Call to action that converts: how to write it and prove it with A/B testing

Updated on October 28, 2026

A red "Get Started" button garnered 9.8% of clicks, while a "Contact Us" button only managed 2.7%. On paper, the former wins hands down. Except that the latter generated three times as many submitted forms. This is what the agency Aspiration Marketing discovered when testing four versions of the same call to action, and it's the trap most teams fall into: writing a CTA for clicks, when the company is paid per conversion.

This guide gives you a method for writing a call to action that really converts, then a protocol to prove it through A/B testing, with supporting figures, without telling you stories.

What a call to action that converts is (and what it isn't)

A call to action that converts is a button or link whose text announces a specific benefit, corresponds exactly to what the visitor will find after clicking, and whose effect on the business objective has been measured by an A/B test. All three conditions matter. Without the third, you have a CTA that "appears to convert".

Traditional guides only address best writing practices. These remain unverifiable until tested against your visitors. Ronny Kohavi, who led the experiment at Microsoft, reports that of the ideas tested online, about a third improve the targeted metric, a third make no difference, and a third worsen it (presentation “Online Controlled Experiments”, StanfordDepending on the field, 60 to 90% of ideas fail to produce the desired effect. Your button rewrite, however brilliant, follows the same principle.

Therefore, remember this distinction: writing produces a hypothesis, the test produces proof. If the vocabulary (variant, control, significance, MDE) is less familiar to you, the A/B testing glossary covers every term used in this article.

Click or conversion: the trap that reverses the verdict of a CTA test

Measuring a CTA by click-through rate often amounts to rewarding the vaguest promise, not the most profitable one. The correct indicator is the final conversion rate relative to the number of visitors exposed, not the button click.

Let's return to the test published by Marketing Aspiration, which tested four buttons leading to its contact page:

We recalculated these figures using a z-test for comparing proportions. Three key findings that the original article did not highlight:

• On click, A overwrites D (z-score of approximately 18, the difference is not due to chance). A team that drives by click would have deployed "Get Started".
• On the forms per view, D beats A (z-score of approximately 4.5, well above the threshold of 1.96 which corresponds to 95% confidence). The verdict is completely reversed depending on the metric chosen.
• A versus B, same text, different color 0.24% versus 0.23% of forms per view, z-score of approximately 0.2. No detectable difference. The color increased clicks, not business.

The explanation is simple: "Get Started" promised something vague; visitors clicked out of curiosity and left upon discovering a contact form. "Contact Us" honestly announced what was next: fewer clicks, but qualified ones. The authors acknowledge that the variations weren't run simultaneously, which weakens the comparison. How do you declare the right metric from the start? That's the purpose of this page. objectives and tracking in BlastOptimize : the main objective (form sent, purchase, registration) is defined here before launching any variation.

The 4-block formula for writing the button text

An effective CTA text combines four blocks: a verb that describes what the visitor gets, the concrete object of this gain, a precise element, and if necessary a risk reducer placed right next to the button. You won't always use all four, but you'll know which one is missing.

Block 1 — the verb of obtaining

“Send”, “Submit”, “Confirm” describe what you expect from the visitor. “Receive”, “Get”, “Discover”, “Compare” describe what they gain. Michael Aagaard, a Danish CRO consultant, achieved +14.79% conversions by replacing “Order Information and Prices” with “Get Information and Prices” — a single word changed, the one that shifted the effort to the benefit (Report published by Unbounce).

Block 2 — the object of the gain

"Start" doesn't say what begins. "Start my audit" does. The subject line should name what the visitor will have in their hands: a quote, access, a comparison, a 12-page PDF.

Block 3 — precision

A specific number or timeframe makes the promise credible: "Receive my quote in 24 hours," "Try it for 14 days." Precision reduces uncertainty, and uncertainty is the primary cause of hesitation when faced with a button.

Block 4 — the risk reducer

It's often a tiny line of text below the button rather than inside it: "No credit card required," "No commitment," "Response from a human, not a robot." It addresses the objection that arises at the last moment.

Test the formula using the "I want..." principle: read your button preceded by "I want." "I want to receive my quote in 24 hours" sounds right. "I want to submit" is meaningless. If the sentence doesn't work when spoken, the button won't work either.

Examples of rewritten CTAs: before / after

The most cost-effective rewrite almost always involves replacing a generic verb with the concrete benefit on the following page. Here are typical cases of B2B and e-commerce sites, with the assumption that each rewrite would allow testing.

For an e-commerce site, product page and shopping cart CTAs deserve specific treatment (shipping costs, delivery times, stock): the use cases described on the page A/B testing for e-commerce give an overview. Be careful, though: these rewrites are hypotheses, not recipes. “See the 20-min demo” can scare off an audience that expected a 2-minute video. Only testing settles it.

What surrounds the button matters as much as the button itself.

A CTA converts when its promise is consistent with the title that precedes it and the page that follows; excellent text placed on an inconsistent page remains a bad CTA. The Aspiration test shows that it was the consistency of "Contact Us" → contact page that made the difference, not the wording itself.

Three contextual elements to check before modifying the text:

• The title above the button. If the title mentions "reduce your costs" and the button says "request a demo," the visitor should make the connection themselves. Align both with the same benefit.
• The number of competing CTAs. Two primary buttons with the same visual weight on the same screen divide attention. Keep one primary CTA and, if necessary, a more discreet secondary link.
• The arrival page. If the button promises "24 hours" and the confirmation page announces "within 72 hours", you lose the trust gained by clicking.

This point has a direct consequence on the test: if you modify the button and the headline at the same time, you'll be testing a combination, not a CTA. When two elements interact, a multivariate test can make sense; the page A/B test or multivariate test explains when to switch from one to the other and why multivariate testing requires much more traffic.

Color, size, position: what the tests really say

No color converts better in absolute terms: what matters is the contrast of the button with the rest of the page. “Red versus green” studies have been circulating for years, but their outcome depends entirely on the surrounding design.

In the Aspiration test, red and blue yielded the same business result for identical text. A red button on a red page disappears; a green button on a gray page stands out. Before testing a color, ask yourself one question: Is the button the most contrasting element on the screen? If so, the color test is unlikely to produce a measurable difference with your traffic.

On positioning, HubSpot reports an observation by Aagaard: on a long page, placing the CTA at the bottom, after the argument, increased conversions by 304%. It's not a universal rule. On a short page with a simple offer, a CTA above the fold remains relevant. The logic is the same as for the copy: the button must appear when the visitor has enough information to decide.

Formulate a sound CTA test hypothesis

A good CTA test hypothesis can be written in one sentence: "By replacing X with Y, we think we will increase the metric M, because observation O shows that visitors are hesitant about Z." Without the "because", you are testing randomly.

Concrete example for a quote request form:

“By replacing ‘Submit’ with ‘Receive my quote within 24 hours’, we expect to increase the number of quote requests per visitor, because session recordings show that 40% of visitors fill out the form and then stop at the button.”

The observation can come from a click map, an exit survey, or the support itself. Personalization is a promising avenue: HubSpot measured across more than 330,000 CTAs that versions tailored to the visitor's segment converted 202% better than generic versions. A different CTA for an existing customer and a new visitor is therefore a legitimate hypothesis, provided you have enough traffic per segment.

One hypothesis = one single variable. If you change the verb, the object, and the reassurance micro-line all at once, and the variant wins, you won't know what to generalize to the rest of the site.

How much traffic is needed to prove that a CTA is successful?

To detect a relative increase of 20% on a conversion rate of 3%, approximately 13,000 visitors per variant are needed; for an increase of 10%, more than 50,000. This is what the approximate rule gives n ≈ 16 × p(1 − p) / δ², with a risk of error of 5% and a power of 80%.

Two practical consequences. First, on a low-traffic site, don't test subtle nuances: test clear-cut changes in the promise ("Send" versus "Receive my quote within 24 hours"), the only ones capable of producing a 20 or 30% difference. Second, measure the CTA at the stage closest to conversion that still maintains sufficient volume: if you only have 15 sales per week, the quote request might be a more realistic indicator than revenue.

And don't stop the test as soon as the dashboard displays "significant." Checking the results every day and stopping at the first spike is the number one cause of false winners, detailed in our article on the subject. false positives in A/B testing.

The A/B testing protocol for a CTA, step by step

A reliable CTA test follows six steps: define the objective, calculate the sample size, create a single variant, distribute traffic randomly simultaneously, wait the planned duration, and then analyze on the initially declared metric. Skipping a step produces a number, not proof.

1. State the main objective before launch: form submitted, trial started, purchase. The click can be a secondary metric, never the final arbiter. In BlastOptimize, this is set up in the configuration of the test's conversion goals.
2. Calculate the duration from the table above and your weekly traffic. Round up to full weeks (at least two) to smooth out the differences between Monday and Saturday.
3. Create the variant by changing only the element in the hypothesis. Editing button text shouldn't require a developer: that's the job of thevisual editor.
4. Distribute the traffic 50/50, at the same time. Comparing "the September CTA" to "the October one" conflates the button's effect with seasonality, campaigns, and current events. Also, verify that the variant displays without a flickering effect, which unfairly penalizes the new version; this is the key issue. ultra-fast dissemination of variants.
5. Let it run until the scheduled end., even though one variant seems to be winning after three days.
6. Analyze based on the main metric, then segment (mobile/desktop, new/recurring) for exploratory purposes only. The analytics dashboard shows the gap, the confidence interval and the probability that the variant is actually better.

A losing variant isn't a failure: it teaches you what your visitors don't want to read. Record it in a test log, with the hypothesis and the result, so you don't retest it in six months. The most common pitfalls (stopping early, too many variants, a metric changed along the way) are listed in our article on classic A/B testing mistakes.

Checklist before launching your test

Check each point before starting: if even one is missing, the test result will be questionable.

• ☐ The button text passes the "I want..." test when spoken aloud.
• ☐ The button contains a verb of obtaining and the concrete object of the gain.
• ☐ The promise of the button corresponds exactly to the following page.
• ☐ The title above the button refers to the same benefit.
• ☐ Only one primary CTA per screen.
• ☐ The hypothesis is written in one sentence, with its "because".
• ☐ Only one variable changes between the control and the variant.
• ☐ The primary metric is a conversion, not a click.
• ☐ Sample size and duration (full weeks) are calculated and noted.
• ☐ Traffic is distributed randomly, simultaneously, without the variant flashing.
• ☐ The end date is fixed and will not be brought forward.
• ☐ The result, winning or losing, will be recorded in the test register.

If you lack the time to chain CTA tests by hand, the Autopilot mode in BlastOptimize is designed to automate part of this optimization loop.

FAQ: Frequently asked questions about CTAs

Does a first-person call to action ("my") convert better than a second-person one ("your")?

No general rule guarantees this. The idea is that the visitor reads "Receive my guide" as their own sentence, similar to the "I want…" test. However, the expected effect is small and varies from one audience to another. This is a hypothesis with limited scope: reserve it for high-traffic pages.

Should a CTA test be measured by click-through rate or conversion rate?

The final conversion rate is reported per visitor exposed. The click-through rate can increase while sales or form submissions decrease, as in the Aspiration test where the most clicked button generated three times fewer forms per view.

How long should an A/B test on a button last?

The time required to reach the calculated sample size, rounded to the nearest full week, with a minimum of two weeks. On a site with a 3% conversion rate aiming for +20%, expect approximately 13,000 visitors per variant.

My site has little traffic: does testing a CTA make sense?

Yes, provided you test clear changes in the promise and choose a metric close to the button that retains volume (quote request rather than signature). Avoid tests based on color or verb nuance, which your traffic won't be able to differentiate.

Does urgency ("Only 3 spots left") work on a CTA?

It can increase short-term conversion rates if it's true. If it's false, it constitutes deceptive business practices and erodes the trust of returning visitors. Only test it when the scarcity is genuine, and also monitor the return or cancellation rate.

Can we test multiple CTAs from the same page at the same time?

Yes, if the tests focus on independent areas and non-overlapping audiences; otherwise, the effects will be mixed. If the button and the title interact, a multivariate test is preferable, keeping in mind that it requires significantly more visitors.

Key takeaways

A call to action that converts is written like a precise promise and validated like a scientific hypothesis.

• Write using the formula: verb of obtaining + object of gain + clarification + risk reducer.
• Align the button with the title that precedes it and with the page that follows it.
• Judge a CTA on conversion per visitor, never on click alone.
• Color matters less than contrast; positioning depends on the length of the argument.
• Calculate the traffic needed before launching: at 3% conversion, a 20% increase requires approximately 13,000 visitors per variant.
• Let the test run its course and record each result, including the loser.

To move from theory to your first CTA test, consult the offers and prices BlastOptimize or compare our approach to that of an established player with BlastOptimize facing AB Tasty.