Time to First Byte (TTFB): how to measure a slow server response and what can reduce it
Understand what TTFB measures, how field and lab results differ and which network, hosting or application work may reduce delays.
TTFB is the time from starting to load a page until the first byte of the response arrives. Everything the visitor sees comes after it. This article summarises two web.dev pages, as read on 15 November 2026 (both pages carry late-2025 update dates). It is for people who commission or review a site: what to ask for, and how to tell a real problem from a measurement artefact. Our LCP and CLS workflow covers where TTFB sits among the other loading steps.
What TTFB includes
web.dev says TTFB is the sum of redirect time, service worker startup (if there is one), DNS lookup, connection and TLS negotiation, and the request up to the arrival of the first byte. So a slow server is one cause among several. Slow redirects or a distant server can look the same from the outside.
What counts as good, and why not to chase it blindly
- web.dev's rough guide: most sites should aim for a TTFB of 0.8 seconds or less; above 1.8 seconds is poor.
- It says the aim is that the 75th percentile of users gets a good First Contentful Paint, and that TTFB is not a Core Web Vitals metric, so it is not strictly necessary to hit the threshold if it does not hold back the metrics that matter.
- It explains why the numbers are a rough guide: a server-rendered site can have a higher TTFB but better FCP and LCP than a fully client-rendered one, while for a single-page app that fills the page with JavaScript a low TTFB is especially important.
- A caveat on comparison: 103 Early Hints, or flushing a response early, count as the first byte, so TTFB is hard to compare across platforms. Chrome changed what TTFB measures in version 115 and reverted the change in version 133, per the page. Know what your tool measures.
How to measure it
- Use real-user (field) data first. The optimisation page says field data is the primary source because it includes redirects, while lab tools often test the final URL and miss that delay. PageSpeed Insights shows both field and lab data for public sites in the Chrome User Experience Report.
- The Lighthouse "server response time" audit excludes DNS and redirect time, so it covers only part of TTFB, per the page.
- If lab TTFB is much larger than field TTFB, the lab setup is more constrained than typical users. If field is much larger than lab, the page lists server-side caching, redirects and network differences as likely causes, and suggests testing less common pages or URL parameters to bypass caches.
- The Server-Timing response header lets the backend report named timings, such as database queries, server-side rendering time, disk seeks and CDN cache hits or misses. They show in the browser's Navigation Timing data and in Chrome DevTools. If that is not practical, the page says an application performance monitoring tool is an alternative.
What the page lists as ways to reduce it
- Check the platform. Its examples say WordPress speed depends on plugins and themes, and that customised platforms are similarly affected. Read the platform's own guidance.
- Hosting. The page says to consider hosting first: enough memory, a provider that keeps the backend stack current, and, for larger personalised applications with database work, a choice that matters more. Shared hosting is generally slower, which it says is probably fine for a small, mostly static site.
- A CDN. Even with a fast backend, distant users can see high TTFB. A CDN caches on edge servers closer to users, and the page lists extras such as fast DNS, HTTP/2 or HTTP/3, modern TLS, edge workers and well-handled compression. It calls adopting a CDN a high priority if you do not have one.
- Caching. Use Cache-Control so edge servers can reuse responses. Even short cache times help busy sites, the page says. Unique query-string parameters for analytics can defeat the cache. It warns that caching can hide a slow backend from your own tests, so keep a way to bypass it, and that older or rarely visited pages may not be cached.
- Fewer redirects. The page names ads and newsletters that redirect through analytics services as a common source, and says to fix same-origin redirects you control: missing https in links, and inconsistent trailing slashes. The HTTP to HTTPS redirect can be avoided on repeat visits with the Strict-Transport-Security header, and for the first visit by joining the HSTS preload list.
- Stream the markup. Browsers handle HTML in chunks as it arrives, so a backend should send the start of the page early. The page notes newer React versions have streaming server rendering, that Node.js and Deno support this, and that other runtimes may not. Static rendering at build time suits pages that need no personalisation.
- A service worker. A stale-while-revalidate strategy can make TTFB nearly instant for rarely changing documents, but not for markup that changes per user. An app shell model suits single-page apps.
- 103 Early Hints. Lets a server tell the browser to start fetching render-critical files while the backend is still preparing the page. The page says it helps most when the backend does a lot of work, and little for static sites. It also warns Early Hints can make TTFB look fast while the server is still slow, so measure real server time with Server-Timing or finalResponseHeadersStart.
Omnistu advice for a site project (not from the source)
- Ask for field TTFB for your important pages, by device type, before agreeing any work. If there is no field data, ask the developer to add real-user measurement.
- Ask for a Server-Timing breakdown on a slow page: how much is the database, how much is rendering, how much is the network. It turns "the site is slow" into a specific cost.
- Check your own links, ad URLs and newsletter links for extra redirects before touching the server.
- When someone proposes a CDN, caching or Early Hints, ask how they will check that the improvement is real and not a cached or early-flushed measurement.
- Treat the 0.8 second figure as a rough guide, as the page does.
Limits
Two web.dev pages, read in full on 15 November 2026. We did not read the pages they link to (the HSTS, Early Hints, Cache-Control, service worker and Server-Timing references), and we tested nothing. We make no claim about how much any change reduces TTFB for a particular site; the pages do not either. Browser and tool behaviour, such as what TTFB reports, has changed before and may again.
By Omnistu.