Accessible video in shops: subtitles and transcripts

A product video explains in ninety seconds what takes three paragraphs of text. Except: anyone who cannot hear the audio sees lips moving. Anyone who cannot see the screen hears music and a casual "up here on the right". And anyone on a train mutes the sound anyway. Subtitles and transcripts are therefore not an add-on for a small group – they are the part of the video that carries meaning without sound and without picture. Since the German Accessibility Act took effect, they are also a legal requirement for newly published shop videos. This article shows what the standards actually demand, how far everyday practice is from that, and what a video page looks like that serves both routes. Our accessibility optimisation covers the implementation.

Who cannot follow the audio

The World Health Organization counts 430 million people worldwide (WHO) with disabling hearing loss, measured as more than 35 decibels of loss in the better hearing ear. That is not a fringe group but more than five percent of the world's population. And the figure is not standing still: by 2050 the same source expects around 2.5 billion people (WHO) to have some degree of hearing loss. A shop publishing videos without a text track today is building in a barrier that affects more people every year.

The link with age is the most important lever for retail. Among people aged 60 and over, more than 25 percent (WHO) live with disabling hearing loss. Anyone addressing an older, financially strong audience is therefore talking about a substantial share of their target group, not an exception. The economic dimension underlines this: unaddressed hearing loss causes annual global costs of nearly one trillion US dollars (WHO) according to the same account.

For the German market it becomes even more tangible. At the end of 2025, just over 7.8 million people (Federal Statistical Office) with a severe disability lived here, meaning a degree of disability of at least 50 with a valid pass. Relative to the population that is 9.4 percent (Federal Statistical Office), close to one in ten people in the country. That figure covers only officially certified severe disability; the number of people with milder but everyday-relevant impairments is higher.

The breakdown is where it gets interesting. For 4 percent (Federal Statistical Office) of people with severe disabilities the most severe impairment is hearing loss, a balance disorder or a speech disorder – exactly the group for whom subtitles make the difference. For another 4 percent (Federal Statistical Office) it is blindness or a visual impairment; for them the subtitle is not what counts, but the transcript and the audio description. And because disability mostly develops over the course of life, just over a third, or 34 percent (Federal Statistical Office), of people with severe disabilities were 75 or older on the reporting date.

Subtitles stopped being a niche feature long ago

The most common false assumption is that subtitles only serve those who need them. The BBC's editorial guidelines contradict that with a measurement of their own. Around 10 percent of the broadcast audience turn subtitles on regularly, and for some online content the share rises to as much as 35 percent (BBC). The majority of these viewers have no hearing impairment – they are in a noisy environment, sharing a room, less familiar with an accent, or simply reading faster than anyone speaks.

Subtitles are primarily aimed at people with hearing loss but are used by a much wider group: around 10 percent of the broadcast audience turn them on regularly, and for some online content it is 35 percent.

BBC Subtitle Guidelines, version 1.2.5 (paraphrased)

The number does not come from a survey but from an analysis of actual playback. A one-week sample in March 2016 showed subtitle usage of around 18 percent across all content on the BBC's on-demand service (BBC Research & Development), with tablets above 20 percent. This is the oldest source in the article, and for a sober reason: no newer published measurement of this kind exists, and the current editorial guidelines carry the same order of magnitude forward.

That moving images are part of everyday customer life can be shown independently: 33 percent (Eurostat) of internet users in the EU had a subscription for film, series or sports streaming in 2025. Anyone who watches television with subtitles every day expects them in a shop too. How strongly video supports sales is described in our look at product videos and conversion – the reach of that effect depends on whether the video remains understandable without sound.

• Ambient noise: platform, workshop, open-plan office – the audio is there but unusable.
• Consideration: in a waiting room or next to sleeping children, the sound stays off.
• Second language: written text is easier for many people to absorb than speech.
• Technical terms: part numbers, material names and dimensions are read more reliably than heard.
• Searchability: only what exists as text can be found again inside a video.
• Hardware: a broken speaker or a silent autoplay rule makes any audio track worthless.

What WCAG and EN 301 549 actually require

The requirements are older than the German act and more precise than the debate suggests. Captions for prerecorded video sit in success criterion 1.2.2 at conformance level A (W3C), the lowest level. Captions are required for all prerecorded audio content in synchronised media – the only exception is where the media itself is already an alternative to text and is clearly labelled as such. An explainer video that supplements a text page does not fall under that exception.

The transcript is not an optional extra either. Success criterion 1.2.3, also level A (W3C), requires either an audio description or a media alternative for synchronised media. That alternative is the full text: everything spoken plus a description of what happens on screen and matters for comprehension. So anyone shying away from audio description can replace it at level A with a clean transcript – a route many shops underestimate because it needs no recording studio.

At the next level up, that is no longer enough. Success criterion 1.2.5 at level AA (W3C) requires audio description for all prerecorded video content in synchronised media. Because the market aims at level AA in practice, this is where the cost of a video is decided: either the narration is planned from the outset to describe the picture as well, or a second audio track is produced. We broke the criteria and their changes down in our overview of the new WCAG 2.2 success criteria.

The European standard that practice orients itself by does not invent this anew. For web pages, clause 9.1.2.2 (ETSI/CEN/CENELEC) points directly at WCAG criterion 1.2.2, and clause 9.1.2.5 (ETSI/CEN/CENELEC) at 1.2.5. What is genuinely its own is the synchronisation requirement: under clause 7.1.2, captions in prerecorded material must appear within 100 ms (ETSI/CEN/CENELEC) of the associated timestamp. That is the value against which every machine-generated subtitle set should be measured before it goes live. A formal presumption of conformity, however, is tied by section 4 of the Accessibility Strengthening Act to harmonised standards whose reference has been published in the Official Journal of the European Union, and for Directive (EU) 2019/882 that publication is still outstanding. The regulation under the act points instead to the state of the art and to the lists issued by the Federal Agency for Accessibility. EN 301 549 is therefore the written benchmark practice works with, not a legally presumed conformity.

The act draws the line at the publication date

The rule that decides matters for video is not in an annex but in the scope itself. Under section 1 (4) number 1, prerecorded time-based media published before 28 June 2025 (Accessibility Strengthening Act) are excluded. The date is therefore a publication date and not a retrofit deadline: the archive video from 2023 stays outside, today's new product video falls inside. Anyone who re-edits an old video and publishes it again leaves the exception – a point regularly overlooked during relaunches.

The second exception is narrower than many hope. Under section 2 number 17, a microenterprise is a business that employs fewer than ten people and either has an annual turnover of at most 2 million euros or an annual balance sheet total of at most 2 million euros (Accessibility Strengthening Act). Both conditions must be met, and the exception applies only to services, not to products. We worked through how the line is drawn in B2B trade in our analysis of the exemptions for microenterprises.

The framework for breaches is set. Anyone offering or providing a service contrary to the requirements commits an administrative offence that can be penalised with a fine of up to 100,000 euros (Accessibility Strengthening Act). The lower ceiling of 10,000 euros applies to the remaining offences the act lists individually, not to offering a service that is not accessible. More significant in practice than the amount is that market surveillance acts on its own initiative and consumers can file applications. Our review of enforcement in 2026 shows how that has developed.

The transitional rule is frequently misunderstood. Under section 38 (1), service providers may continue to provide their services until 27 June 2030 (Accessibility Strengthening Act) using products they lawfully used before 28 June 2025. That concerns deployed products and legacy contracts, not the content itself. A video player installed in 2024 may keep running; the video uploaded in 2026 still needs captions.

What is actually shipped on the web

Between requirement and delivery there is a gap that can be quantified. According to the analysis of the HTTP Archive crawl, 7.7 percent of desktop pages (Web Almanac) contain at least one video element, and 6.7 percent on mobile. Video is therefore no longer a special case but a regular page component. And the share is growing: the mobile figure is 32 percent higher (Web Almanac) than in the 2022 survey, in relative terms.

The matching caption rate is the real finding. Only 0.5 percent of pages with a video element (Web Almanac) use a track element at all, and 0.65 percent on mobile. That does not mean nearly all videos run without captions – media embedded via iframe are not counted, and some portals ship their own tracks. It does mean that the native route, visible in the markup and therefore testable, is used almost nowhere. And that is exactly the route the standard envisages.

There is an aggravating pattern on top. 23 percent of video elements carry the autoplay attribute, but only 8 percent carry muted (Web Almanac). A video that starts by itself and emits sound is an imposition for hearing users and, for everyone else, an information source that stays silent. Anyone using autoplay needs muted, visible controls and captions – otherwise the start carries no information, only motion.

The overall state fits the picture. The annual automated test of the one million most-visited home pages found detectable WCAG failures on 95.9 percent (WebAIM) of pages, averaging 56.1 errors per page (WebAIM). Automated tests cover only part of the criteria, so the real situation is likely worse rather than better. In practice that means: those who measure will find something – and those who do not measure still have findings.

One partial finding from the same survey is particularly telling: embedded video from a large video platform was detected on 6.2 percent of home pages, and those pages carried 9.4 more errors than average (WebAIM). Embedded video therefore does not automatically improve a page's accessibility; on average it worsens it. Anyone hoping for an overlay tool instead should know the risks we gathered in our look at accessibility overlays.

Captions that stay readable in the time available

Producing a subtitle track is technically easy today. Producing one that can be read is not. The BBC's editorial guidelines give a recommended subtitle rate of 160 to 180 words per minute (BBC), which works out at 0.33 to 0.375 seconds per word. Anyone displaying faster produces text the eye can no longer catch – and the subtitle becomes decoration instead of information.

That implies a minimum display time any tool can be measured against: a subtitle should stay on screen for at least around 0.3 seconds per word (BBC), so roughly 1.2 seconds for a four-word subtitle. This is the metric that lets you check a machine-generated file in a few minutes: block duration divided by word count. Where the value falls short, the block should be split or shortened, not accelerated after the fact.

• At most two lines per cue, broken at a sensible point in the sentence.
• Display time of at least around 0.3 seconds per word, measured across the whole file rather than by sampling.
• Deviation from the audio below 100 milliseconds, as European standard EN 301 549 sets out.
• Mark speaker changes so a dialogue is not read as a monologue.
• Write down sounds that matter for comprehension, such as a catch clicking into place.
• Spell out numbers, dimensions and part numbers instead of abbreviating them.
• Ship the file with the correct language tag, otherwise the screen reader picks the wrong voice.

WEBVTT
Kind: captions
Language: en

1
00:00:04.120 --> 00:00:07.480
The load capacity filter
sits in the top left of the menu.

2
00:00:07.640 --> 00:00:09.900
[catch clicks into place]

3
00:00:10.050 --> 00:00:13.720
-And if I switch pages?
-Then the selection is kept.

Three details in this file decide its quality. The sound cue is not decoration: anyone who cannot hear the catch click into place will not understand the hand movement being shown. The dash before the spoken lines separates two voices without inventing names. And the line breaks sit at meaning boundaries, not wherever the character counter happens to trip. The same care applies to image content as to alternative text, which we described in our piece on alt text for product images.

The transcript is the other half

A transcript is more than typed-up audio. It is the complete, linear text version of the video: everything spoken, plus a description of what happens on screen and matters for comprehension. Blind users read it with a screen reader. Deafblind users read it on a braille display – the only route by which a video reaches that group at all. And anyone simply short on time skims it in thirty seconds instead of ninety.

The side effect is the strongest one commercially: a transcript turns a video into content that search engines and assistive systems can read. Whether that content is actually retrieved can be checked in the access log – our article on logfile analysis for shops describes how. The same logic applies to manuals and datasheets, which we covered in our piece on accessible documents.

The player has to be operable

Captions and transcripts are of little use if the player cannot be operated. The browser's native video element brings keyboard operation, caption switching and volume control out of the box – a custom player has to reproduce all of it. So it is worth asking whether a bespoke interface is really necessary. If it is, it belongs in proper development and not in a pile of scripts nobody can survey any more.

<figure>
  <video controls preload="metadata"
         poster="/media/gt-400-poster.avif"
         width="1280" height="720">
    <source src="/media/gt-400-1080.mp4" type="video/mp4">
    <track kind="captions" src="/media/gt-400-en.vtt"
           srclang="en" label="English" default>
    <track kind="descriptions" src="/media/gt-400-en-ad.vtt"
           srclang="en" label="Audio description">
    <p>This video cannot be played here.</p>
  </video>
  <figcaption>
    Assembling the GT-400 in 1:48 minutes.
    <a href="/en/media/gt-400-transcript/">Read the full transcript</a>
  </figcaption>
</figure>

• Visible entry to the transcript: a text link directly below the video, not in an unlabelled disclosure drawer.
• Complete keyboard support: play, pause, seek, volume and captions reachable without a mouse, with a visible focus indicator.
• No time pressure: controls must not fade out while focus sits in the control bar.
• Caption contrast: light text on a darkened background so it stays legible over bright picture content.
• No information carried by motion alone: whatever only the animation shows belongs in the text as well.
• Keep delivery testable: the subtitle file has to be served with the correct content type, or the browser silently ignores it.

The first command is the most important and the most frequently forgotten: a 200 status code on the subtitle file proves nothing as long as the content type is wrong. If the server delivers text/plain instead of text/vtt, the browser simply does not show the track – no error message, no console entry. The same goes for load timing: a track that arrives after playback has started arrives too late. How we measure delivery and loading behaviour is described in our PageSpeed optimisation.

For shops with their own app the question arises again on a second surface: what works on the web has to be demonstrated separately in the app, because different assistive technologies apply there. Anyone facing that decision will find the trade-off in our article on the choice between an app and a PWA.

Four weeks to an accessible video page

The effort is usually overestimated because people think of the hardest video first. The reverse order works better: count, then prioritise, then produce. In most shops the bulk of the benefit hangs on a handful of videos sitting on category and product pages with high reach.

1. Week 1 – inventory: list every video with its publication date, embedding method and page views. The date decides the scope, the views decide the order.
2. Week 2 – text version: write the transcript for the highest-reach videos and edit it properly. The subtitle set is derived from the transcript, not the other way round.
3. Week 3 – timing and integration: generate the subtitle file, check display times and synchronisation, add track elements, link the transcript, verify delivery and content type.
4. Week 4 – evidence: walk through with keyboard and screen reader, record findings per criterion, update the accessibility statement and define the process for new videos.

More important than the four-week plan is the fifth step: the rule applies from now on to every new video, otherwise the backlog grows faster than it is cleared. If you want to know where your shop stands overall, our shop check is the entry point. Part of that order is the question of how long raw files, transcripts and subtitle versions are kept – which fits our article on the deletion concept in a shop.

Treat video as part of the shop fittings

The most convenient error of thinking on this topic is splitting it into duty and extra. Captions would be the duty, the transcript the extra, audio description the luxury. The standards see it differently: captions and the media alternative both sit at level A (W3C) and are therefore equally binding, with audio description following one level later. The sequence still makes practical sense – just as an order of implementation, not a ranking of obligation.

Anyone building a physical shop plans the ramp before the floor goes down. With video the arithmetic is the same: the transcript emerges almost incidentally during production, because the script gets written anyway. Retrofitting is expensive, because someone has to sit through a finished video. Make the text version a fixed part of video production and you pay the effort once and save it on every further video – while keeping your own accessibility statement at a level you can evidence. If you would like to work through the inventory in an orderly way, get in touch.