Does page speed affect conversion rate? What the primary studies actually show, and what they don’t.

By Zach Wennstedt, founder · Published October 22, 2026 · 12 minute read

Does page speed really change conversion rate?

It does, and the direction has never been in doubt. What is in doubt is the size, because almost every number you have seen comes from one of three kinds of evidence, and they prove different things.

• Controlled experiments split real traffic between a faster and a slower version of the same page. Only these show cause and effect. There are few of them, and the best public ones are published by Google’s web.dev team.
• Large correlational studies compare many sites or stores and find that faster ones convert better on average. They are useful for size estimates, but fast sites also tend to be better sites in other ways.
• Vendor claims and press releases that once had a study behind them and now mostly circulate as quotes. Some were real; few can be checked today.

The rest of this article keeps those three apart, because a conversion decision built on a press-release number and one built on an A/B test deserve different levels of confidence.

What did the largest page speed studies find?

Milliseconds Make Millions (Deloitte for Google, 2020)

The most quoted modern study. Deloitte and Fifty-Five analyzed 37 brands across retail, luxury, travel and lead generation in Europe and the US, about 30 million sessions over four weeks, and modelled how small speed changes related to each step of the journey. The headline: “a mere 0.1s change in load time can influence every step of the user journey.”

Two cautions the report itself makes clear if you read the method section. The speed data came from Lighthouse lab tests, and the relationship was estimated with a regression, so these are associations rather than experiments. And the lead-generation results were about engagement (bounce rate down 8.3% on informational pages, page views up 7%), not form submissions. If you run a service business, the retail conversion number is not your number.

Portent site speed study (2022)

Portent looked at 20 sites (6 ecommerce, 14 B2B lead generation) and more than 27,000 landing pages over 30 days. For ecommerce, conversion fell steeply over the first few seconds:

For B2B lead generation the article reports a conversion rate of almost 40% at one second, 34% at two seconds and 29% at three. We left the four-second row out of the chart because the article’s own worked example for it does not match its stated rate. The bigger limitation is size: 20 sites and an unnamed load metric. It is a good illustration of shape, not a benchmark for your industry.

Shopify store speed and conversion (2026)

The newest large dataset, and the one with the clearest method. Shopify compared actively selling stores over a 28-day window in early 2026, bucketed them by Core Web Vitals, compared median conversion, excluded the slowest 5% and controlled each metric for the other two. Its findings:

• “For every 100 milliseconds slower a store loads, conversion tends to be about 3.5% lower” (measured by Largest Contentful Paint).
• “Stores with 2.5 second LCP report roughly 30% lower conversion than stores with 1.5 second LCP.”
• “For every 32 milliseconds slower a store responds to interactions, conversion tends to drop by about 1.5%” (Interaction to Next Paint).
• Layout shift (CLS) did not follow the same pattern.

The author calls it correlation, not causation, which is the honest framing. It is also ecommerce on one platform; a plumber’s quote form is a different decision. But as an estimate of slope, it is the best public number available in 2026, which is why our calculator below uses it.

Which speed results are experiments, and which are correlations?

Google’s web.dev case studies are the closest thing to experiments in public. They are company-reported and selected because they worked, so they are not a random sample, but several used a proper split test.

Read the table top to bottom and a consistent picture appears. Where speed was the only thing changed (Vodafone, Rakuten 24), the gains were real and sizable. Where it was estimated across many sites, the slopes are in the same family. Nothing here says speed is the biggest lever on every site, but nothing says it is a myth either.

Which famous page speed statistics have no source you can check?

Several of the most repeated numbers have lost their trail. That does not make them false; it makes them unsuitable for a business case.

• “A 100 ms delay cuts conversions by 7%.” It traces to an Akamai press release from April 2017 based on SOASTA data. The release exists, but we could not open its text tonight, and later coverage says the 7% applied to smartphones with smaller figures for tablets and desktops. Quote it with that context or not at all.
• “Bounce probability rises 32% as load time goes from 1 to 3 seconds.” Published by Think with Google from 2017 Google/SOASTA research. Google has since retired the page; every link to it now redirects to a general hub. The figure may be fine, but nobody citing it today can show you the source.
• “Amazon loses 1% of sales per 100 ms,” “Walmart gained 2% per second.” These come from conference talks over a decade ago. We found no primary document for either.
• “1 second delay equals 7% fewer conversions.” A garbled version of the Akamai 100 ms claim, ten times larger than the original. You will find it in many roundups.

If a proposal for speed work leans on these, ask for the numbers above instead, or better, for a test on your own traffic.

Which speed metric matters most for conversions?

Google’s Core Web Vitals define “good” as LCP within 2.5 seconds, INP of 200 milliseconds or less and CLS of 0.1 or less, measured at the 75th percentile of real page loads on mobile and desktop. In the 2026 Shopify data, LCP had the clearest relationship with conversion, INP a real but smaller one per millisecond, and CLS no consistent pattern. That fits intuition: a page that has not shown its main content cannot persuade anyone, and a button that lags makes a visitor doubt the form went through.

Most sites still miss at least one threshold. HTTP Archive’s 2025 Web Almanac (July 2025 data, published January 2026) found:

LCP is the metric most sites fail, and it is also the one most closely tied to conversion in the newest data. That makes it the place to start. If you want the engineering detail on the interaction side, the Denver SEO team’s explainer on Interaction to Next Paint walks through how INP is measured and fixed.

How much is page speed worth on your site?

Here is the calculator. It multiplies your traffic, conversion rate and value per conversion by the Shopify slope, then shows a cautious version at half the slope. Use your own field LCP (PageSpeed Insights shows it when Chrome has enough data for your site), not a lab score.

Two ways to read the result. If even the cautious number is small next to the cost of the work, speed is not your first conversion project; look at the offer, the headline and the form first. If the cautious number pays for the work inside a year, do it, and measure it properly (section 8) so the next decision rests on your data.

When is page speed not your real conversion problem?

Speed is a multiplier on everything else. A fast page with an unclear offer converts a small number quickly. Three situations where we would put speed second:

• You already pass LCP comfortably. Going from 1.8 s to 1.5 s is worth less than fixing a headline that does not say what you do or where. The five-second homepage test is a faster check.
• The form or the phone is the leak. A slow page loses people before they read; a long form loses them after they decide. Our write-up on form fields and completion rate covers that side.
• Your traffic is mostly returning or branded. Visitors who already chose you tolerate a slower page. Speed matters most for first-time visitors from search and ads, who are comparing you with the next result.

How do you prove speed’s effect on your own site?

The Vodafone design is the gold standard and is more achievable than it sounds. Build the faster version of one high-traffic landing page, split paid traffic 50/50 between old and new for two to four weeks, and compare conversion rate with a significance check. Because only speed changed, the difference is attributable to speed.

If a split test is not practical, use a before-and-after with guardrails: record two to four weeks of baseline conversion rate and field LCP by device, make the speed change and nothing else on those pages, then compare the same number of weeks, and check that traffic mix and seasonality did not shift. Segment mobile and desktop separately, since mobile is where most sites are slow. Our A/B testing page explains how long a test needs to run for the result to mean anything.

What should a service business fix first?

For the typical service-business site (a homepage, service pages, a quote form, a call button), the slow parts are almost always the same, in roughly this order of payoff:

1. The hero image. Usually the LCP element. Serve it in a modern format at the size it displays, give it width and height, and do not lazy-load it.
2. Third-party scripts. Chat widgets, review carousels, heatmaps, multiple tag managers and old pixels. Each adds main-thread work that slows both LCP and INP. Remove what no one reads and load the rest after the page is interactive.
3. Fonts. Two families and a few weights, preloaded or system fonts, with a fallback that does not shift the layout.
4. Hosting and caching. A slow server response pushes every other metric back. A cache or CDN in front of the site is usually the cheapest fix of all.
5. The form itself. Heavy form plugins and spam tools can make the submit button lag. The form is where a slow interaction costs you the lead.

Then measure again with field data, not just a lab score. Speed work that does not move the 75th-percentile LCP your visitors experience has not happened yet, whatever the lab score says.