How to test changes on your links page without fooling yourself with the numbers

A/B tests sound like something reserved for big companies with data teams, but the idea is very simple: compare two versions of something to learn which works better. Anyone with a links page can do it, without special tools, as long as they respect a few rules. Without them, the risk is drawing false conclusions: believing a change worked when it was chance, or discarding a good idea because of a bad week.

What can be tested

Almost any element that influences visitor behavior:

• The text of a button ("Book" versus "Book my first session").
• The order of the links.
• The link that goes first.
• The header sentence.
• The profile photo.
• The theme or colors.

A good test changes only one of these and leaves everything else equal. If you change the button text, the order and the photo at once, you will not know which of the three produced the result.

The method, in five steps

1. Formulate a hypothesis. Not "let's see what happens", but a concrete sentence: "If the button says what happens on tap, it will get more clicks". A hypothesis forces you to know what you expect and why.

2. Choose one metric. Just one. Normally, clicks on the button being tested or the page's click rate. Choosing it beforehand prevents you from picking, afterwards, the one that looks best.

3. Fix the duration. Decide how long the test lasts before starting, for example two weeks per version. Cutting early because "it already looks clear" is one of the most common ways to fool yourself.

4. Compare equivalent periods. Version A and B should run under similar conditions. A normal week against a week with a viral post is not a fair comparison.

5. Decide with a rule. Before seeing the results, define what difference would make you switch. For example: "if the new version improves the click rate by at least a fifth, I adopt it".

How to do it without a testing tool

The simplest way is sequential: you run version A for a period, switch to B for an equal period and compare. It has a clear limitation: periods are never identical, and external factors (a key date, a post) can influence. To reduce that risk:

• Use periods of at least one full week, to cover every day.
• Note any event that may alter traffic.
• If possible, repeat the sequence (A, B, A) to confirm.

An alternative is to use different versions on different channels, for example one address for one social network and another for another, but that complicates reading because audiences differ.

How much traffic you need

With few visits, small differences mean nothing: two clicks more or less are noise. A practical rule: if the change is not clearly noticeable (more than a modest difference) and you have few visits, repeat the test for longer or try a more ambitious change. Small changes on low-traffic pages can rarely be detected.

When your traffic is small, focus tests on big changes (the main link, the header proposition) and not on details like an adjective.

Biases worth knowing

• Confirmation bias. We tend to see what we want to see. A decision rule written in advance helps.
• Novelty effect. A change may work at first just because it is new. Test long enough.
• Regression to the mean. An extraordinarily good or bad week is usually followed by a normal one.
• Looking at too many metrics. If you look at ten numbers, one will come out favorable by chance.

What to do with the result

If the new version won, adopt it and record the change. If it lost, go back to the previous one and note what you learned. If there is no clear difference, keep the simpler one and move on to test something else. A tie is not a failure: it tells you that element is not the factor that matters.

Keep a record, even a sheet with the date, the change, the hypothesis and the result. Over time, that testing notebook becomes the best knowledge about your audience.

What Zona's analytics offer

Zona's analytics show visits, clicks and click rate day by day, with 7 days of history on the free plan and up to 90 on Pro and Business, plus the star link. To see week-by-week comparison, the longer history helps. You can see it on the analytics page and in the guide to metrics that matter. To start, create your zone for free.

Common mistakes

• Changing several things at once.
• Cutting the test when the result "seems clear".
• Not defining in advance which metric counts.
• Comparing periods with very different events.
• Not recording what was learned.

Frequently asked questions

Do I need a special tool?

Not to begin with. A sequential comparison with a record and equivalent periods is a good starting point.

How long should a test last?

At least one full week per version, and more if you have little traffic.

What do I test first?

The first button and the header text: they are the elements that influence the most.

Can I test with few visits?

You can, but results will be less reliable. Focus tests on big changes and extend the duration.