There is a number from Microsoft’s experimentation programme that ought to be printed above every conversion team’s desk. Reviewing well-designed and well-executed experiments built to improve a key metric, Kohavi, Crook and Longbotham reported in 2009 that only about one in three actually improved it. That finding has held up across other large experimentation platforms since. It means most confident opinions about call-to-action wording, including ours, are wrong two times out of three.
Key takeaways
- Only about one third of well-run experiments improve the metric they targeted.
- Baymard puts average cart abandonment at 70.22%, pooled from 50 studies.
- WCAG 2.2 requires interactive targets of at least 24 by 24 CSS pixels at level AA.
- The EU’s Digital Fairness Act, expected Q4 2026, targets manipulative interface design.
A CTA is a request for commitment, priced in effort
The useful way to think about a call-to-action is as a transaction where the reader pays first. Attention, an email address, a phone number, fifteen minutes of a diary, a card number. Everything else about CTA design follows from making that price legible and proportionate to what is being offered in return.
This is why the classic advice to use strong verbs is true but shallow. “Start” is a better verb than “submit”, and it changes nothing if the reader cannot tell whether starting costs them thirty seconds or a sales call. The wording that reliably outperforms is the wording that answers the unasked question: what happens to me immediately after I click this.
The rewrite that usually does the work
Below are the four patterns we see most often in audits, with what we would replace them with. The right-hand column is the point: in each case the improvement is information, not enthusiasm.
| Common CTA | Stronger version | What changed |
|---|---|---|
| Learn more | See the pricing tiers | Names the destination, so the click is predictable |
| Get started | Create an account, no card needed | States the price of the commitment |
| Request a demo | Book a 20 minute call | Converts a vague obligation into a known one |
| Download the guide | Download the 12 page guide, PDF | Sets expectation of length and format |
None of these is clever. That is deliberate. Cleverness in a CTA transfers cognitive load from the writer to the reader, which is the opposite of what the element exists to do. The place for a distinctive voice is the headline above it, and we have written separately about what makes a headline worth clicking.
Where conversions are actually lost
The evidence on abandonment points consistently away from the button. Baymard Institute’s pooled figure across 50 studies puts average cart abandonment at 70.22%. Asked why they left, 47% of shoppers cited extra costs appearing during checkout, such as shipping, taxes or fees. Close to one in five cited a checkout process that was too long or too complicated, which is unsurprising when Baymard’s benchmark database records an average US checkout flow displaying 23.48 form elements by default.
Two conclusions follow. First, the strongest CTA on the internet cannot survive a cost revealed after the click, because the problem is a broken promise rather than a weak invitation. Second, the compounding wins sit downstream: Baymard estimates, on ten years of large-scale checkout testing, that an average large ecommerce site could gain around a 35% relative increase in conversion rate through better checkout design alone.
The constraints that stopped being stylistic
Two developments have moved parts of CTA design from taste to compliance, and both are recent enough that plenty of design systems have not caught up.
Target size is now a conformance criterion. WCAG 2.2, which became a W3C Recommendation in October 2023, added success criterion 2.5.8 Target Size (Minimum) at level AA: pointer targets must be at least 24 by 24 CSS pixels, measured on the interactive hit area rather than the visible graphic. Five exceptions exist, covering adequate spacing, an equivalent control elsewhere on the page, inline targets constrained by line height, targets sized by the user agent, and cases where a particular presentation is essential or legally required. In the European Union this matters commercially as well as ethically, because the European Accessibility Act has applied since 28 June 2025 to a defined set of products and services, including ecommerce.
Manipulative patterns are heading for explicit regulation. The European Commission’s work programme lists the Digital Fairness Act as a Q4 2026 legislative initiative, aimed at dark patterns, addictive design, influencer marketing and unfair personalisation, with particular attention to minors. It has not been tabled yet, and analysts tracking the file expect negotiation through 2026 and 2027. Nothing is in force. But a countdown timer that resets on refresh, or an opt-out hidden behind a low-contrast link, is now a thing to remove on a schedule rather than defend in a design review.
How to test one without fooling yourself
Given the one-in-three base rate, the point of a CTA test is to protect you from your own confidence. That requires a discipline most teams skip.
- Write down the metric and the expected effect before you start. A prediction recorded afterwards is a story, and stories always fit.
- Check that the traffic can detect the effect you care about. A page with a few hundred weekly conversions cannot resolve a two percent difference, and running the test anyway produces a number that feels like evidence.
- Change one thing. Rewriting the label, the colour and the position at once tells you the combination won, not which part earned it.
- Measure the outcome, not the click. A CTA that lifts click-through and drops completed purchases has made the funnel worse while looking like a success in the report.
- Let it run to the planned end. Stopping when the result looks good is the single most common way teams manufacture wins that never reappear.
Objections we hear most often
Does urgency still work?
Genuine scarcity does. Manufactured scarcity increasingly does not, and it is precisely the category the Digital Fairness Act is being drafted to address. If the deadline is real, state it plainly with the date. If it is not, removing it costs less than defending it later.
How many CTAs should a page carry?
One primary action, repeated as often as the page length justifies, plus at most one secondary path for readers who are not ready. Competing primary actions do not split traffic evenly between them, they add a decision that was not there before.
Should the CTA differ by channel?
The wording should match what the reader can actually do next in that context, which usually means shorter on social and more explicit in email, where the click leaves the platform. The commitment being requested should stay identical across all of them, because a mismatch there reads as a bait and switch even when nothing dishonest was intended.
What we would do first on any page that is underperforming: read the five words after the click before touching the five words on the button. That order is unfashionable, unglamorous and, in our experience, where the recoverable conversions actually are.
The button is one step in a longer sequence
A CTA only performs as well as the path it sits on, which is a structural question rather than a copywriting one.
Sources: Ron Kohavi, Thomas Crook and Roger Longbotham, Online Experimentation at Microsoft, 2009, for the finding that only about one third of well-designed and well-executed experiments improved the key metric they targeted. Baymard Institute for an average documented cart abandonment rate of 70.22% pooled from 50 studies, 47% of shoppers citing extra costs at checkout, close to one in five citing an overly long or complicated checkout, an average US checkout flow of 23.48 form elements displayed by default, and an estimated conversion rate gain of around 35% available to an average large ecommerce site through better checkout design, based on ten years of large-scale checkout testing. W3C, WCAG 2.2, a W3C Recommendation since October 2023, success criterion 2.5.8 Target Size (Minimum), level AA, requiring pointer targets of at least 24 by 24 CSS pixels with five defined exceptions. Directive (EU) 2019/882, the European Accessibility Act, applicable since 28 June 2025. European Commission work programme listing the Digital Fairness Act as a Q4 2026 legislative initiative covering dark patterns, addictive design, influencer marketing and unfair personalisation; the proposal had not been tabled at the time of writing and no obligation flows from it yet. Figures on abandonment describe ecommerce and do not transfer directly to lead generation. Updated August 2026.
