The uncomfortable part of this question is that the people who leave never explain themselves. You get a number that went down and no account of what happened, so the temptation is to guess, change six things at once, and watch the number for a week.
It is more tractable than that. Leaving is not one event, it is several different events that happen at different moments, and each has a different cause and a different fix. Working out which moment you are losing people at narrows the problem enormously.
1. They left before the page appeared
This is first because it is the most commonly missed. If the page takes several seconds to show anything readable, a meaningful share of people are gone before they have seen a single word you wrote. No amount of copywriting reaches them.
What matters is not page weight or a score out of a hundred. It is the gap between the tap and the first thing a person can actually read. On a weak mobile connection that gap is often several times longer than it is on your laptop, and your laptop is where you tested it.
2. They could not tell what the page was for
Five seconds is roughly what a stranger gives a page before deciding whether to invest more. In that time they are answering one question: is this the thing I was looking for? If the page cannot answer it, they go back.
The reason this is hard to see yourself is that you cannot un-know what your page is about. You look at it and the meaning is already there. A stranger looks at it and sees a headline, an image, and a button, and has to reconstruct the purpose from those three things.
The test is free and takes a minute: show the page to somebody who has never seen it, for five seconds, then hide it and ask what the page was for and what they would do next. If they cannot answer, nothing further down the page matters.
3. The page is doing more than one job
Pages accumulate. Somebody asked for the newsletter box, somebody else wanted the case studies above the fold, and the founder wanted the mission statement. Each addition was individually reasonable and the result serves nobody.
A page with one job can be judged: does the visitor do the thing? A page with four jobs cannot be judged at all, and it usually turns out that the most visually prominent element is not the one the page exists for.
4. There were too many decisions
Every choice you present is a small cost paid by somebody who has not yet decided to trust you. Below a certain number this is fine. Above it, people stop choosing rather than choosing badly, and stopping looks exactly like leaving.
This is where the popular version of the psychology overstates things. The famous jam study is much less settled than its fame suggests, and later work found the average effect close to zero. What does survive is narrower and more useful: choice hurts when the options are hard to tell apart, and helps when they are clearly different. Six clearly distinct plans is fine. Three plans whose differences require a table to understand is not.
5. They did not believe you were real
A stranger arriving on an unfamiliar site is asking a question that has nothing to do with your product: is there a real business behind this, and what happens if something goes wrong?
The signals that answer it are unglamorous. A physical address. A named person with a real photograph, which outperforms a logo. A way to reply that is not only an email address to copy out. Specific social proof about people like the reader, rather than a vague number of anonymous customers.
6. The form asked for too much
Every field has a price and most forms are collecting things nobody has a use for. The phone number field is frequently the most expensive box on a site: it is optional to the business and it reads, to the visitor, as a commitment to being telephoned.
The test for each field is simple and almost nobody applies it: if this were blank, would anything actually stop working? If not, it is costing you completed forms in exchange for data nobody reads.
7. The offer is wrong, and the page is fine
This one is last and it is the one nobody wants. Sometimes every mechanical thing on the page is correct and people still leave, because the price is wrong for what is being offered, or the thing itself is not wanted, or the visitors arriving were never going to buy anything.
No amount of page work fixes that, and a consultant who takes the fee without saying so is being paid to avoid an inconvenient conversation. If your traffic is badly targeted, the audit that matters is of your traffic sources, not of your page.
How to work out which one is yours
| What you observe | Most likely cause |
|---|---|
| Lots of visits under 3 seconds | They never saw the page — speed |
| People arrive and leave from the same page | Cannot tell what it is for, or wrong traffic |
| They browse several pages then go | Doubt, or the offer |
| They start the form and abandon it | The form is asking too much |
| They reach checkout and stop | Trust, unexpected costs, or the phone checkout |
| Everything looks fine and nobody buys | The offer, the price, or the audience |
That table is not a diagnosis, but it narrows a vague problem to one or two candidates, and that is most of the work. Everything after it is measurement.
Questions people also ask
What is a normal bounce rate?
There is no useful universal number, and the ones published online average across businesses that have nothing to do with each other. A blog post with a high bounce rate may have served its purpose perfectly. Compare your page to itself last month, not to somebody else's benchmark.
Should I add a chat widget?
Sometimes. It genuinely helps when doubt is the problem and somebody is actually there to answer. An unattended widget that promises a reply and does not deliver one is worse than no widget, because it converts a small doubt into a broken promise.
Will making the page shorter help?
Only if length was the problem. Long pages work well when the reader is committed and the content earns each section. Cutting a long page that was working is a common and expensive way to make things worse.
How much traffic do I need before this is worth studying?
For testing, thousands of visits a month. For measuring mechanics, none — the fourteen checks work on a page with no visitors at all, which is also the best time to run them.