What actually converts and holds traffic on websites?

Evidence. Deloitte Ireland LLP published Milliseconds Make Millions on 24 March 2020, commissioned by Google and based on data provided by Fifty-Five. Deloitte analysed mobile data from 37 brand sites — retail, travel, luxury, and lead generation brands across Europe and the US — covering over 30 million sessions over 4 weeks. Based on a 0.1 second improvement across 4 mobile speed metrics, with every page in the journey improving by that amount, it observed retail conversions up 8.4% and average order value up 9.2%, travel conversions up 10.1%, luxury page views per session up 8.6%, and lead generation information pages’ bounce rate improved by 8.3%. The design was observational, so the percentages describe a correlation and do not promise the same lift on any single site.

What this means for a small operator. Measure the mobile load of pages that carry enquiries or sales, fix the largest delays first, and re-measure.

Evidence. Google’s web.dev collection, The business impact of Core Web Vitals, reports that Vodafone in Italy improved Largest Contentful Paint (LCP, the time until the main content appears) by 31% and achieved 8% more sales, using server-side rendering, reduced render-blocking JavaScript, and image optimisation.

Rakuten 24, a Japanese online store, ran a one-month A/B test reported on web.dev. Half of traffic went to a page optimised for Core Web Vitals and half to the original, with no other functional or visual differences. The optimised version produced 53.37% more revenue per visitor, a 33.13% higher conversion rate, and a 35.12% lower exit rate. Rakuten started from a page with room to improve, so read the size as tied to that starting point and the direction as the lesson.

What this means for a small operator. A controlled test on your own highest-value page gives a number you can trust for your own business.

Evidence. Portent analysed about 100 million page views across 20 B2B and B2C sites. Its April 2022 update reported e-commerce conversion rates of 3.05% at a one-second load, 1.68% at two seconds, and 1.12% at three seconds (chart above). Portent’s own summary states that a one-second site converts at 2.5 times the rate of a five-second site for e-commerce, and 3 times for B2B (5 times at ten seconds for B2B).

Google and SOASTA’s 2017 mobile benchmarks, published through Think with Google, used a model trained on mobile visit data. It estimated bounce probability rises by 32% as load time grows from one to three seconds, by 90% from one to five seconds, and by 123% from one to ten seconds. These are relative model estimates: a 32% rise applied to a 20% bounce rate gives 26.4%. The page puts the model’s prediction accuracy at 90%.

Akamai’s spring 2017 State of Online Retail Performance report, built on SOASTA data from about 10 billion retail visits over one month, reported a 100-millisecond delay hurt conversion rates by up to 7.1% on smartphones, 3.8% on tablets, and 2.4% on desktop (Akamai’s press release rounds the top figure to 7%), and a two-second delay increased bounce rates by up to 103%. Akamai sells performance services and had just acquired SOASTA, so read this as a vendor’s report on its own customers’ data.

What this means for a small operator. The first three seconds carry the largest measured losses. Fix the slowest money pages first.

Evidence. Google’s web.dev documentation defines 3 Core Web Vitals, judged at the 75th percentile of real page loads (the value that 75 of 100 measured visits came in under), separately for mobile and desktop: LCP of 2.5 seconds or less, Interaction to Next Paint (INP, how quickly the page responds to a tap or click) of 200 milliseconds or less, and CLS (Cumulative Layout Shift, how much a page’s content jumps around while loading) of 0.1 or less. INP replaced First Input Delay on 12 March 2024. The HTTP Archive’s Web Almanac 2025 Performance chapter, using July 2025 CrUX data, puts the share of origins passing all three Core Web Vitals at 48% for mobile and 56% for desktop, which means 52% of mobile origins did not pass in that measurement. The 2024 edition reported 43% for mobile and 54% for desktop; the 2025 edition restates the 2024 mobile figure as 44%.

Google Search Central’s page experience documentation states Core Web Vitals are used by Google’s ranking systems, and that Google seeks to show the most relevant content even if page experience is sub-par. Page experience helps most when competing pages offer similar content.

What this means for a small operator. PageSpeed Insights reports the same 3 numbers for any page, free. Passing all three on mobile puts a small site ahead of roughly half the measured web.

Evidence. Backlinko analysed about 4 million Google search results. Its chart reports average click-through rates of 27.6% for position one, 15.8% for position two, and 11.0% for position three. The three sum to 54.4%, matching the study’s reported share for the top 3 combined, and the position-two value is cross-checked by Backlinko’s own text: moving from position two to position one gives 74.5% more clicks, and 27.6 / 1.745 = 15.8. Position one earned about 1.75 times the rate of position two in that dataset. Backlinko is owned by Semrush, which also provided the data. The dataset predates AI summaries, so read it as a pre-AI-summary baseline.

What this means for a small operator. Pages already ranking on the first page are usually the cheapest traffic available. Improve those first, and check their click-through rate in Search Console against their position.

Observed browsing. Pew Research Center published its analysis on 22 July 2025, tracking 900 US adults during March 2025 across 68,879 unique Google searches, of which 12,593 produced an AI summary — 18.3%, matching Pew’s published 18%. Users clicked a traditional result in 8% of visits to pages with a summary, against 15% without one — a recomputed 46.7% fewer clicks. They clicked a link inside the summary itself in 1% of visits. They ended their session after 26% of pages with a summary, against 16% without — a recomputed 62.5% higher rate, or 10 percentage points. Summaries appeared on 60% of queries beginning with question words, against 8% of one- or two-word searches. Pew compared different kinds of searches, so the full difference cannot be assigned to the summary alone.

Keyword data. Ahrefs published an update on 4 February 2026, using December 2025 data for 300,000 keywords (150,000 with an AI Overview, 150,000 informational without one) and aggregated Google Search Console desktop data. Position-one click-through rate for AI Overview keywords fell from 0.073 in December 2023 to 0.016 in December 2025. Ahrefs forecast 0.03734 without AI Overviews against an actual 0.015677, and reported a 58% lower rate for the top-ranking page. Recomputing gives 58.0%. Ahrefs describes the result as a correlation.

Client accounts. Seer Interactive’s September 2025 update covered 3,119 informational terms across 42 organisations, with 25.1 million organic impressions from June 2024 to September 2025. Organic click-through rate on queries with an AI Overview fell from 1.76% to 0.61%. Seer reports a 61% decline; recomputing from Seer’s own published endpoints gives 65.3%, and Seer does not explain that gap on the page, so quote the endpoints, not the 61%. Paid rate fell from 19.70% to 6.34% (Seer reports 68%; recomputing gives 67.8%). Brands cited inside an AI Overview earned a 35% higher organic rate than brands not cited (0.70% against 0.52%, Q3 2025 averages). Seer states citation may not cause the higher rates, because stronger brands are more likely to be cited. Queries without an AI Overview still fell, reaching 1.62% in September 2025.

What this means for a small operator. Rankings alone no longer predict traffic on question-style searches. Measure success in enquiries, sales, and Search Console clicks. That recommendation is an inference from the three studies above.

Evidence. Nielsen Norman Group published How Long Do Users Stay on Web Pages? in September 2011. Its summary states users often leave web pages in 10–20 seconds, and pages with a clear value proposition can hold attention for much longer. Its advice is to communicate the value proposition clearly within 10 seconds. The study predates the current mobile web, so treat the seconds as a guide to the shape of behaviour: the risk of leaving is highest at the start.

What this means for a small operator. The first screen has one job: say what this is, who it serves, and what the visitor can do next, in plain words.

Evidence. Baymard Institute’s average of documented cart abandonment studies is 70.22% across 50 studies (Baymard’s own page, checked November 2026). The complement is 29.78%: fewer than 3 in 10 carts became orders. Baymard’s survey of US shoppers who abandoned during checkout, excluding shoppers who were just browsing, reports 40% citing extra costs such as shipping, tax, and fees as too high, 20% citing slow delivery, 19% citing distrust of the site with card details, and 18% citing a forced account. Shoppers could give more than one reason. Baymard also estimates that the average large e-commerce site can raise conversion by 35.26% through better checkout design. Applying that lesson to a contact form or booking step is an inference: the measured behaviour is about checkouts, and the shared element is friction at the moment a visitor acts.

What this means for a small operator. Show the full price early. Let people act without creating an account. Keep the form to fields you will actually use.