Does page speed matter for AI search? Core Web Vitals explained
Core Web Vitals are a Google ranking input, but no AI assistant documents using them. Here is what page speed work actually helps crawlers reach and read your pages.
By Crawlito · 7 min read
In short
- Google says Core Web Vitals are used by its ranking systems, that there is no single page experience signal, and that it shows the most relevant content even when page experience is sub-par.
- Google's AI features page lists no speed or Core Web Vitals requirement for AI Overviews or AI Mode beyond being indexed and eligible to show with a snippet.
- OpenAI, Anthropic and Perplexity do not document using page speed or Core Web Vitals in their crawler documentation at the time of writing.
- Core Web Vitals are field metrics from real users, so a crawler does not produce or see them; what a crawler experiences is server response time, errors and how much HTML it has to download.
- Google lowers its crawl capacity limit when response times grow or a site returns 5xx or 429 responses, so a fast, stable server helps crawling even when it does not change rankings.
On this page
- What Google documents about Core Web Vitals
- Do Core Web Vitals affect AI Overviews and AI Mode?
- What AI assistants document about page speed
- Why crawlers care about server response, not LCP
- Speed work that actually helps crawlers
- How to measure speed the way a crawler sees it
- Frequently asked questions
- Check what a crawler receives
Page speed matters for AI search in a narrower way than most advice suggests. Google says Core Web Vitals are used by its ranking systems, and Google's AI Overviews and AI Mode are part of Search, but Google lists no extra speed requirement for those features. ChatGPT, Claude and Perplexity do not document using page speed or Core Web Vitals at all. What every crawler does experience is your server: how fast it answers, whether it returns errors, and how much HTML it sends.
So the useful question is not "will a better LCP get me cited?", which nobody can answer, but "is my site fast and stable enough that crawlers get the page every time?" That one you can measure and fix.
What Google documents about Core Web Vitals
Google's page experience documentation makes four points that are worth reading carefully:
- Core Web Vitals are used by Google's ranking systems.
- There is no single page experience signal. Core ranking systems look at a variety of signals.
- Relevance comes first. Google says it always seeks to show the most relevant content, even if the page experience is sub-par, and that page experience matters most when many pages are similarly helpful.
- Good scores in Search Console or third-party tools do not guarantee top rankings, and Google suggests that chasing perfect scores purely for SEO may not be the best use of your time.
The three Core Web Vitals, as defined on web.dev, are:
| Metric | Measures | Good |
|---|---|---|
| Largest Contentful Paint (LCP) | Loading | 2.5 seconds or less |
| Interaction to Next Paint (INP) | Responsiveness | 200 milliseconds or less |
| Cumulative Layout Shift (CLS) | Visual stability | 0.1 or less |
A page passes when it meets all three at the 75th percentile of real page loads, measured separately for mobile and desktop.
Do Core Web Vitals affect AI Overviews and AI Mode?
Google's AI features page says that to appear as a supporting link, a page must be indexed and eligible to be shown in Google Search with a snippet, and that there are no additional technical requirements. It does not mention Core Web Vitals or speed. It does list "providing a great page experience for users" among general best practices, which is the same advice Google gives for Search as a whole.
So the documented position is: AI Overviews and AI Mode draw on Google Search, Search ranking uses page experience signals among many others, and there is no separate speed threshold for the AI features. Our post on AI Overviews requirements covers the eligibility conditions that do apply.
What AI assistants document about page speed
OpenAI's crawler documentation, Anthropic's crawler help article and Perplexity's bots page describe their user agents and how robots.txt applies to them. None of them mentions Core Web Vitals, page speed, LCP or a fetch timeout at the time of writing.
That leaves two honest statements:
- No documented effect. There is no published evidence from these operators that a faster LCP changes whether a page is chosen or cited.
- No documented tolerance either. A crawler that gives up on a slow response is still a crawler that did not get your page. Undocumented timeouts exist in every HTTP client; they are just not published.
Be wary of claims that a particular score improves AI visibility. They are not backed by operator documentation, and AI search myths covers several similar claims.
Why crawlers care about server response, not LCP
Core Web Vitals are field metrics. They are collected from real Chrome users interacting with your page, which is why Lighthouse cannot measure INP without user input. A crawler does not scroll, click or wait for a hero image to paint. It sends a request and reads the response.
That makes a different set of numbers relevant to crawlers:
| What the crawler experiences | Related metric | Why it matters |
|---|---|---|
| Time until the response starts | Time to First Byte (TTFB) | A slow origin can exceed a crawler's timeout, and Google crawls less when responses slow down |
| Errors under load | Rate of 5xx and 429 responses | Google lowers its crawl capacity limit when it sees them; any crawler gets no content from an error |
| Size of the HTML | Uncompressed bytes | Googlebot reads only the first 2 MB; larger pages also take longer to fetch |
| Text present without JavaScript | Raw versus rendered content | Many AI crawlers do not run JavaScript, so text that appears only after scripts run is missing for them |
TTFB is not a Core Web Vital. web.dev calls it a supporting metric and gives 0.8 seconds or less as good. For crawler purposes it is the closest thing to a direct speed signal, because it is the part of the page load a crawler actually waits for.
Google's crawl capacity limit
Google is explicit about one effect of speed. Its crawl budget documentation says the crawl capacity limit goes up when response times stay stable or improve, and goes down when the site slows or responds with 5xx errors or 429 rate-limit responses. Google also says that if it can load and render pages faster, it may be able to read more content from the site.
For a small site this rarely matters, because Google can crawl it all anyway. For a large catalogue or a site that publishes often, a slow or unstable origin means fewer pages refreshed per day. Crawl budget for AI search goes into when this becomes a real constraint.
Speed work that actually helps crawlers
If your goal is AI search readiness rather than a Lighthouse score, spend the effort here first:
- Cache HTML at the edge. Most crawler requests are for pages that rarely change. A CDN cache hit returns in milliseconds and keeps load off your origin. Make sure crawlers are not bypassing the cache through query strings or cookies.
- Fix slow uncached pages. Find the templates with the slowest server response in your logs, usually search results, large category pages and pages that call several APIs. See slow TTFB and crawler timeouts.
- Stop errors under load. A burst of crawler requests that produces 502 or 503 responses is worse than a slow page. Scale the origin or rate-limit politely with 429 and
Retry-After. - Send the content in the HTML. Server rendering or static generation removes the need for a crawler to run JavaScript. This helps both speed and readability. Do AI crawlers run JavaScript? explains which ones do.
- Keep HTML lean. Inline images and large serialized data slow every fetch and can push content past Googlebot's limit, as covered in the Googlebot file size limit.
Improving LCP, INP and CLS is still worth doing for your visitors and for Google Search's page experience signals. It is simply a separate project from crawler access.
How to measure speed the way a crawler sees it
Time a plain request from outside your network:
curl -s -o /dev/null --compressed \
-A 'Mozilla/5.0 (compatible; speed-check)' \
-w 'status=%{http_code} ttfb=%{time_starttransfer}s total=%{time_total}s bytes=%{size_download}\n' \
https://www.example.com/category/shoesRun it several times, including against a URL with a cache-busting query string so you see the uncached origin. Then look at your server or CDN logs for real crawler requests and their response times. In nginx, add $request_time to the log format if it is not there:
log_format timed '$remote_addr [$time_local] "$request" $status '
'$body_bytes_sent "$http_user_agent" rt=$request_time';Filter for Googlebot, OAI-SearchBot, Claude-SearchBot and PerplexityBot and look at the slowest requests and any 5xx or 429 responses. For Core Web Vitals themselves, use field data: the Core Web Vitals report in Search Console or PageSpeed Insights, which shows Chrome user data when there is enough of it.
Frequently asked questions
Do Core Web Vitals affect whether ChatGPT or Perplexity cites my page?
There is no documented effect. OpenAI and Perplexity do not mention Core Web Vitals or page speed in their crawler documentation, and nobody outside those companies can measure a causal link.
Are Core Web Vitals required for Google AI Overviews?
No. Google says a page must be indexed and eligible to show with a snippet, and that there are no additional technical requirements. Core Web Vitals remain one of many signals in Google Search ranking.
Is TTFB a Core Web Vital?
No. web.dev describes Time to First Byte as a supporting metric and says good values are 0.8 seconds or less. It is the speed number most relevant to crawlers, because it is what they wait for.
Will a slow site stop AI crawlers from reading it?
Sometimes. A response that is slow enough to time out, or a server that returns 5xx errors under crawler load, gives the crawler nothing. The operators do not publish their timeouts, so aim for consistently fast responses rather than a specific number.
Check what a crawler receives
The free check fetches your page as an automated client and reports the status, the response and whether the main content is in the raw HTML. For a site with many templates, an audit samples them and points to the slow or failing ones.
Filed under Rendering and content. Topics: google, rendering, ai-crawlers, crawl-budget, ai-overviews