What is a good bounce rate, really?

Ask what a good bounce rate is and you will usually be told "around 40%". It is a tidy answer to a question that changed in 2023, and knowing what changed makes the number genuinely useful again.

The number means two different things depending on which tool you ask, and it deserves your attention far less often than people think.

One word, two measurements

Bounce rate used to mean one simple thing: someone arrived, looked at one page, and left without going anywhere else. One page, then gone.

Then GA4 arrived and changed it. In GA4, bounce rate is the percentage of visits that weren't engaged — and "engaged" has three ways to happen. A visit counts as engaged if it lasted ten seconds or more, or it included a conversion, or it included a second pageview. Fail all three and you've bounced.

So the ten-second rule does most of the work. A visitor who reads your page for twelve seconds and leaves is engaged in GA4 and a bounce nearly everywhere else. Same person, same visit, two different answers.

Nearly every "what's a good bounce rate" article was written before 2023, which means it's describing the old number. If you read that 40% is healthy and then compare it to what GA4 shows you, you're comparing two different measurements and drawing conclusions from the gap.

A bounce is sometimes the best outcome

That was true of the old definition too, and it's the part benchmarks can't fix.

Picture a plumber's website. Someone searches for an emergency plumber, lands on the page, sees the phone number, calls it, and closes the tab. One page. Gone in forty seconds. By the classic definition that's a bounce, and it's also the single best outcome that website can produce.

Now run it the other way. Someone lands on your pricing page, can't find what a plan costs, clicks three times looking for it, gives up and leaves. Four pageviews. Not a bounce. A complete failure.

Bounce rate counts movement. It doesn't know anything about satisfaction, and movement and satisfaction aren't the same thing — they're sometimes opposites. A page that answers the question immediately produces a bounce. A page that hides the answer produces a healthy-looking click path.

Skip the benchmarks

People want a number to compare themselves to. Here's the rough shape of what gets published, and then why you shouldn't lean on it.

Three reasons to look past it.

The bands overlap almost completely. Nearly any number between 30% and 70% is "normal" for something. If your answer is inside that range, you've learned nothing.

Nobody is measuring the same thing. Those figures come from different tools, different years, and now two different definitions. Averaging them produces a number with no meaning.

Your traffic mix moves it more than your website does. Paid social bounces hard — people are interrupted, not searching. Branded search bounces softly — they came looking for you. Shift your spend between the two and your bounce rate changes without a single page being edited.

Three sharper numbers

Bounce rate is a fine smoke alarm and a terrible dashboard. Use it to notice that something changed, then look at something sharper to find out what.

• Time on page, by page. Not the site average — that's dominated by whichever page gets the most traffic. A page where the time drops by half after a redesign is telling you something specific.
• Scroll depth. If most people stop at 30% of a long page, the thing that lost them is at 30%. This is the single most actionable number on this list and almost nobody looks at it.
• Where the dead clicks are. People clicking something that isn't clickable — a heading that looks like a link, an image they expect to zoom. Every one is a small act of confusion you can fix. Click maps surface these as their own report.
• Bounce rate split by source. One number for the whole site hides everything. Split it and you'll usually find one channel dragging the average around.

What we report

We use the classic definition: a visit that saw one page and went no further. No ten-second rule, no engagement test.

GA4's version captures something the old one misses, so this isn't a claim that ours is better. We use it because it's the definition nearly every tool outside Google uses, and because a number you can explain in one sentence beats one that needs a paragraph and three conditions. If we imported your GA4 history, those bounce rates stay labelled as GA4's rather than being quietly mixed in with ours. They aren't the same measurement and we won't pretend they are.

One more thing if you're comparing tools. A consent banner sits between every visitor and your content, and some people leave rather than deal with it. Those departures land in your bounce rate — and you can't size the effect from your own data, because the people who refused are the ones you lost permission to count. Removing the banner changes that number — which is a reason to be sceptical of any before-and-after comparison that also changed the consent setup. We write about when a banner is actually required separately, and how counting works without one too.

The short version

1. Check which definition your tool uses before you compare yourself to anything.
2. Ignore benchmarks. Compare the page to itself, last month.
3. A bounce isn't a failure. Sometimes it's the fastest possible success.
4. When the number moves, go and look at scroll depth and dead clicks. That's where the actual answer is.
5. If it hits exactly 0% or exactly 100%, you have a tracking bug, not a marketing problem.