A third of a second
Most web pages already move, for about a third of a second at a time. Generative tools now make long motion cheap as well, but the eye answers to what starts moving, and starts are what stay scarce.
· 4 min read
Motion on a web page costs more to make than words or a photograph, the forecast goes, so it is saved for ads, launches and the tops of home pages. Once a model can make a bespoke animation for the price of a stock image, that reason disappears, and every page can move.
Most pages already move. In the HTTP Archive's 2022 crawl, the CSS animation property appeared on 77% of mobile pages, the same share as the year before, and the transition property on 85%.1 The motion is small. The median transition lasts 300 milliseconds, the median animation has two keyframes, and the commonest animation names describe rotations, bounces, slides and fades.1 That is a menu opening, a button answering a tap, a loader turning. It became the default once a line of stylesheet could produce it, with no motion designer and no generative model. For interface motion, the forecast came true before its cause arrived.
What is still rare is motion as content, a clip or an animated drawing, anything that runs for seconds. In 2024 one mobile page in fifteen (6.7%) contained a video element, while the median page held 13 images; video had grown by about a third since 2022.2 More than half of those videos run under 30 seconds, and the crawl's authors read a rise in autoplaying inline video as short clips doing the job animated GIFs used to do.2 This is the kind of motion generative video makes cheap. Google sells video from one of its generation models at about ten cents per second of 720p output, a dollar for a ten-second clip.3 The reader still pays to receive it. In 2025 the median video file on a mobile page was 384 kilobytes, against 2.56 megabytes for the whole median mobile home page.4 Motion written as code weighs a few lines of stylesheet, and motion delivered as footage costs every reader the download. If production cost is what kept content motion rare, it should now spread the way interface motion did.
Motion has been nearly free before. The blink tag made text flash on and off for the price of a single tag; it became much maligned for overuse, many people found it annoying, and browsers no longer support it.5 The HTML standard lists blink and the scrolling marquee among elements that are entirely obsolete and must not be used by authors, and calls CSS transitions and animations a more appropriate mechanism for moving content.6 The standard dropped two elements that kept text moving, and what took their place, on three pages in four, is mostly a transition lasting a fraction of a second.
In a laboratory, letters that were moving drew no more attention than letters standing still, while items that had just begun to move did, even when the motion told the viewer nothing.7 The authors, Richard Abrams and Shawn Christ, describe a relative insensitivity to ongoing motion.7 Among news headlines, banner ads slowed people looking for a headline they had just been shown, and animated banners slowed them no more than still versions of the same ads; afterwards people recognized the still banners better than chance and the animated ones at chance.8 That study had 24 participants, all aged 19 to 22.8 Narrow as they are, the two results place motion's hold on a reader in its first instant. For most readers a loop that keeps going becomes wallpaper, and each start takes the eye whether or not anything worth seeing has begun. Motion the reader starts begins where the reader is already looking, so its pull costs nothing; motion that starts by itself pulls the eye from the text. Cheap production multiplies the second kind of start and gives the reader no more attention to spend on them.
Some readers pay far more than attention. The web's accessibility guidelines require a way to pause, stop or hide anything that starts moving by itself, lasts more than five seconds and runs alongside other content; five seconds was chosen as long enough to catch a user's attention and short enough to wait out.9 Motion triggered by scrolling or other interaction can bring dizziness, nausea and migraine to people with vestibular disorders, potentially enough to need bed rest.10 About half of the pages the HTTP Archive crawls now include a check for whether the reader has asked their device to reduce motion, up from 34% in 2022, and the share barely moved in 2025.1112 How many readers have actually asked is not publicly measured.
Long motion earns its cost in one place, where the thing shown is itself a movement. Across 26 studies, animations taught better than still pictures by a medium margin overall, somewhat more when the animation depicted the subject than when it decorated it, and by a large margin when the lesson was a procedure or a physical skill.13 Cheap generated motion has an obvious use there, in showing how a knot is tied or how a part fits.
This site publishes animations with its articles.1415 Claude, the model that drafted these sentences, is Anthropic's. Both have a stake in what the web does next. Away from processes, the animation-first web is likely to look like the one that already exists, extended. Generated motion should spread in the shape interface motion already has, short, begun by the reader's arrival or action, light enough to write as code, and then still. If cost stops limiting motion, the limit that remains is the reader's supply of beginnings. The most valuable thing an animation can do, after it starts, is stop.
Sources
- 1
- 2
- 3
- 4
- 5
- 6
- 7
- 8
- 9
- 10
- 11
- 12
- 13
- 14
- 15
About this piece
- Kind
- Research essay. An argument grounded in published research.
- Evidence
- Moderate
- Research
- Standard, 15 sources
- Made
- My idea, researched and drafted with AI. Every sourced claim was checked against its source, with an AI model, before I published it.
- Reviewed
- Oct 29, 2026