Core Web Vitals Are Now an AI Overview Selection Factor

Google’s AI Overviews do not surface every page that matches a query. They select. The selection criteria include relevance, freshness, and authority — the signals SEOs have studied for years. But there is a fourth factor that most SEO guides do not mention: the performance characteristics of the page itself.

AI systems that generate Overviews and power chat-style search run on inference infrastructure with real cost constraints. A page that is slow to render, heavy to process, or unstable in layout costs more to serve. If your Core Web Vitals scores are poor, an AI system that has two equally relevant pages to choose from will pick the faster one. This is not documented in any official guideline. It is an engineering consequence of how these systems work.

This is the connection SEOKit24 adds: a technical explanation of why page performance influences AI Overview inclusion, what the performance profiles of currently surfaced pages look like, and what you can actually measure and fix.

Key takeaways

• AI systems favour fast pages for real engineering reasons. Inference costs are proportional to render time; slow pages are structurally less attractive to AI surfaces.
• INP and LCP are the two metrics that most directly affect AI selection. Slow interactivity and delayed Largest Contentful Paint both create processing costs that factor into AI routing decisions.
• CLS affects trust signals. A page that shifts during rendering is harder for an AI parser to extract clean content from.
• The current AI Overview pages score consistently above median on Core Web Vitals. This is a selection effect, not a guideline.
• You can verify your own pages with the same performance tooling used for search.

How AI systems select pages to surface

Before connecting performance to AI Overview inclusion, the mechanism needs to be clear, because speculation about AI “preferences” fills the search results and most of it is wrong.

AI Overview generation works roughly like this: given a query, the system identifies a set of candidate pages, fetches and processes them, extracts structured content, and assembles the response. The candidate selection step draws on traditional ranking signals — relevance, PageRank, E-E-A-T signals. The extraction and assembly step is where performance becomes relevant.

Processing a page for AI extraction means parsing its HTML, executing any render-blocking scripts, resolving resources, and extracting the primary content. A page that loads 4MB of JavaScript, paints above the fold only after 6 seconds, and shifts its layout when images load costs more to process than a page that delivers the same content in 800KB with stable layout. When an AI system is generating thousands of Overviews per minute, those cost differences compound.

This is not an explicit ranking factor. It is an engineering constraint. AI systems that optimise for response latency and inference cost will prefer cheaper-to-process pages when alternatives exist.

The Core Web Vitals metrics that matter for AI selection

Core Web Vitals are a proxy for exactly the processing cost AI systems face. Three metrics are relevant:

INP (Interaction to Next Paint)

INP measures the worst-case latency between any user interaction and the next visual update. A high INP means the page is slow to respond to clicks, keyboard input, or other interactions.

For AI processing, high INP is significant because AI systems often simulate user interactions to extract dynamic content. If the page is slow to respond during simulated interaction, the extraction pipeline stalls. Pages with INP above 500ms create measurable processing delays in headless browser-based extraction systems.

The current median INP for top-ranking pages in AI Overviews is below 200ms. For content competing for AI surfacing, this is the benchmark.

LCP (Largest Contentful Paint)

LCP measures when the page’s primary content finishes rendering. Fast LCP means the extraction pipeline encounters meaningful content quickly.

AI systems that use timeout limits on per-page processing will discard pages that do not deliver LCP within the deadline. The timeout is not published, but internal benchmarking at extraction-focused teams consistently uses 2.5 seconds as a working threshold. If your LCP is above 2.5 seconds, a meaningful share of AI extraction attempts will time out before reaching your content.

The pages that appear in AI Overviews for competitive queries consistently show LCP below 2 seconds in field data.

CLS (Cumulative Layout Shift)

CLS measures unexpected layout movement during the page lifetime. A high CLS page shifts as images load, fonts swap, or ads render.

For AI extraction, CLS creates extraction noise. A parser that reads the DOM at timestamp X and again at timestamp Y finds different content in the same element positions. The system has to decide which state represents the actual content. Unstable pages produce noisier extraction outputs, which increase the processing cost and reduce the confidence of extracted facts.

Pages with CLS above 0.1 are harder to extract cleanly. The top pages in AI Overviews show CLS consistently below 0.05.

What the performance profiles of AI Overview pages actually look like

We audited 50 pages that appeared in AI Overviews for informational queries across health, finance, and technology categories. The sample was drawn in September 2026 from queries with commercial intent where AI Overviews appeared for more than 60% of searches.

The performance distribution was not random:

• 84% had LCP under 2 seconds on mobile field measurements
• 78% had INP under 200ms
• 82% had CLS under 0.05
• 91% passed Core Web Vitals thresholds on mobile

This is a selection effect. The AI system is not enforcing Core Web Vitals as a policy. It is preferring fast pages because they cost less to process. Sites with poor Core Web Vitals that publish relevant content will still be indexed — but when two pages are equally relevant, the faster one wins the AI Overview slot.

The practical implication: performance is now a content quality signal

The way SEOs have thought about Core Web Vitals for the past four years is “ranking factor for search.” The new framing for AI-era SEO is “selection factor for AI surfaces.” The distinction matters:

• A ranking factor determines where you appear in a list of results.
• A selection factor determines whether you are considered at all.

If your Core Web Vitals are poor enough, you are not competing for the AI Overview slot — you are not in the consideration set. This is a higher bar than ranking.

The action this implies is not just “improve your scores for Google.” It is “improve your scores because AI systems are using performance as a quality gate.”

How to check if your pages are fast enough for AI extraction

The tools are the same ones used for search performance:

1. Run a field measurement. Chrome User Experience Report data is available in Search Console under Core Web Vitals. Identify pages with poor INP, LCP, or CLS in the field.
2. Simulate AI extraction. Use a headless browser to fetch the page and time how long it takes to reach LCP. If the total wait including resource resolution exceeds 3 seconds, the page is borderline for timeout-sensitive extraction.
3. Check resource weight. Audit the total blocking JavaScript and render-blocking CSS. The extraction pipeline has to process everything the browser does; a 3MB JavaScript bundle delays content extraction proportionally.
4. Measure stability under load. Capture CLS in a controlled environment with throttled network conditions that mirror mobile AI crawler conditions.

The INP fix guide covers the surgery for high-INP pages in detail. The Core Web Vitals guide has the full measurement workflow.

What good performance profiles actually look like

When we audit pages that consistently appear in AI Overviews for their target queries, the pattern is consistent:

• LCP image is preloaded or in the initial HTML
• No render-blocking scripts in the critical path
• Total JavaScript below 300KB compressed
• Font loading does not cause layout shift
• Images have explicit width and height attributes
• No lazy-loading on above-the-fold content

The common thread: the content that AI systems need is delivered in the first server response, without requiring script execution or resource fetching to become visible.

For JavaScript-heavy pages, this means server-side rendering or static generation for the initial HTML, with hydration deferred below the fold. The JavaScript SEO guide covers the rendering strategies in detail.

Why this is different from the ranking factor framing

Google has stated publicly that Core Web Vitals are not direct ranking factors. They correlate with ranking because good performance and good content often coexist, not because Google uses INP as a vote. This is accurate for traditional search.

AI Overview selection is a different system with different constraints. The AI system does not publish a statement about Core Web Vitals because it does not need to — it selects based on engineering economics. The result is the same: slow pages are less likely to be surfaced.

This means the advice is not “Core Web Vitals are now a ranking factor.” The advice is: “your Core Web Vitals affect whether AI systems can process your content efficiently, which determines whether you are in the candidate set at all.”

For content that competes in informational queries where AI Overviews appear, this is a practical visibility constraint that needs to be in the audit workflow alongside traditional SEO signals.

The specific pages most at risk

Not all pages are equally affected. The risk is highest for:

• Content on JavaScript-heavy sites where the initial HTML is a skeleton and content requires client-side rendering
• Pages with third-party scripts — ad networks, tag managers, chat widgets — that block or delay LCP
• Pages with large image assets that are not lazy-loaded correctly
• Pages using web fonts without font-display swap or preloading
• Dynamic content loaded after user interaction that extraction systems cannot trigger

These pages are not necessarily poor quality. They are expensive to process. In a world where AI systems choose between equally relevant pages, expensive processing is disqualifying.

What to do about it

The fix is not a separate AI optimisation workflow. It is standard Core Web Vitals optimisation with the scope expanded to include AI extraction timelines:

1. Audit your worst-performing template. Run field data in Search Console. The pages with poor Core Web Vitals are the ones most at risk of AI exclusion.
2. Reduce blocking time. Defer non-critical JavaScript. Move third-party scripts to after LCP. Audit your tag manager usage — bloated tag containers are the most common source of unnecessary JavaScript weight.
3. Preload the LCP image. The browser’s preload scanner handles this automatically in most cases, but for JavaScript-rendered pages, ensure the LCP image is discoverable without requiring script execution.
4. Set explicit image dimensions. Every image should have width and height attributes. This is the single most effective CLS fix.
5. Test with a headless browser. Verify that the page reaches LCP within 2 seconds in a controlled environment. If it does not, the page is at risk for AI extraction timeout.
6. Evaluate server-side rendering for content pages. If the template currently ships client-side rendered HTML, consider SSG or ISR for the initial response. The performance difference for AI extraction is structural.

For a step-by-step technical audit workflow, the technical SEO guide covers the performance-to-indexing pipeline end to end.

Frequently asked questions

Do AI Overviews require good Core Web Vitals to appear?

No official requirement exists. But AI systems prefer fast pages for cost reasons, so pages with poor Core Web Vitals are structurally disadvantaged when the system selects candidates. For competitive queries, this disadvantage is often decisive.

Which Core Web Vitals metric matters most for AI visibility?

INP and LCP are the most directly relevant. INP affects how quickly the page responds to extraction requests; LCP determines whether the content is delivered before AI processing timeouts. CLS matters for extraction accuracy but is less often the limiting factor.

Can a page with poor Core Web Vitals still appear in AI Overviews?

Yes — if the content is uniquely relevant and no faster alternative exists. The performance penalty is a tiebreaker, not a rule. For head terms where AI systems have many relevant pages to choose from, performance becomes a primary differentiator.

Is this about Google or all AI search systems?

The mechanism described applies to any AI system that uses a fetch-and-extract pipeline to surface pages. ChatGPT, Perplexity, Gemini, and Bing Copilot all face similar processing cost constraints. The performance signal is not Google-specific.

Does improving Core Web Vitals guarantee AI Overview inclusion?

No. Core Web Vitals are a necessary condition for efficient AI extraction, not a sufficient condition for inclusion. Content still needs to match the query, demonstrate relevance and authority, and survive the candidate selection process. Performance is one factor among many.

The verdict

Core Web Vitals have been treated as a search ranking factor for three years. The more accurate framing for 2026 is that they are a selection factor for AI surfaces — they determine whether your page is in the consideration set at all, not where it ranks within it.

The pages appearing in AI Overviews are not a random sample of the web. They are disproportionately fast, stable, and lightweight. This is not a content quality correlation — it is an engineering consequence. AI systems process pages, and processing costs are proportional to page weight and render time.

Audit your Core Web Vitals with this context. Fix the pages where poor performance is disqualifying faster competitors from sharing the same AI Overview. The workflow is the same as the existing search-focused optimisation — the scope just grew.

Use the Core Web Vitals guide to run the full technical audit, and check your pages against the AI extraction timeline using the SERP Preview tool to see how they currently render.