Vertical Video Subtitles: Burned-In Captions Aren't Enough

Captions burned into a vertical video make it watchable with the sound off. They do not make the video accessible. For that, the text must convey all the audio information, stay in sync and exist outside the pixels: according to Orange's digital accessibility guidelines, subtitles "must be in text format and synchronised with the video's audio", not burned into the image. Subtitles and vertical video accessibility are therefore two separate deliverables: kinetic on-screen text for performance, and a proofread .srt or .vtt file for compliance. Most brands pay for the first and think they have bought the second.

Do burned-in captions really make your vertical video accessible?

No. Captions burned into the image make the video watchable without sound, which has nothing to do with accessibility. Burned-in text isn't text: it's pixels. A screen reader can't see it. A user who has set their phone to a larger font size can't enlarge it. A machine translation engine can't pick it up. And nobody can extract it to reread the content at their own pace.

:French public guidelines are clear on this point: subtitles must be attached to the video as a separate text file and must not be displayed directly in the video. It's the sentence that nobody, on the pages covering the subject, connects to the social workflow. Yet it has a direct impact on your quote.

That doesn't mean we're going to stop burning in text. Animated text in 9:16 does part of the retention work, just like the three layers of a 3-second hook. On our e-commerce batches, removing the text layer from a creative systematically causes 3-second retention to drop. Burn-in stays. It simply doesn't replace anything.

TikTok's case illustrates the misunderstanding well. When the platform launched its auto captions in April 2021, it stated in its official announcement that "Auto captions automatically generate subtitles, allowing viewers to read or listen to content", and that viewers can turn them off from the share panel. The result: information carried only by auto captions can disappear during playback. Information carried only by burn-in doesn't disappear, but remains invisible to anyone using a screen reader. Each route has its own gap. That's why we deliver both together.

The rule we have applied to all our batches since 2025: a vertical video is delivered with two text tracks. The burned-in on-screen text, which serves performance. The proofread .srt or .vtt file, which serves accessibility and reuse. Billing for the first while thinking you're selling the second is a quoting error, not a post-production error.

Captions, subtitles, transcripts: what exactly are you buying?

Three words, three different deliverables, and a confusion that derails half of all briefs. In English, captions refer to text in the original language that transcribes the dialogue and meaningful sounds: a slamming door, a ringtone, off-screen laughter. That's the accessibility deliverable. Subtitles, on the other hand, translate the dialogue into another language: commercially useful, but not an accessibility measure. A batch of Reels subtitled in English for the UK market is not an accessible batch.

In France, the reference term is SME: subtitling for the deaf and hard of hearing. The quality charter signed in 2011 under the aegis of the CSA, now Arcom, sets conventions the web has largely forgotten: a colour code by type of speaker, white for the person visible on screen, yellow for an off-screen voice, red for sound effects. Nobody will ask you for that level on a branded TikTok. But the underlying principle applies everywhere: if the audio carries information, the text must carry it too.

The Transcript, the Forgotten Deliverable

A transcript is a linear text document, without timecodes, that reproduces the entire content of the video. It's useless in the feed. It's extremely useful elsewhere: on your product page, in a blog post, on a landing page where the video is embedded, and for answer engines that have no way of reading your creative. Orange's social media guidance is explicit: "All videos must have subtitles and, if possible, a transcript." Real marginal cost when the .srt file already exists: five minutes of copying, pasting and cleaning up. Almost nobody does it.

Where should you place subtitles in 9:16 so they don't end up under the interface?

The trap in 9:16 is the bottom of the frame: all four platforms stack the caption, username, audio, progress bar and navigation there. A subtitle placed "bottom centre" as in cinema disappears under the UI. For everyone, and above all for those who depend on it. It's not an editing problem, it's a framing problem, and therefore a shoot decision, made before the camera rolls.

Here are our internal measurements, taken on an iPhone 15 and a Pixel 8 in October 2026 with up-to-date apps, as a percentage of the height of a 1080 × 1920 frame. These are in-house measurements, not official specifications: platforms move their interface without warning, so redo the measurement twice a year.

Two practical consequences. First, the usable band sits in the lower central third, not at the foot of the frame: on our productions, the text block is locked between 32% and 70% of the height, with an 18% right margin to avoid the column of buttons. Second, the framing must leave this space free from the shoot onwards: a face centred low and a subtitle in the same place means a shot that has to be redone.

LinkedIn deserves a warning: the feed crops vertical videos, and text placed too high gets cut off in the preview. We often make different trade-offs there, because the aspect ratio is chosen by placement, and the choice of ratio mechanically determines the available text area.

Are platforms' auto captions good enough?

Unproofread auto captions are worthless for compliance, and sometimes worse than worthless commercially. Speech recognition has made huge progress in French since 2023; nobody disputes that. But it still fails exactly where it's costly: brand names, product references, dosages, prices, legal notices and negations.

Negation is the case that should worry you. "Do not apply to broken skin" turning into "apply to broken skin" is no longer a typo, it's a reversed message. On our 30-second French batches, we find on average one to three corrections per video in everyday language, and far more as soon as a product name, a figure or an acronym enters the script. Proofreading takes around four to eight minutes per video. That's not much. Yet it's the line everyone cuts.

The "social media" guidance in Orange's framework puts it bluntly: "even if tools can automatically generate subtitles, it is often necessary to check them". The WCAG criteria, for their part, require captions for all pre-recorded video with audio from level A (criterion 1.2.2); inaccurate captions satisfy no criterion, since a conformance criterion concerns accuracy, not presence.

The Real Problem: Auto Captions Aren't a Deliverable

Even when corrected in the app, auto captions remain locked into the platform. You can't export them, reuse them on your website or supply them to a buyer who asks for a file. When a client hands us an "already subtitled" batch produced by a previous agency, we check just one thing: is there an .srt in the folder? Nine times out of ten, there isn't. Everything has to be transcribed again. Allow 10 to 20 minutes per video to make up for what would have cost six minutes at the editing stage.

Which videos should you subtitle, and to what standard?

All those carrying meaningful audio, and the required standard depends on where the video ends up, not where it was shot. A creative that lives only in Meta Ads Manager and a video embedded on your product page aren't subject to the same rules, even though they are often the same master file.

The sheer size of the audience concerned is enough to settle the debate on whether it's worth it. According to the World Health Organization fact sheet on deafness and hearing loss, nearly 2.5 billion people will have some degree of hearing loss by 2050, and at least 700 million will require hearing rehabilitation. That's not a niche. It's a significant share of your paid audience.

The Regulatory Framework, and What We Don't Know

Three texts frame the subject in France. Article 47 of the law of 11 February 2005 requires online public communication services in the public sector and the bodies that depend on it to be accessible, with the RGAA as the technical standard; the criteria in its "multimedia" section explicitly cover synchronised subtitles. Large companies above a certain turnover threshold are also covered. And Directive (EU) 2019/882, known as the European Accessibility Act, has applied since 28 June 2025, with a scope that notably includes e-commerce.

What we don't know, and we'd rather say so: whether a Reel published on a third-party platform falls within the scope of these obligations has never, in our cases, received a clear-cut legal answer. We're not lawyers. What we do observe, however, is that the question comes up more and more often in tenders from major accounts and in public procurement, as a line in the specifications: "supply of subtitle files". Either you have them, or you don't.

The Figure We Refuse to Repeat

You've read everywhere that 80 or 85% of social videos are watched without sound. We looked for a recent, traceable primary source for that percentage. We didn't find one: it has been passed from agency article to agency article since a 2016 study with a very different scope. Sound-off viewing in the feed is a reality our own retention data confirms. Its exact level, no. Be wary of any provider who sells you subtitling with a percentage copied without a URL.

How much does an accessible vertical video cost, and on which line of the quote?

The extra cost of making a 30-second vertical video accessible is between €8 and €15 when it is built into the edit, and between €25 and €60 for full compliance. These are marginal costs recorded across our 2026 production batches, not published prices: they assume the editor already has the project open. Picked up cold, on a batch delivered three months ago, expect to pay two to three times as much.

How we decide in practice. For a high-volume organic TikTok batch, level 4 every time: the file costs ten euros and it belongs to you. For a creative ad meant to run for six weeks in paid media, level 4 as well, and the file is then used to produce variations and translated versions without transcribing again. For a video embedded on your website, a landing page or LinkedIn, level 5: that is where the regulatory framework bites. For a public tender or a key account with an accessibility department, level 5, and the clause goes into the contract before signing.

On a quote, this line must be named. ‘Subtitling’ on its own means nothing: ask for ‘burned-in text overlay’ on one side and ‘proofread subtitle file, .srt or .vtt format’ on the other. If the supplier lumps the two together, you will be sold kinetic text. What to expect from a batch is set out in the 12-line acceptance appendix, and how a short format is priced is detailed in the line-by-line quote for a Reel.

How Do You Quality-Check a Batch of 20 Videos in 15 Minutes?

You don't check 20 videos line by line: you check 20 files in thirty seconds each, then three randomly chosen videos in depth. This is the protocol we apply on delivery, and the one we recommend to our clients for checking their other suppliers.

Step 1, five minutes: open the delivered folder. Are there as many .srt files as .mp4 files? If not, the check stops there and goes back to the supplier. Step 2, ten minutes: pick three videos at random and run through the following eight points.

• The .srt opens in a text editor and contains timecodes in the format 00:00:01,200. A file without timecodes is a transcript, not a subtitle file.
• Brand names, product references, prices and dosages: exact spelling. This is the first place where ASR gets it wrong.
• Negations are present. Search for ‘ne’, ‘pas’ and ‘sans’ in the file and cross-check against the script.
• 40 characters maximum per line, two lines maximum per block. Beyond that, it is unreadable on a 6-inch screen.
• Reading speed of around 15 characters per second. This is the comfort convention we apply; above 20, nobody can keep up.
• Meaningful sounds appear in square brackets: [ringing], [off-screen laughter], [oven beep]. If there are none in the entire file even though the video contains some, the subtitling is incomplete.
• Safe zone: open the published video on a real phone, not in your editing software's player. The text must stay visible beneath the expanded caption.
• Mute test: turn off the platform's captions and mute the sound. Does the video still make sense? If so, the text overlay is doing its job. If not, the critical information exists only in the audio.

This protocol does not replace an accessibility audit carried out by an expert, nor does it claim to. In a quarter of an hour, it eliminates 90% of the defects we see in batches that come to us for rework. And it works all the better when the thresholds are already in the initial brief, following the logic described in the three layers of a useful brief. Checking for things you never asked for has never worked.

What Remains Inaccessible in a Vertical Video, Beyond Subtitles?

The most common case, and the least addressed: the silent video where all the information sits in the on-screen text. A before-and-after for a hair treatment, a furniture assembly demo, a price carousel, all set to royalty-free music without a single spoken word. A blind user scrolling with a screen reader gets nothing out of it. Zero. There is no sound to transcribe and no machine-readable text.

The workaround is not elegant, but it works: repeat the key information in the post caption, and provide a transcript of the visual information in the delivery pack for reuse outside the platform. We now make this mandatory on all ‘text-only’ formats. Five lines of caption, three minutes of writing.

Contrast, Flashes, Synthetic Voices

Three other issues come up in our quality checks. Contrast of burned-in text: pure white on a light background, very common in trendy overlays, becomes unreadable as soon as the subject moves; we add a drop shadow or a semi-opaque banner, not a thin outline. Flashes: fast strobing transitions pose a real risk for people with photosensitivity, and we refuse them above three alternations per second. Finally, TikTok's synthetic voice helps some users but mangles brand names: if you use it for a hook, listen to it before publishing.

A note on Stories, which the Orange guidelines flag as poorly accessible: the very nature of the format, with its fixed duration and automatic sequencing, makes subtitles hard to follow for slow readers. We know of no satisfactory solution. The workaround is not to put any information there that appears nowhere else, and to repeat the message in a permanent post.

The platforms are making progress on their side, but only on the viewing end: automatic captions, enlarged text, contrast settings, text-to-speech. None of this produces the file your legal department will ask you for. The creative side stays on your side of the table, and that is precisely where the question of subtitles and accessibility in vertical video is settled: at the planning stage and at export, not at publication. The day a buyer asks you for your .vtt files for 200 creatives produced last year, the catch-up quote costs ten times what the line would have cost on the original invoice. We priced this twice in 2026. Both times, the client paid.

Frequently Asked Questions

Is an SRT file required if my subtitles are already burned in?

Burned-in text is not text: it cannot be read by a screen reader, resized or exported. Orange's guidelines explicitly call for a separate text file synchronised with the audio. Burned-in text serves retention in the feed; the file serves accessibility and reuse. Deliver both: it's a ten-euro line on a quote.

Are TikTok's automatic captions enough for a brand?

No, not unless they are proofread. TikTok states that its auto-captions generate subtitles automatically and that viewers can turn them off. On our 30-second French batches, we correct an average of one to three errors per video, more on product names and negations. The corrected text also remains locked inside the app.

Where should subtitles go so they don't sit under the interface in 9:16?

In the lower central band, never at the very bottom of the frame. Our October 2026 measurements put the usable zone between 32% and 70% of the height on TikTok, with an 18% right margin to avoid the column of buttons. This zone is reserved at the framing stage, during the shoot, not in the edit.

How much does accessible subtitling cost for a 30-second vertical video?

Between €8 and €15 in marginal cost when the proofread .srt file is produced during the edit, based on our 2026 batches. Allow €25 to €60 for a full compliance pack with a transcript and description of the visual information. Picked up cold, several months after delivery, the cost is two to three times higher.

What's the difference between subtitles, captions and a transcript?

Captions reproduce dialogue and meaningful sounds in the original language: that is the accessibility deliverable. Subtitles translate into another language and are not an accessibility matter. A transcript is linear text without timecodes, useful on a website or a product page. Three different deliverables, three different uses.

Is my company required to subtitle its social media videos?

The public sector and large companies above a turnover threshold fall under Article 47 of the 2005 law and the RGAA. The European Accessibility Act has applied since 28 June 2025, notably to e-commerce. The exact status of a video published on a third-party platform is still debated: have your legal team decide.