Core Web Vitals in 2025: What Still Matters (And What Does Not)
Core Web Vitals have evolved. Here is what still matters for SEO and AI search in 2025, what you can safely ignore, and how to measure the metrics that count.
Key takeaways
- INP replaced FID in March 2024 and is now the most important interactivity metric.
- LCP and CLS remain core, but the thresholds have not changed in 2025.
- AI crawlers care less about Core Web Vitals than Googlebot does, but they care about crawlability and rendering.
- Focus on INP under 200ms, LCP under 2.5s, and CLS under 0.1 — anything beyond is diminishing returns.
Table of contents
What changed in Core Web Vitals
The biggest change in the last two years was the March 2024 swap of FID (First Input Delay) for INP (Interaction to Next Paint). FID measured the delay before the first user interaction; INP measures the full interaction latency throughout the page lifecycle. INP is a stricter, more representative metric.
Beyond that, the thresholds have been stable. LCP (Largest Contentful Paint) targets 2.5 seconds or less. CLS (Cumulative Layout Shift) targets 0.1 or less. INP targets 200 milliseconds or less. Hit all three and you are in the green; miss one and your site is "needs improvement" in Google's eyes.
The 2025 landscape is mostly about enforcement, not new metrics. Google has tightened how Core Web Vitals factor into rankings, and AI crawlers have started using page-render signals to decide which pages to index deeply.
INP: the metric that replaced FID
INP measures the time from a user interaction (click, tap, keypress) to the next painted frame. It is a worst-case metric — your INP score is the slowlest interaction during the page lifecycle, with a few statistical adjustments.
The 200ms threshold is strict. Most sites miss it on mobile because of long JavaScript tasks. The fix is almost always the same: break up long tasks, defer non-critical JavaScript, and use requestIdleCallback for background work.
Common INP killers are third-party scripts (analytics, chat widgets, ad scripts), large React re-renders, and synchronous event handlers. Audit your site with Chrome DevTools' Performance panel and look for tasks longer than 50ms — those are your INP problems.
LCP: still the hardest to fix
LCP (Largest Contentful Paint) measures when the largest visible element renders. For most pages, that element is a hero image. For text-heavy pages, it is a large heading or paragraph block.
The 2.5-second threshold is achievable on desktop but hard on mobile, especially on slow networks. The fixes, in order of impact: preload your hero image, serve modern image formats (WebP, AVIF), use a CDN, and remove render-blocking CSS.
A common mistake is optimizing LCP on a fast connection and ignoring the long tail. Google measures the 75th percentile of page loads. If your site loads in 1.5s on fiber but 5s on 4G, your LCP score is 5s. Test on throttled connections.
CLS: the easiest to fix
CLS (Cumulative Layout Shift) measures unexpected layout changes. A CLS of 0.1 or less is the threshold, and unlike LCP and INP, CLS is almost entirely fixable with discipline rather than engineering.
The four common CLS causes: images without explicit width and height, ads and embeds without reserved space, web fonts that cause FOIT/FOUT, and dynamically injected content above the fold. Fix each one and your CLS will be near zero.
The fix for images is to always include width and height attributes. The fix for ads is to reserve space with min-height. The fix for fonts is to use font-display: optional or swap. The fix for injected content is to not inject above the fold.
Do AI crawlers care about Core Web Vitals?
AI crawlers (GPTBot, ClaudeBot, PerplexityBot) care about Core Web Vitals less directly than Googlebot does, but they are not indifferent. What they care about is whether the page renders cleanly within their crawl budget.
A page with a 6-second LCP may not finish rendering before the crawler's timeout, which means the LLM gets a partial page. A page with a high CLS may have shifted content by the time the LLM parses it, which means the LLM may extract the wrong text.
The practical advice: hit the green thresholds for Google, and you will be fine for AI crawlers. Do not over-optimize beyond the thresholds; the marginal SEO benefit is small and the AI-crawler benefit is negligible.
What to ignore in 2025
A lot of Core Web Vitals advice from 2020-2022 is now outdated or counterproductive. Three things you can safely stop worrying about.
- FID is gone. INP replaced it. Any advice that still references FID is stale.
- Preconnect and dns-prefetch hints have minimal impact on modern browsers. Use them but do not obsess.
- Critical CSS inlining has diminishing returns once you ship modern CSS frameworks. The complexity is not worth the 50ms savings.
seosights team
Editorial at seosights. We build the operating system for AI search — Three Sights, one unified engine.
Put this into action
Run a full Three Sights audit on your site. 8 AI agents, 90-day roadmap, 14-day free trial — no credit card.
Start free trial