesheet. We gave the sticky header a fixed height declaration so the browser reserved space correctly from first paint. We added explicit width and height attributes to the hero iframe container.
None of this changed the site's content. None of it touched the SEO score. But it stabilised the site — for visitors, for search engines, and for the AI crawlers that increasingly determine whether a business gets heard.
The site went from a CLS of 0.8 to well within acceptable range. The performance score climbed. The SEO score stayed at 85. And for the first time, the foundation was actually solid enough for that score to mean something.
Most sites we audit don't have an optimisation problem. They have a structural problem they don't know exists. SEO scores don't surface it. Dashboards don't prioritize it. But it determines whether your content is even considered by AI systems.
If your CLS is in the "poor" range, your site is already compromised — whether your SEO score says 85 or 95.
Frequently asked questions
Not sure where your site stands?
We help organisations figure out what to fix first — and what AI actually hears when it listens for them.
Run a Quick Sound Check