Testing landing pages: what to vary, what to leave alone.

Test one thing at a time, change the parts of the page that carry the decision, and leave the parts that build trust alone. Most landing-page tests fail because they change too much at once or too little that matters. You end up with a result you cannot read and a page that looks different but converts the same.

I have run these tests for law firms, dental practices, and trades. The pattern holds. The pages that improve are the ones where someone decided what question they were asking before they touched the page. This post covers what to vary, what to protect, and how to know when a test is worth running at all.

What should you vary first?

Vary the parts of the page that sit closest to the decision: the headline, the offer, and the call to action. These three carry most of the weight. A visitor decides whether to stay in the first few seconds, and the headline is what they read. They decide whether to act based on what you are asking them to do and what they get in return.

Start with the headline. It is the highest-leverage change on any page, and it is cheap to test. A San Diego personal-injury firm ran two versions of the same page. One headline said "Injured in an accident? We can help." The other said "No fee unless we win your case." The second version doubled the form fills. Same page. Same photos. One sentence changed.

Next, test the offer. The offer is what the visitor gets for taking action. "Book a consultation" and "Book a free 20-minute case review" are different offers, and they pull different numbers. Be specific about what happens next. Vague offers make people hesitate. Concrete ones lower the risk of clicking.

Then test the call to action itself — the button text and its placement. "Submit" is weak. "Get my free review" tells the visitor what they are getting. Try moving the button higher. Try adding a second button lower on the page for people who read to the end. These are small changes with measurable effects.

What should you leave alone?

Leave alone the parts of the page that build trust and the parts that already work. Your logo, your license numbers, your real client reviews, your years in practice — none of that should move because someone read a blog post about button colors. Trust signals earn their place slowly. Ripping them out to test a hunch usually costs you more than it gives back.

Also leave alone anything you have already tested and won. If a headline beat its rival by a clear margin last quarter, that is your control now. Do not re-litigate settled questions because you are bored. Testing is expensive in traffic and time. Spend it on open questions.

And resist the urge to test cosmetics before substance. Button color, font choice, and background shade rarely move revenue for a small business. They get tested constantly because they are easy to change and easy to measure. Easy is not the same as important. A downtown San Diego dental practice spent six weeks testing shades of blue on a booking button. The button color did nothing. The problem was the page loaded in four seconds on a phone, and half the visitors left before they saw any button at all. The real fix was speed, not color.

How do you run a test you can actually read?

Change one variable at a time and let the test run until you have enough visitors to trust the result. This is the rule most people break. They change the headline, the image, and the button on the same day, get a lift, and cannot tell which change caused it. Now they are guessing again next month. One variable per test. Every time.

Here is the sequence I use:

1. Write down the question. "Does a no-fee headline beat a we-can-help headline?" If you cannot state the question in one sentence, you are not ready to test.
2. Decide the metric before you start. Form fills. Booked calls. Not clicks. Not time on page. Pick the number closest to money.
3. Change one thing. Keep everything else identical.
4. Split the traffic evenly and wait. For a small business, this can take weeks. A page that gets 200 visitors a month cannot resolve a test in three days.
5. Read the result against your metric, not your gut. If the new version won on form fills, it won. If it won on clicks but lost on form fills, it lost.

Sample size is where small businesses struggle. You do not have enterprise traffic. A test that needs 5,000 visitors per version to reach confidence is not a test you can run — it is a test you will abandon halfway and misread. If your page gets low traffic, test bigger swings. A whole new headline and offer beats a subtle word change, because the effect is large enough to see with fewer visitors. This is one of the questions a good Fractional CMO helps you answer: what is worth testing given the traffic you actually have.

What breaks a landing-page test?

The most common thing that breaks a test is running it on a page that has a bigger problem underneath. You can test headlines all day on a page that loads slowly, hides the phone number, or targets the wrong search intent. None of the tests will help, because the page was never going to convert well no matter what the headline said.

Fix the foundation before you optimize the surface. A page that ranks for the wrong query brings the wrong visitors, and no amount of copy testing fixes a mismatch between what someone searched and what your page offers. If people search "emergency plumber near me" and land on a page about scheduled maintenance, the test is dead on arrival. That is a positioning problem, not a headline problem.

Another killer is testing during a traffic spike or a quiet season. If you run a test the week a competitor pauses their ads, your numbers are polluted. If you test during the two weeks around a holiday, they are polluted a different way. Pick a normal period. Run through it cleanly. Do not read a test that overlapped an event you cannot control.

The last one is impatience. Most tests get killed too early because the first two days look good and someone declares victory. Early results swing wildly. A test with 30 conversions per side can flip its winner in a week. Wait for the numbers to settle. The discipline to wait is worth more than the cleverness of the test.

How do you know a test was worth it?

A test was worth it when the result changed a decision and you can tie that decision to revenue. If you tested two headlines, the winner brought 20% more consultations, and those consultations turned into signed clients, the test paid for itself. If you tested two shades of blue and the winner brought a 2% lift in clicks that never turned into calls, you spent traffic on nothing.

The honest measure is downstream. Clicks are cheap and easy to move. What matters is whether the extra clicks became phone calls, booked appointments, and paying clients. A landing page that lifts form fills but fills them with unqualified leads is worse than the page it replaced. Track the whole path, not the top of it.

When we worked with McShanes Solicitors, the wins that mattered were the ones we could trace to signed matters — not the ones that only moved a vanity number on a dashboard. That is the standard. If we cannot tie a test result to revenue, we do not celebrate it.

Testing is part of a larger loop. It works best when it sits alongside clear positioning and a page that already loads fast and ranks for the right terms. If you are wondering whether you need outside help to run this loop, read when you actually need a Fractional CMO (and when you don't). And if you are weighing whether an agency or a fractional lead is the right fit for growth experiments, the difference that matters walks through it.

Where this breaks down

A/B testing does not help a business with too little traffic to reach a clear result. If your page gets a few dozen visitors a month, you will not resolve a test in any reasonable window — spend that effort on ranking and positioning instead, and test later when you have volume. Testing also will not fix a broken offer or a page pointed at the wrong audience. It sharpens what is already close to working. It does not rescue what was never going to work.

Start with the headline and the offer. Change one thing at a time. Protect your trust signals. Measure against revenue, not clicks. Foundations first, then the test.