First, check the definition and split by source and device
Bounce rate is not one metric across tools. In Google Analytics today, as the reports and dimensions guide explains, Bounce RateShare of sessions that were not engaged (GA4); historically single-page sessions. is the share of sessions that were not engaged, and a session counts as engaged if it lasted beyond a documented threshold, included a key event or viewed more than one page. Older analytics setups, page builders and ad platforms count a bounce differently, often as any single-page visit. A rise that coincides with a tool change, a new consent setup or a page that now loads as one long screen may be a change of definition rather than a change in behaviour. Write down which definition the report uses and compare only figures measured the same way.
Then break the page's bounce down by source, campaign and device for the same period. A page that bounces heavily on paid social phones and lightly on branded search desktops has a traffic or mobile problem, not a page-wide one. Read bounce beside the page's conversion rate (CVRShare of visits or clicks that convert.): a page whose bounce rose while conversions held may simply have lost visitors who were never going to act. If the visitors do stay but the rate fell after a launch, the conversion rate troubleshooting hub routes to the redesign and sudden-drop pages.
Diagnosis table
Diagnosis
| Symptom | Check | Likely cause | Fix |
|---|---|---|---|
| Ads or emails earn clicks, and those visitors leave fastest | The ad's or link's promise against the page's headline and first screen | Page breaks the click's promise | Restate the promise and offer in the first screen |
| Bounce is far higher on phones and slow connections | Field data for the largest element's load time by device | Slow first load | Shrink and prioritise hero media; cut blocking scripts |
| Visitors from every source leave quickly; recordings show little scrolling | A stranger's reading of the first screen: what, for whom, what next | First screen unclear | Lead with the offer, the audience and one action |
| One query, placement or audience bounces far above the rest | Search terms and placements behind the bouncing sessions | Traffic with no intent to act | Exclude or re-route that traffic; match intent |
| Bounce rose after a pop-up, banner or consent notice was added | The page on a phone in a fresh session: what covers what, and when | Overlays covering the content | Delay, shrink or remove overlays; reserve space |
| The figure jumped on a date with no change to the page or traffic | Which tool and definition the report uses, before and after | A different bounce definition | Pick one definition; compare like with like |
Fix, verify and prevent each cause
A page that does not continue the click
People click because an ad, email or post promised something, and on arrival they scan for that promise. If the headline talks about something else, many leave without reading further.
Fix: read each traffic source's ad or link beside the page it lands on, and restate its promise, offer and any price in the headline and subhead, using the audit method in the message match guide; where several ads with different promises share one page, give each its own variant.
Verify: bounce from the corrected sources falls and their conversion rate rises, while untouched sources stay where they were.
Prevent: add the destination page to every ad and email review, and re-audit pairs whenever either side is rewritten.
A first load that keeps visitors waiting
A visitor on a phone over a mobile network who sees a blank or half-built screen often leaves before the headline appears. web.dev documents Core Web Vitals, including Largest Contentful Paint for loading, as measures taken from real visitors, which is the right lens for a page bought with paid clicks.
Fix: read field data for the page by device, find what delays the largest element, typically an oversized hero image, late fonts or render-blocking tags, and fix it in the template as the speed and mobile guide explains.
Verify: the loading measure moves into a better band in field data, and bounce on phones falls over the following weeks.
Prevent: test every new tag and hero asset on a throttled phone before publishing it.
A first screen that does not say what is on offer
Some pages load quickly and match the ad, yet the first screen is a mood image and a slogan, and the visitor cannot tell what is offered, for whom or what to do.
Fix: rebuild the first screen around a headline that names the outcome, a subhead that says for whom and how, the offer stated plainly and one visible action, as the anatomy of a landing page sets out; on phones, keep all of it within the first screen.
Verify: a five-second test with people unfamiliar with the business shows they can say what is offered, and scroll depth and clicks on the main action rise.
Prevent: run the same five-second test on every new Landing PageThe page a visitor arrives on from a campaign, built to turn that visit into one action. before it receives paid traffic.
Visitors who never meant to act
Broad keywords, loose audience settings and placements in games or content apps send people who were curious, early in their thinking or there by accident. They leave because the page asks for a decision they are not ready to make.
Fix: find the queries, placements and audiences behind the bouncing sessions and judge their intent with the intent signals guide; exclude what has none, and send early-stage traffic to a page that offers a lighter next step instead of the main form.
Verify: bounce for the remaining traffic falls and cost per conversion improves, even if visits drop.
Prevent: review search terms and placements on a fixed rhythm and record the intent level each campaign is meant to serve.
Overlays that hide the page
A newsletter pop-up on arrival, a chat window, an announcement bar and a consent notice can together cover most of a phone screen before the visitor has read a word; some push the page down as they appear and cause mis-taps.
Fix: open the page on a phone in a fresh session and record what covers what; delay or remove promotional pop-ups on paid landing pages, keep the consent notice compact, and reserve space for any banner so nothing shifts, following the layout advice in speed and mobile. Keep the reassurance the trust and proof guide places in the first screen visible, not buried under a notice.
Verify: bounce on phones falls after the change and the first screen shows the headline and offer without dismissing anything.
Prevent: require sign-off from the page owner before any overlay or tag that displays content goes live on a landing page.
A bounce figure measured differently
If the number changed because the tool or definition changed, no page fix will move it back.
Fix: confirm which definition each report uses, as the reports and dimensions guide describes for Google Analytics, and rebuild the comparison on one definition; where single-page visits are valuable, such as a call or a direction request, record those actions as events so the page is credited.
Verify: the trend read on a single definition shows whether there was ever a behaviour change.
Prevent: note the definition beside every bounce figure in reports and annotate tool or consent changes on the date they happen.
A prevention routine for landing page exits
Most early exits are caused before a page goes live, so the routine sits at launch and at every change to the ads or the page. Before a page receives paid traffic, read each ad or link beside it, run a five-second test on the first screen with people outside the team, load it on a mid-range phone over a throttled connection, and check what covers the screen in a fresh session. Agree one bounce definition and one report for each page, and note it beside the figure. Weekly, read bounce and conversion by source and device rather than as a page average, so a single bad source or a slow template shows up on its own. Monthly, review search terms, placements and audiences for traffic with no intent, and re-audit the message match for any campaign whose ads changed. Treat every new tag, pop-up and banner as a change to the page that needs the same checks as a redesign.
A boiler installer's quote page
A boiler installer's quote page showed a bounce rate of 78% against 52% the previous year, and the team blamed the new design. The report had moved to a tool that counted any single-page visit as a bounce; on the old engaged-session definition, bounce was 61%. Split by source, display placements in mobile games bounced at 93% and made up a third of visits; search ads promising a fixed-price boiler landed on a headline about "home comfort" with no price. Field data showed the hero video delaying the largest element on phones. The installer excluded the game placements, put "Fixed-price boiler replacement, fitted in two days" and the price in the headline, and replaced the video with a compressed image. Bounce on the engaged definition fell to 44% and quote requests rose from 96 to 141 a month.
Common mistakes
Comparing bounce figures from different tools. Redesigning a page whose problem is one traffic source. Leading the first screen with a slogan instead of the offer. Stacking a pop-up, a chat window and a consent notice over the headline. Judging bounce without conversion beside it. Treating a call from the page as a failed visit.
