What does time to first byte measure?
TTFB is a few waits added up, and only the last one is the server's. Redirects come first, if the link points to an old address. Then the DNS lookup finds the server, the browser connects (with a TLS handshake for https) and waits for the answer.
I timed those parts on the homepage of corewebvitals.io, my consultancy site, with 15 loads from Cloudflare's cache and 15 past it. The medians, October 29, 2026:
| DNS | Connect | TLS | Waiting for the server | TTFB | |
|---|---|---|---|---|---|
| From the cache | 1 ms | 10 ms | 16 ms | 26 ms | 52 ms |
| Past the cache | 1 ms | 9 ms | 13 ms | 61 ms | 85 ms |
DNS, connecting and TLS barely moved between the rows. Only the wait grew, from 26 to 61 ms, because past the cache the request goes on to the site's own server.
How fast is a good TTFB?
Google itself calls the 0.8 seconds a rough guide. PageSpeed Insights flags a server that takes over 600 ms under Document request latency. That's the server's part in one test; the 0.8 seconds is the whole TTFB of real visitors.
All five of my sites' homepages answer in 61 to 103 ms from the Netherlands, and not one test load took 200 ms. So aim well under 200 ms for a cached page. Several hundred milliseconds on a CMS page is worth a look, and over a second hurts everyone.
Does a slow server hurt your rankings?
Through Largest Contentful Paint, it can. TTFB itself isn't one of Google's Core Web Vitals, but LCP is, and every millisecond of TTFB is part of it.
On very large sites, Google lets Googlebot crawl more while response times, "including latency and Time-to-First Byte", stay stable.
How do I measure TTFB?
Start with real visitors. PageSpeed Insights has a TTFB row in its top section, "Discover what your real users are experiencing", if the page or the site gets enough Chrome traffic.
For one load, open DevTools in Chrome (right-click, Inspect), go to Network, reload, click the first request and open Timing. "Waiting for server response" is the server's part plus one round trip. The DNS, connection and SSL bars above it complete the TTFB.
Then test past the cache, because caching can hide a slow backend. Add a made-up query string (?test=123) or check the header: Cloudflare's cf-cache-status says HIT when the answer came from its cache. The homepage from the cache says little about a product page, a cart or a logged-in page.
The first step is always the cache. Then the cache hit ratio. You see it in CoreDash: some page groups or query strings have a much longer TTFB. After that we go page by page and attribution element by element to find what's slow and why, and fix it: first at site level, then template, then page, so the first rounds already catch most of it.
MarketingTracer's uptime monitor checks only the homepage, every 30 minutes from our servers, and shows its median response time (TTFB) with a chart of every check. It sends no alerts yet. For the client report, take PageSpeed Insights' real-user TTFB, and the monitor's chart for the trend.
How do I lower a slow TTFB?
Cache first, then redirects, then distance, and the server last.
A full-page cache (a CMS plugin or the server) keeps a finished copy of the page, so the server doesn't rebuild it for every visit. Where a page is the same for everyone, let the CDN keep the HTML too: on my homepage, Cloudflare's copy cut the wait by more than half.
Every redirect adds a full round trip before the first byte. Link straight to the final address; a site crawl finds the internal links that still redirect.
Distance counts too: every round trip takes longer for a visitor far from the server, and a CDN brings the connection closer. My TTFB page on corewebvitals.io shows real-user TTFB per country and goes deeper.
When the wait past the cache stays long, the server itself is slow: heavy database queries, too little memory, a crowded shared plan. That's for the developer or the host, and a new host rarely fixes what a cache would have fixed.