A website’s performance has gone from an infrastructure detail to one of the variables that most directly influence conversion, rankings and media cost. This article gathers what data from primary sources - Google, Deloitte, Portent and the case studies published on web.dev - shows about that impact, with the correct, up-to-date numbers.
Key numbers
- Only 48% of mobile sites pass the three Core Web Vitals metrics simultaneously (Web Almanac 2025, CrUX field data). Good performance is still a minority - and, for that reason, a differentiator.
- A 0.1-second improvement in mobile load time increased retail conversions by 8.4% (Google/Deloitte, “Milliseconds Make Millions”, 2020).
- Sites that load in 1 second convert 2.5x more than sites that take 5 seconds (Portent, 2022).
- Real 2025 cases (web.dev): Ray-Ban +101% mobile conversion, QuintoAndar +36% conversions, Fotocasa +27% leads.
- Google’s official thresholds for “Good”: LCP ≤ 2.5s, INP < 200ms, CLS < 0.1 (Google Search Central).
The quality standard: Core Web Vitals
Google uses Core Web Vitals (CWV) as the main yardstick for page experience. They are three field metrics, measured from real users (Google Search Central - Core Web Vitals):
These thresholds are the official ones and remain in force - there was no threshold change in 2025 or 2026. The recent structural news was something else: on March 12, 2024, INP definitively replaced FID as the responsiveness metric, measuring every interaction in the session, not just the first (web.dev - INP is a Core Web Vital). It is a more demanding yardstick, because it captures the user’s entire experience (web.dev - INP).
The figure that turns this into a competitive opportunity: according to the Web Almanac 2025 (HTTP Archive, on CrUX field data), only 48% of mobile sitesreach “Good” on all three metrics at the same time. Less than half. Since more than half of global web traffic is already mobile (Statcounter, 2026), the field where the majority fails is exactly the one that matters most.
The math of conversion
The relationship between speed and revenue is consistent across serious, independent studies.
The “Milliseconds Make Millions”study, conducted by Deloitte Digital at Google’s request in 2020, analyzed 30 million sessions across dozens of sites. The result: an improvement of just 0.1 second in mobile load time lifted retail conversions by 8.4% and average order value by 9.2% (web.dev - Milliseconds Make Millions).
At the other end of the yardstick, the Portent (2022) research, covering more than 27,000 landing pages, measured the effect of slowness: e-commerce sites that load in 1 second convert 2.5 times more than sites that take 5 seconds. In B2B lead generation, the difference reaches 3x (Portent - Site Speed Is Hurting Everyone’s Revenue).
Classic industry studies were already pointing to the same pattern nearly two decades ago: the Aberdeen Group (2008) reported a 7% drop in conversions and 16% in satisfaction for every 1 second of delay. Dated numbers, but ones that recent research confirms in direction and magnitude.
The practical takeaway is direct: the traffic you already pay to attract is wasted in the fraction of a second between the click and the page being ready. It is revenue that never shows up in Analytics, because it only measures those who waited for the load.
Performance, SEO and AI search
In organic rankings, speed works as a page experience signal - a tiebreaker. Google itself is clear: useful content weighs more, but among pages of equivalent relevance, the one that delivers a better experience gets the edge (Google - Page Experience). There is no formula of “X positions per second”; there is a quality floor that, when you don’t meet it, leaves you behind those who do.
In AI search, there are still no official published thresholds for speed, timeout or crawling criteria from the main agents. But log analyses and practical tests already point in an important direction: slow, unstable or hard-to-process pages create more friction for any system trying to crawl, interpret and retrieve content. That’s why performance and semantic structure don’t guarantee AI citation, but they lower the technical barriers to the page being read correctly.
Efficiency in Google Ads
In paid media, landing page performance influences the landing page experience, one of the components evaluated in the Google Ads Quality Score. Google doesn’t publish a fixed table of discounts or penalties per Quality Score point, but it confirms that higher-quality ads and pages tend to perform better, including better positions and lower costs. In practice, a slow or unstable landing page can reduce media efficiency, wasting part of the budget before the conversion even happens.
High-impact techniques
Reaching these standards in 2026 takes engineering, not a plugin:
- Edge computing. Processing requests close to the user (Cloudflare Workers, Vercel Edge) reduces Time to First Byte (TTFB) and speeds up LCP.
- Next-generation images. Converting to AVIF (about 50% smaller than JPEG at equivalent quality) and WebP (25-35% smaller) cuts weight without visible quality loss (Cloudinary - AVIF vs WebP).
- JavaScript management. Breaking up long tasks with
scheduler.yield()frees the main thread and directly improves INP (Chrome - scheduler.yield()). - Prerendering. The Speculation Rules API preloads pages the user is likely to visit next, making the following navigation practically instantaneous (Chrome - Prerender pages) - exactly what Ray-Ban used (below).
Case studies: measured results
The most solid numbers don’t come from blogs - they come from companies that measured before and after and published on web.dev, Chrome’s official documentation:
- QuintoAndar. Reduced INP by 80% (from ~1,000ms to ~216ms on mobile) by optimizing JavaScript and removing third-party pixels. Result: +36% conversions year over year.
- Ray-Ban (EssilorLuxottica). Implemented prerendering via Speculation Rules. Mobile LCP dropped 43% and mobile conversion rose 101% (on desktop, +156%).
- Fotocasa. Improved filter responsiveness and eliminated unnecessary re-renders, improving INP and generating +27% in contact and phone leads.
- T-Mobile. A data-driven approach cut user complaints by 20% (and slowness-specific complaints by 34%) and lifted the conversion rate of purchase-intent visits by 60%.
Conclusion
In 2026, web performance is not an infrastructure cost - it is a revenue strategy. Speed protects the ad investment, sustains visibility in traditional search and in AI search, and is what separates, in fractions of a second, conversion from abandonment. And since less than half of all sites meet the standard, doing this well is still a scarce competitive advantage.
At Inodus, 90+ performance on mobile and desktop is a starting requirement, not phase 2. Want to know your site’s speed score today? Run the free online audit.
Frequently asked questions
What are Core Web Vitals and what are the thresholds in 2026?+
They are Google's three page experience metrics: LCP (loading) with a 2.5s threshold, INP (responsiveness) below 200ms and CLS (visual stability) below 0.1. These thresholds remain in force in 2026 (Google Search Central).
Does performance improve Google rankings?+
It works as a page experience signal and a tiebreaker between content of similar relevance. It doesn't replace useful content, but, when absent, it leaves the site behind equivalent competitors (Google - Page Experience).
How much does speed affect conversion?+
Studies from primary sources show a direct impact: +8.4% retail conversion per 0.1s faster (Deloitte/Google, 2020) and 2.5x more conversion on 1s sites vs. 5s (Portent, 2022).
Why does a slow site make Google Ads more expensive?+
Because the landing page experience feeds into the Quality Score. A high score reduces CPC; a slow landing page drags the score down and makes the click more expensive (Google Ads - Quality Score).
Does performance matter for appearing in AIs like ChatGPT?+
Yes, indirectly: AI crawlers can abandon slow pages before processing them. A fast, well-structured page has a better chance of being read and cited.
How we interpret the sources in this article
This content distinguishes four types of evidence: official documentation, case studies published by recognized sources, proprietary market studies, and emerging research or analyses. Official data is treated as a normative reference. Proprietary studies and benchmarks are used as a signal of direction, not as a universal rule. Academic research and log analyses about AI are presented as evolving technical evidence, especially while there are no public thresholds defined by the vendors.
Methodology and sources
Data from primary sources, with a link on every citation in the text: Google Search Central and web.devofficial (Core Web Vitals thresholds and case studies); HTTP Archive - Web Almanac 2025research/analysis (CWV pass rate); Deloitte Digital/Google - “Milliseconds Make Millions”, 2020case study; Portent, 2022proprietary; WordStreamproprietary (CPC benchmarks). Dated statistics (2008) are presented as historical reference, not as current data.