How fast is fast enough for a website?
Short answer
Fast enough is Google’s ‘good’ line on three measures, for three visits in four on phones: the main content in 2.5 seconds, a reply to a tap within 200 milliseconds, and a page that barely jumps. Below that line, fix speed first. Above it, I would usually spend the money on the offer.
Contents 7 sections
Nobody can tell you what one second is worth on your site, and that includes me. What I can tell you is where the line is, what companies found when they measured crossing it, and what usually keeps a marketing site on the wrong side of it.
Where is the line?
Google publishes three measures, the Core Web Vitals, and gives each a “good” mark.11The thresholds and the 75th percentile rule are from web.dev, Web Vitals (updated 31 October 2024). The “poor” marks are from the pages for each measure: LCP, INP and CLS. All checked 11 November 2026.
| Measure, and what it times | Good |
|---|---|
| Largest Contentful Paint (LCP). How long until the biggest thing on screen has loaded. | 2.5 s or less |
| Interaction to Next Paint (INP). How long the page takes to answer a tap, click or key. | 200 ms or less |
| Cumulative Layout Shift (CLS). How much the page jumps while you use it. | 0.1 or less |
Over 4 seconds, 500 milliseconds or 0.25, Google calls it poor. The band in between “needs improvement”.
Google measures them at the 75th percentile of real visits, with phones and computers counted apart. In plain words: three visits in four have to be good. Not the average visit, and not your own laptop on the office wifi.
How many sites are over it?
Fewer than half, on phones.
- Fig.
- Source
- HTTP Archive Web Almanac
On phones, the first measure is the one sites fail most often: 62% pass LCP, against 77% for INP and 81% for CLS.22HTTP Archive, Web Almanac 2025, Performance chapter, with data from July 2025. INP is counted per website, LCP and CLS per page. almanac.httparchive.org, checked 11 November 2026. So a slow site is most often slow to show its main content, not slow to respond once it is there.
What did getting faster earn, where someone measured it?
These are published results with a method behind them.
More sales after a 31% better LCP on one landing page, in an A/B test on paid traffic split 50/50.
Source: web.dev, Vodafone, 2021
More retail conversions from a mobile site 0.1 s faster. 37 brands, 30 days.
Source: 55 and Deloitte for Google, 2019
Vodafone’s changes are worth reading, because they are ordinary ones. It moved a widget’s rendering from the browser to the server, sent the important HTML ready-made, and resized the hero image.33web.dev, Vodafone case study, 17 March 2021. Each version had about 34,000 visits a day. The faster version also raised lead-to-visit rate by 15%. web.dev, checked 11 November 2026. Nothing exotic, and both versions looked the same.
Rakuten 24, an online shop for everyday goods in Japan, ran the same kind of test on one landing page for about a month. The faster version had a 33.13% higher conversion rate. Its caption says version A loaded “within 1.6 seconds” and version B “within 2 seconds”.55web.dev, Rakuten 24 case study. The A/B test ran on one high-traffic landing page, with traffic split evenly. The page shows no date. web.dev, checked 11 November 2026. The page does not say which measure that is. If it is anything like LCP, both versions were already under the 2.5 second mark, and the faster one still won. Treat the line as a floor, not a ceiling.
Is there a formula from seconds to dollars?
No. Say it plainly, because someone will try to sell you one. Every study above measured someone else’s site, often one page, often one step. None of them says that ten tenths of a second are worth ten times one.
What you can do is size the question with your own numbers. Drag the page weight below and watch the download time against Google’s mark, then put in your own traffic to see what one published step would be worth if it held for you.
- Fig.
- Interactive
- Page weight, time and money
- Use it
- Drag the weight. Type your own numbers.
On this line every extra 100 KB adds 0.50 s. That is before the browser has parsed or painted anything.
What one tenth of a second is worth, at your numbers
Example numbers. Replace them with yours. The study measured one step of 0.1 s on retail and travel sites. It does not say ten steps are worth ten times as much, and your site was not in it. Read the result as the size of the prize, not a forecast.
Benchmarks used
- “Slow 4G” is 1.6 Mbps down with 150 ms of latency: roughly the bottom 25% of 4G connections. Lighthouse throttling guide, checked Nov 2026
- The median home page weighs 2.56 MB on mobile and 2.86 MB on desktop. HTTP Archive Web Almanac, Page Weight, July 2025 data
- A good Largest Contentful Paint is 2.5 seconds or less, at the 75th percentile of page loads. web.dev, Largest Contentful Paint, updated Sept 2025
- A 0.1 s faster mobile site raised conversion rates by 8.4% for retail sites and 10.1% for travel sites (37 brand sites, 30 days). 55 and Deloitte for Google, Milliseconds Make Millions, data from late 2019
What slows a marketing site down?
Four things, in the order I look at them.
- The hero. The big image or video at the top is usually what LCP times. On 76% of mobile pages, the largest element is an image.2 Size it for the screen, compress it, and load it first, never lazily.
- Fonts. 87% of mobile pages load at least one web font.2 Each file is something the text may wait for. Two families and a few weights are plenty.
- Third-party scripts. Tag managers, chat widgets, heatmaps, consent tools. The Almanac names JavaScript from “third-party widgets and analytics” among the causes of slow responses.2 Every script needs an owner and a reason to stay.
- Building the page in the browser. A page that assembles itself in JavaScript shows little until the script has run. HTML rendered on the server arrives ready to read. It was one of Vodafone’s changes.
How much speed should I pay for?
Get every key page over all three lines on phones, measured on real visits. Below the line, I fix speed before anything else, because the fixes are known and the tests above say they pay.
Over the line, the case gets thinner. Rakuten says faster can still win. But each further tenth of a second costs more to find, and Google is clear that speed is not the whole game in search. It uses Core Web Vitals in ranking, yet it “always seeks to show the most relevant content, even if the page experience is sub-par.”66Google Search Central, page experience documentation. developers.google.com, checked 11 November 2026. Past good, I would usually spend the next dollar on what the page says.
How do I check my own site?
Use field data, not a lab score. PageSpeed Insights shows the Chrome User Experience Report for your address: real visits over the previous 28 days, at the 75th percentile. Its lab test is a simulated load on one device and one connection.77Google, PageSpeed Insights documentation. developers.google.com, checked 11 November 2026. When the two disagree, the field data is the one your buyers had.
My site check reads a page’s HTML the way a crawler does, including how heavy it is. It does not time real visits, so use PageSpeed Insights for these three.
Speed is part of every website I build, checked against these three lines before launch.
Notes and sources
-
The thresholds and the 75th percentile rule are from web.dev, Web Vitals (updated 31 October 2024). The “poor” marks are from the pages for each measure: LCP, INP and CLS. All checked 11 November 2026. ↩︎
-
HTTP Archive, Web Almanac 2025, Performance chapter, with data from July 2025. INP is counted per website, LCP and CLS per page. almanac.httparchive.org, checked 11 November 2026. ↩︎ ↩︎ ↩︎ ↩︎
-
web.dev, Vodafone case study, 17 March 2021. Each version had about 34,000 visits a day. The faster version also raised lead-to-visit rate by 15%. web.dev, checked 11 November 2026. ↩︎
-
Michael Wiegand, Portent, 20 April 2022: 20 sites, a 30-day snapshot of page speed. portent.com, checked 11 November 2026. ↩︎
-
web.dev, Rakuten 24 case study. The A/B test ran on one high-traffic landing page, with traffic split evenly. The page shows no date. web.dev, checked 11 November 2026. ↩︎
-
Google Search Central, page experience documentation. developers.google.com, checked 11 November 2026. ↩︎
-
Google, PageSpeed Insights documentation. developers.google.com, checked 11 November 2026. ↩︎
Written by Dion Zeneli. AI helped with research and editing. Sources are linked where they are used. Read this note as markdown.