Core Web Vitals (LCP/INP/CLS)

1. What Core Web Vitals are

Core Web Vitals (CWV) is Google’s real-user UX measurement set, introduced in 2020 and rolled into Google Search ranking in 2021. The current three metrics are Largest Contentful Paint (LCP) for loading, Interaction to Next Paint (INP) for responsiveness, and Cumulative Layout Shift (CLS) for visual stability. Google’s own definition: “Core Web Vitals are the subset of Web Vitals that apply to all web pages, should be measured by all site owners, and will be surfaced across all Google tools” (Web Vitals overview).

The signal is Google’s, not the AI ecosystem’s. Field data comes from the Chrome UX Report (CrUX) — Chrome users who opted in via sync. There is no AI-engine equivalent that independently weighs your LCP or INP. CWV’s effect on AI-answer visibility therefore depends on whether a given engine reuses the host search system that uses CWV.

Three nearby concepts get conflated in practitioner writing. They are worth separating before any further claim is made:

The third row is where the most common error starts: improving CWV does not improve a bot’s chance of completing your page fetch in time. That is a different problem with a different fix — see SSR for AI Crawlers. The rest of this page traces CWV’s actual, bounded effect on the engines that produce AI answers.

2. The three current metrics — LCP, INP, CLS

Thresholds are the current “Good” values per Google’s stable lifecycle; “Needs Improvement” and “Poor” bands are listed on the per-metric web.dev pages and revised occasionally. For ranking and AI-Overview eligibility purposes, the operational question is whether your 75th-percentile field value is inside the Good band — extra optimization beyond Good has no documented additional ranking weight (§4).

History footnote: INP replaced FID as a Core Web Vital on March 12, 2024 (web.dev announcement, Jan 31, 2024; original Google Search blog, May 2023). Older audit tools and articles still showing FID (First Input Delay) thresholds are historical; INP is the current metric. INP captures the full interaction-to-frame latency for all interactions, whereas FID captured only the input delay of the first interaction — INP is the stricter, more representative measurement.

3. Per-engine effect on AI visibility — the canonical table

The single most-quoted asset on this page. CWV’s effect on an AI engine is determined by whether that engine reuses a host search system that already uses CWV — and the engines split cleanly into three groups.

The load-bearing read of this table: CWV affects AI visibility where the AI surface inherits a search ranking that already uses CWV. It is not a universal AI-engine signal, and it is not bidirectional — “optimizing CWV for AI” is at best a Google-only investment.

4. Why the effect is indirect — the mechanism, named

The causal chain on Google’s surface, end to end:

CWV → Google Search ranking signal → page is indexed and ranked for the underlying query → page enters the candidate pool AI Overviews draws from → AIO selects a subset of candidates to ground and cite → your page is (or is not) one of the supporting links.

CWV is one Google ranking signal among many, weighted alongside content quality, E-E-A-T, link signals, and topical relevance. Google’s own scoping sentence: “Beyond Core Web Vitals, other page experience aspects don’t directly help your website rank higher in search results” (page experience) — which also implies the converse: CWV does contribute to ranking, but other page-experience attributes do not earn ranking weight beyond it. The signal is real but bounded.

What CWV is not, even on Google’s surface: an input to the AIO selection step itself. AIO’s selection from the candidate pool is governed by the same quality systems that rank classic results, plus AI-specific liftability factors. The signals that drive whether AIO actually picks your page out of the candidate pool — once you are in it — are citability (chunk shape, quotable claims), E-E-A-T, entity recognition, and Knowledge Graph presence. GEO did not promote CWV into one of those load-bearing levers.

On the other engines, the chain simply doesn’t exist. Perplexity, ChatGPT Search, and Claude maintain their own retrieval indices and pick their own citations. No equivalent of “Bing’s page load time signal becomes a Perplexity citation factor” has been documented by any of them.

5. Core Web Vitals ≠ AI crawler perf — the disambiguation

The second load-bearing distinction on this page, and the source of the most frequent practitioner confusion. CWV measures real users in Chrome via CrUX field data; AI crawlers are not in that dataset, because CrUX eligibility requires a signed-in Chrome user with sync history enabled (CrUX methodology).

The operational test for AI performance questions is one question, not two: can the bot fetch and parse your HTML in the time it allows itself? That answer is independent of your CWV. A site can have green CWV across the board and still ship a CSR shell that Perplexity’s fetcher can’t read; a site can have failing INP for human users while serving a perfectly bot-readable SSR HTML payload on first byte. The two layers solve genuinely different problems and the fixes do not interchange.

A small but useful corollary: when a perf question comes up in a GEO audit, ask which layer it lives at before reaching for an answer. CWV problems get web-performance work; bot-fetch problems get rendering and infrastructure work — and you can have either, both, or neither.

6. Anti-patterns — the three failure modes

Each row below describes a mistake that sounds right on first telling and fails for a different reason. The first two are the canonical mirror failures — over-investment and over-dismissal of the same fact. The third is a category error in measurement.

The line that closes the section: CWV is not a GEO signal. It is a Google ranking signal whose AI relevance is bounded by which engines reuse Google’s ranking.

7. Why this matters for GEO + how to act

CWV is a keep-this signal under GEO, not a new lever. It still matters where Google’s ranking matters — and the bar to clear is the published Good threshold, not aggressive over-optimization. Keep is not the same as lever: a signal you keep maintaining at baseline is not the place to spend marginal GEO effort.

References

Primary — Google web performance and AI documentation:

• Google web.dev — Web Vitals (overview) · INP — Interaction to Next Paint
• Google Search Central — Understanding page experience in Google Search results · AI features and your website · Optimizing for generative AI features
• Google — INP becomes a Core Web Vital, March 12, 2024 (web.dev) · Introducing INP to Core Web Vitals (Search Central, May 2023)

Primary — Other engines + data sources:

• Microsoft Bing — Bing Webmaster Guidelines (page load time named among ranking factors; CWV not specifically named)
• Google Chrome for Developers — Chrome UX Report (CrUX) methodology (real opt-in Chrome users only — bots not in the dataset)

Secondary — Independent measurement and coverage:

• HTTP Archive — Web Almanac 2024 — Performance chapter (industry CWV pass rates)
• Google web.dev — Rakuten 24 case study (CWV ↔ business KPI correlation, one published example)
• Search Engine Land — Google Core Web Vitals to add Interaction to Next Paint on March 12 (independent coverage of the FID → INP transition)