Does a red button sell more?

Short answer

Not on its own. The famous test where red beat green was run on one specific design where red was the only red element on the page, and the person who ran the test explicitly wrote that the result shouldn't be carried over to other sites. What usually makes a button color work isn't a particular hue but how much the button stands out from what surrounds it — and you can only learn that with a properly built test on your own page.

The claim

A recommendation that keeps recurring in digital marketing articles and talks goes like this: make your "Buy" or "Sign up" button red, and your conversion rate will rise. Sometimes the reasoning rests on color psychology: red evokes urgency and excitement, and it prompts action. Sometimes a number gets attached: a red button brought 21 percent more clicks.

The claim's appeal lies in its simplicity. You don't need to redesign a site, rewrite its copy, or change the product to raise its sales; changing a single color code is enough.

Where did it come from?

The most widely cited source for the claim is an A/B test that designer Joshua Porter shared in 2010 at the marketing software company Performable. Porter turned the green button on the company's home page red on a duplicated page and changed nothing else on the page. Over a test that ran a few days and covered more than two thousand visits, the red button performed 21 percent better than the green one. The result was later published on the blog of HubSpot, which went on to acquire Performable, and it circulated for years as proof that "red wins."

But the test itself didn't support that reading. Performable's page was green-dominant; the logo and the design were in shades of green too. The red button stood out because it was the only red element on the page. Porter said as much himself in his write-up: the results held, at most, under the conditions in which they occurred — this page design, this site, and the audience that saw it. He advised his readers not to turn their green buttons red without testing first, and to try colors on their own pages with their own audiences.

Stories from big tech companies also played a part in making button-color tests famous. Douglas Bowman, who led the visual design team at Google, wrote in a departure post in March 2009 that a team had tested 41 shades of blue because it couldn't decide between two blues. Bowman meant this as a criticism of a culture where data decided every design choice; but over time the story began to be retold as proof that "color testing earns you millions."

What does the evidence say?

The evidence shows the claim runs into trouble on three separate points.

The first is contrast. Whether a button gets clicked depends first on whether it gets noticed, and what determines that is less the button's color itself than how much it differs from the colors around it. A red button stands out on a green page; on a red-dominant page, the red button disappears and a green or blue button might stand out instead. How a single accent color guides the eye on a page is the subject of GT-03.

The second is statistics. One version coming out ahead in a test doesn't mean the difference is real; in small samples, chance can produce large differences. Consider a hypothetical example: two versions each get a thousand visitors, 50 people click in the green version and 60 in the red version. Red looks 20 percent "better." But a difference of this size could easily arise from chance alone even if the two versions were truly equal; looked at with a standard statistical test, the difference isn't significant. If the same rates had instead been observed across hundreds of thousands of visitors, the difference would become reliable.

In a major reference on running experiments, Ron Kohavi, Diane Tang, and Ya Xu invoke a principle known as "Twyman's law" for results like this: any figure that looks interesting or different is usually wrong. The authors recommend double-checking especially striking positive results and running them through validity tests.

The third is time. A site's loyal visitors notice when an element changes and may give it more attention for a while simply because it's new. That attention fades as it becomes familiar. This is why a short test can make a new color look like a winner, only for the difference to vanish a few weeks later.

A properly built button test accounts for all three traps from the start. The behavior to measure is chosen before the test: clicks, purchases, or signups? The number of visitors needed is also calculated before the test, and the test isn't stopped before reaching that number no matter how exciting the interim results look. The test is run for at least one or two full weeks so that weekday versus weekend visitors and the fading of novelty are both accounted for. Before changing the color, the button's contrast with its surroundings is also measured, so that when the result comes in, you can answer the question "did red win, or did the more distinct one win?" In the end, a meaningful gain is only turned into a lasting design decision once it has been replicated at least once more.

The shared upshot of these three problems is that the finding from one button color test cannot be carried over to another site. A different site has different colors, so its contrast is different too; its audience, traffic, and visitors' habits differ as well. A blanket statement like "red brings 21 percent more" presents a result measured once, on a single page, as if it were a universal law.

Verdict — False

"A red button brings more sales" is false as a general rule. A red button really can perform better on a specific page, but the reason is usually not red itself — it's the contrast it creates on that page. Even the author of the most widely cited evidence asked that the result not be generalized.

Example — two pages, two different winners

A cosmetics brand is testing two different pages for its foundation products. The first page is designed with a white background and light green accents; here, an orange-red button stands out and gets clicked more. The second page is filled with large photos of models in different skin tones; warm coffee, honey, and pink tones dominate the photos. On this page, the same orange-red button can get lost among the warm tones in the photos; a dark navy button stands out more clearly instead.

Same brand, same product, same button; two different results. This example shows that a button's color only takes on meaning together with its surroundings. It's also a reminder that, when working with images featuring a range of skin tones, choosing a button color based on a single light-skinned model's photo can produce unexpected results in the page's other versions.

Common misconception

Summary

1. The famous test in which the red button gained 21 percent was run on a single green-dominant page, and its author explicitly asked that the result not be generalized.
2. What usually decides button tests isn't hue but contrast; small samples, a lack of statistical significance, and the novelty effect can also produce misleading winners.
3. The result of one button color test cannot be carried over to another site; every page requires its own test with its own colors, audience, and traffic.

Test yourself

1. (Easy) In an A/B test, how many elements should differ between the two versions?
2. (Easy) Why might the red button have stood out in the Performable test?
3. (Medium) Why is it a mistake to generalize the result of one button color test to another site?
4. (Medium) How can the novelty effect mislead the result of a short test?
5. (Hard) In a test with a thousand visitors, the red version comes out 20 percent better. Why shouldn't this result be trusted right away, and what should be done instead?
6. (Hard) How does Twyman's law suggest a marketing team should approach a striking test result?

Answers

1. Only one; that way, any difference in the result can be attributed to that element ("The claim").
2. The page was green-dominant and the red button was the only red element on the page; in other words, contrast stood out more than red itself ("Where did it come from?").
3. A different site has different colors, and therefore different contrast; its audience, traffic, and habits differ too. A result measured once on one page may not hold outside those conditions; the test's own author stated this explicitly as well ("What does the evidence say?").
4. Loyal visitors may give a changed element more attention for a while simply because it's new; if the test ends during that period, the new version can look like a winner, but the difference can vanish once the interest fades ("What does the evidence say?").
5. A difference this size in a sample this small could arise from chance alone even if the two versions were truly equal; the difference might not be statistically significant. The sample should be enlarged, the test should run long enough, and it should be replicated if possible ("What does the evidence say?").
6. It suggests assuming that an interesting or unexpected-looking figure is usually wrong, and double-checking and running validity tests especially on striking positive results ("What does the evidence say?").

Sources

• Porter, J., "Button Color Test: Red Beats Green," Performable blog, 2010.
• Bowman, D., "Goodbye, Google," Stopdesign, March 20, 2009.
• Kohavi, R., Tang, D., and Xu, Y., Trustworthy Online Controlled Experiments: A Practical Guide to A/B Testing, Cambridge University Press, 2020.

Next step

GT-03 turns this lesson's contrast finding into a design method: it covers the single-accent-color principle, the power of neutral space, and the blur test for predicting where the eye will go on a page. PS-05 returns to red itself: it weighs claims like athletes in red winning more often and students seeing red doing worse on exams, and what attempts to replicate those studies have found.