Chapter 6
Load time, Core Web Vitals and the reality of crawling
Table of contents
Load time sounds like a topic that has been ticked off. Every store owner knows Core Web Vitals, everyone has run a PageSpeed test. What is missing from this discussion: AI crawlers evaluate something completely different from Google and something different from your human visitors. If you optimize for only one audience, you can still fail with the other.
Which two clocks are running at the same time?
The classic Core Web Vitals – Largest Contentful Paint, Interaction to Next Paint, Cumulative Layout Shift – measure how a person experiences the page in their browser. The thresholds have been stable for years:
LCP under 2.5 seconds
INP under 200 milliseconds – successor to the earlier FID metric since March 2024
CLS under 0.1
Each is measured at the 75th percentile of real Chrome users over a rolling 28-day window. These values remain relevant because Google takes them into account in its ranking systems. And because classic search visibility and AI visibility are closely connected, as Chapter 3 showed, they also remain relevant in an AI readiness context.
For AI crawlers, on the other hand, these three values hardly matter. As a rule, they do not execute JavaScript, as Chapter 2 showed, do not interact with the page and therefore experience neither INP nor CLS. What counts for them is a different metric: how quickly the server starts to respond at all – and whether the relevant content is already in the first HTML.
A store can perform brilliantly on classic Core Web Vitals and still be useless for AI crawlers.
Why does perfection not help here?
For AI crawlers, it is not about millisecond perfection. What matters is that the server responds reliably and quickly and that the relevant content is already in the delivered HTML. Whether CLS then drops from 0.05 to 0.02 makes no difference to a crawler that does not render and use the page like a person.
For the human user experience, things look different – and that is the real reason why this fine-tuning is still worthwhile. In an A/B test by Vodafone Italy, a 31 percent better LCP time on otherwise identical pages led to eight percent more sales. Remarkably, to achieve this improvement Vodafone moved, among other things, the rendering of a page element from the browser to the server – exactly the measure that also makes the difference for AI crawlers. That is not a GEO argument. It is a revenue argument, and Chapter 10 will come back to it.
What does this mean for a Shopware store?
Two concrete areas to check. First: server and hosting configuration. A store behind a cold cache or with sluggish database queries produces exactly the response times that deter crawlers – regardless of how well the frontend is built. Second, and this follows on from Chapter 2: if key content is only created by JavaScript, a crawler misses it either way, no matter how fast the server responds.
Server-side, complete HTML is therefore not only a question of visibility but also of speed: a server that delivers complete HTML immediately does not need an additional rendering round.
Our assessment
The common mistake is to read the combined PageSpeed score as a verdict. It is a weighted lab estimate – the three individual values beneath it are what actually counts, and they move independently of one another. For the AI perspective, there is a fourth metric that does not appear in any of these scores: pure server response time.
If you are in the green on Core Web Vitals, you have done an important part of the job for people. For machines, only one of two questions has been answered.
What you can check in ten minutes today
Measure the server response time of your most important product page directly in the browser's developer tools – not via the combined score.
Check how the store responds under real load, not just at six in the morning when it is idle.
Fetch the source code the way a crawler does, without executing JavaScript: is the actual product content already in the first HTML?
Frequently asked questions
Indirectly yes, directly hardly. Google takes them into account in ranking, and ranking is closely connected with AI visibility. AI crawlers themselves experience neither INP nor CLS – for them, server response time and complete HTML are what count.
No. The score is a weighted lab estimate. What matters are the three individual values from real user data – and for AI crawlers, additionally the server response time, which does not appear in any of these scores.
)