TL;DR
- The thresholds that matter for how a wait feels have been known since the late 1960s and have not moved.
- Below about a tenth of a second a response feels instant. Around a second, thought carries on but the delay is noticed. Past ten, attention goes.
- Google publishes the thresholds it grades sites against, and they are public. You do not need to guess.
- A word of warning: much of what is written about page speed and the brain is invented. This article says what we cut and why.
Waiting for a web page is one of the few genuinely universal irritations. It is also one of the better-studied ones, because the question of how long a person will tolerate a machine taking to answer them predates the web by decades.
What follows is the part that is actually supported. It is less dramatic than the version you may have read, and more useful.
The three limits, and where they came from
Jakob Nielsen of the Nielsen Norman Group, writing in 1993 and drawing on Miller’s 1968 work on response time in man-computer conversational transactions and Card and colleagues in 1991, set out three limits that have not needed revising since. About 0.1 second is the limit for the user to feel the system is reacting instantaneously. About 1.0 second is the limit for the user’s flow of thought to stay uninterrupted, even though the delay is noticed. About 10 seconds is the limit for keeping the user’s attention focused on the dialogue.
Those three numbers explain nearly everything about how a slow site feels. Under the first, an action feels like a direct consequence of your click — you are operating the thing, not asking it for something. Between the first and the second, you notice the machine but you keep your train of thought. Past the second limit you have started waiting, which means you have started thinking about something else. And once you are past the third, whatever you came to do has been displaced by the wait itself. Nobody abandons a slow page because they are impatient. They abandon it because by then they are thinking about something else.
This is why the fixes people reach for are often the wrong size. Shaving two seconds off a page that took twelve is a large technical achievement and changes very little, because you are still on the wrong side of every limit. Getting a menu to respond in a tenth of a second rather than half a second is a small technical change that alters how the whole site feels.
A word about the neuroscience
You will find a great many articles claiming that slow websites spike cortisol, disrupt alpha and beta neural oscillations, slow axonal conduction or put the amygdala on alert. We went looking for the primary research behind those claims and could not find it. They appear to have been invented and then copied, and this article originally repeated several of them.
We have taken them out, and we would rather say so than quietly delete them. The honest position is duller and stronger: waiting is annoying, annoyance uses attention, and attention is what you needed the visitor to spend on your offer. You do not need a brain scan to justify making a page faster, and any article that offers you one without a citation is guessing.
The numbers Google actually publishes
If you want thresholds to work to, use the ones your site is genuinely being graded against rather than invented benchmarks. Google’s Core Web Vitals documentation sets three, each assessed at the 75th percentile of page loads and reported separately for mobile and desktop.
| Metric | What it measures | Google’s “good” threshold |
|---|---|---|
| Largest Contentful Paint (LCP) | How long until the main content of the page has rendered | Within 2.5 seconds |
| Interaction to Next Paint (INP) | How quickly the page responds when someone interacts with it | 200 milliseconds or less |
| Cumulative Layout Shift (CLS) | How much the page jumps about as it loads | 0.1 or less |
The 75th percentile matters more than most people notice. It means the standard is not what your site does on a good connection on your machine — it is what it does for the slower quarter of real visits. Testing on office broadband on a desktop tells you almost nothing about whether you pass.
What the delay is worth in money
The commercial side has been measured too. In Milliseconds Make Millions, research commissioned by Google and carried out by 55 and Deloitte across 37 brands, a 0.1 second improvement in mobile site speed moved 3.2% more people from the product listing page to the product detail page and 9.1% more from the product detail page to add-to-basket. Retail consumers spent 9.2% more, and travel booking rates improved by 10%.
Note the size of the intervention. A tenth of a second — the same threshold Miller identified as the point where a response stops feeling instant — produced movement at every stage of the funnel. That data comes from retail and travel brands on mobile, so treat it as an indication of scale rather than a promise for your sector. But it does mean the payback on speed work is real and measurable, which is more than can be said for most website improvements.
What to fix, in the order that usually pays
Speed work has a bad reputation for being expensive because people start with the hardest parts. In practice the first three items below account for most of the problem on most small business sites, and none of them requires a rebuild.
- ImagesCorrectly sized, compressed and in a modern format. This is almost always the largest single win, and it is the one you can do this afternoon.
- Scripts you do not needOld tracking tags, chat widgets nobody answers, fonts loaded in six weights. Removing things is cheaper than optimising them.
- CachingSo that returning visitors and repeat pages are not rebuilt from scratch every time.
- HostingCheap shared hosting shows up in server response time, and no amount of front-end work fixes it.
- The theme itselfSliders, page builders and icon libraries load whether or not you use them. Last, because it is the only item on the list that can turn into a rebuild.
There is a second lever that costs nothing and is routinely ignored: perceived speed. A page that shows its heading and layout immediately while an image finishes loading feels faster than one that shows nothing until everything is ready, even when the total time is identical. Reserving space for images so the page does not jump about is the same idea — the layout shift is not slow, it just feels careless.
How to check yours properly
- Test on a real phone
Use a mid-range handset on mobile data, not your laptop on office wifi. The gap between the two is usually where the problem is hiding.
- Test the pages people land on
Most speed testing is done on the homepage. Most visitors arrive somewhere else, often the slowest, heaviest page on the site.
- Use field data, not just lab tests
A one-off test tells you what happened once. The report of what real visitors experienced tells you whether you pass.
- Fix the largest thing first
Whatever the report names as the biggest single contributor, do that one, then measure again. Batch fixes leave you unable to tell what worked.
Set a target that means something. “Faster” is not a target. “Main content on screen within two and a half seconds for three-quarters of mobile visits” is one, and you can tell when you have hit it.
Good to know
FAQs
Fast enough to stay inside the thresholds that govern how a wait feels: roughly a tenth of a second to feel instant, about a second before the delay interrupts someone’s train of thought, and ten seconds before their attention has gone entirely. Google’s own “good” grade asks for the main content on screen within two and a half seconds.
We could not find primary research supporting that, despite it being widely repeated, so we do not use the claim. The straightforward explanation is enough: waiting is irritating, irritation takes attention, and attention is what you wanted the visitor to spend on your offer.
Images, almost always — uploaded at full resolution, uncompressed and scaled down in the browser. After that, scripts and plugins nobody can justify, then cheap hosting showing up in server response time.
The Google, 55 and Deloitte study measured movement at every stage of the funnel from a tenth of a second’s improvement on mobile, across retail and travel brands. It is sector-specific data, but the effect is real and measured rather than asserted.
Usually because the test ran on a fast connection on a desktop. Google grades at the 75th percentile of real visits, split by device, so the standard is what the slower quarter of your visitors get — not what you get in the office.
Next step
Site slower than it should be?
We help businesses across Brighton and London make their websites faster, starting with whatever is actually causing the delay.