Blog

Nobody Owns Subtitle Placement

November 21, 2026  ·  Tablingo

The US caption rule has a subparagraph literally titled Placement, and it stops at television. Online, the only standard is “no worse than TV”, plus one non-binding note.

Pause a news clip at the wrong second. A lower third slides in with a name, a title, sometimes a place. The subtitles have been parked at the bottom of the frame the whole time, and they land on top of it. Two pieces of text, one of them unreadable.

The reaction is always the same: it looks bad. That is true about taste and useless about design, because nothing follows from it. Ask a narrower question — what did the subtitle actually cover? — and the answers can at least be checked.

One frame, three collisions

A subtitle can cover three different things, and they are not one problem in three costumes.

It can cover on-screen text: a name, a score, a date, a company. That text is usually information you needed anyway, and if the speech is in a language you do not read, the on-screen text may be in that same language. Replacing it with a translation of the audio is covering one piece of information with another.

It can cover a mouth. For a viewer who reads lips, the mouth is not decoration; it is a second copy of the same sentence, and sometimes the clearer one.

It can cover the action: a goal, a card trick, a chart the speaker is pointing at. Nothing in the audio describes it, and nothing afterwards brings it back.

CoveredWhat is lostWho notices
A name or a dateText you also readAnyone reading
A speaker’s mouthLip-readingDeaf viewers
A score or a chartThe contextEveryone, later

Name the collision and the question changes from “is this ugly” to “who was supposed to prevent it”. That is where a rule would be, if there were one.

The rule called Placement stops at television

American caption regulation does have a paragraph about where a subtitle sits, and its title is the word. 47 CFR §79.1(j) is headed “Captioning quality obligation; standards”, and its second subparagraph opens by naming four properties: “Captioning shall be accurate, synchronous, complete, and appropriately placed as those terms are defined herein.”

Then it defines them, one at a time: accuracy, synchronicity, completeness, placement. The fourth is specific in a way the other three are not.

“Captioning shall be viewable and shall not block other important visual content on the screen, including, but not limited to, character faces, featured text (e.g., weather or other news updates, graphics and credits), and other information that is essential to understanding a program’s content when the closed captioning feature is activated. Caption font shall be sized appropriately for legibility. Lines of caption shall not overlap one another and captions shall be adequately positioned so that they do not run off the edge of the video screen.”

The things it names are the three collisions, in order: faces, on-screen text, and “other information that is essential”. Then two clauses nobody quotes — lines of caption must not overlap each other, and the text must not run off the frame. Even the boring failure modes got written down.

This is the only place in American regulation where a subtitle’s position is a requirement rather than a preference. It is also a television rule: §79.1 puts the duty on video programming distributors, meaning the stations and cable operators that carry a channel.

Online: no worse than TV

§79.4 covers video delivered over the internet. It does not restate those four standards. What it does is compare. An owner must:

“Send program files to video programming distributors and providers with captions as required by this section, with at least the same quality as the television captions provided for the same programming.”

And a distributor must pass them on while “maintaining the quality of the captions provided by the video programming owner”.

Read that twice. The online obligation is not that captions be well placed. It is that they be no worse than the ones television had. The standard is borrowed, and a borrowed standard only exists where the thing it borrows from exists.

So for a programme that went out on American television with captions, a streaming version may not degrade them, and the placement paragraph above becomes binding at one remove. For a programme that never did, there is nothing to compare against — and because the television rule exempts anything whose audio is not English or Spanish, that is most foreign-language video. The chain and where it ends is its own article.

This is the sentence most write-ups get wrong. Quoting (j)(2)(iv) at a streaming service quotes a television rule at a party the rule does not cover. The accurate complaint is narrower and harder to file: not that the captions are badly placed, but that they are worse than the broadcast was.

If you do want to file it, the window is short. §79.4(e)(1) takes written complaints “within sixty (60) days after the date the complainant experienced a problem with captioning”, and (e)(3) gives the distributor thirty days to answer. A real path — for programming the section covers, about a defect it can measure.

The accessibility standard says it once, in a note

The Web Content Accessibility Guidelines are what an accessibility audit is measured against, and they almost never mention position. Success Criteria 1.2.2 and 1.2.4 ask whether captions exist for pre-recorded and for live audio. Neither asks where they sit.

The one sentence that does exist is not a success criterion. It is a note in the glossary entry for captions: “Captions should not obscure or obstruct relevant information in the video.”

In WCAG a note is explanatory. Conformance is judged against success criteria, and a note tells you what a criterion means rather than adding a requirement of its own. So the standard contains a sentence that describes the right outcome, and no mechanism for failing anyone who misses it.

That is not a flaw waiting to be fixed. Placement is expensive to test — an auditor has to sit through the video and judge a frame — while “captions present” is a checkbox. The standard did not decide the question. It declined it.

Two style guides, two units

What is left is the part of the industry that writes this down for money: the style guides. Two are public, and they disagree, because they are not answering the same question.

The BBC’s guidelines give percentages. For 16:9 video, “subtitles should not be placed outside the central 90% vertically and the central 75% horizontally”; for 9:16, “this is reversed: … the central 75% vertically and the central 90% horizontally”. Then the default position, which is not a percentage at all: “The normally accepted position for subtitles is towards the bottom of the screen (Teletext lines 20 and 22. Line 18 is used if three subtitle lines are required).”

Teletext lines. The number comes from a page layout that stopped being how anyone watches television, and it survives because the rule was written while it still was. The instruction underneath it is the part that matters — avoid “on-screen captions, any part of a speaker’s mouth or any other important activity” — and it names the same three things as the American regulation, counted in a different unit.

Netflix’s guide gives a preference. Subtitles “should be center justified and placed at either the top or bottom of the screen, except for Japanese, where vertical positioning is allowed”; where both ends are occupied, “the subtitle should be placed where easier to read”. No coordinates at all.

DocumentCounts inBinds
§79.1(j)(2)(iv)InformationTelevision
WCAG noteOne sentenceNo one
BBC §10Percent and lineIts own output
Netflix §5PreferenceIts deliveries

Three kinds of document, four examples, four units. A regulator wrote a floor for broadcasters; a working group wrote a meaning for auditors; a broadcaster wrote a house style for its own output; a licensor wrote a delivery spec for its vendors. None of them was writing the rule for the video in your tab.

Why the numbers don’t travel

The BBC numbers look like the portable part of this article, and they are the least portable. Two details break them.

They describe a rendering area, not a sentence. 90% and 75% say where the box may sit inside the frame; they do not say whether this particular line should move because a name just appeared in the middle of it.

And they invert with the frame. The same broadcaster reverses both numbers for portrait video. A rule that flips when the aspect ratio changes is not a property of the text. It is a property of the renderer.

The BBC says as much itself a paragraph later. Where a player can draw subtitles under the picture instead of over it, “vertical displacement will be ignored by the device and only horizontal positioning will be used”. On that player the question in this article disappears, because there is no frame left to cover.

That is the structural difference between timing and placement. Timing lives in the file: a cue has an in-point and an out-point, and every player reads the same numbers. Placement lives in the player. Two people watching one stream can see the same sentence in two different places, and both can be within spec.

A test on a paused frame

If no rule settles it, what is left is a judgement, and a judgement needs a question small enough to answer in a second.

Find the densest frame — the one where a name, a score, a chart or a face shares the screen with a sentence. Pause. Look at the strip where the subtitle sits. Ask one thing: is there anything in that strip I also need to read? Three answers:

None of it needs a tool. A pause and one second of looking, on any video, ours included.

What a tool can promise

Since no rule decides, the decision stays with whoever is watching, which changes what you can sensibly ask of a subtitle tool. Not “avoid the face”: that needs a model of the frame rather than a model of the audio, and ours does not have one. Ask for mechanics instead. Can it move, can it resize, can it switch off.

Ours, stated plainly: five positions — top centre, bottom centre, bottom left, bottom right, and a right-hand sidebar — a size slider for each of the two lines, and an off switch. The overlay is not free-dragged; you pick one of the five, and the sidebar itself can be dragged to an edge. If your video needs the caption at 40% height, clear of a scoreboard in the corner, none of the five is that. The judgement is still yours; the tool’s only job is to be out of the way while you make it.

Which is the honest ending for a question nobody was assigned. Where a subtitle goes is decided by the person in front of the screen, one pause at a time, with one question: is anything in that box something I also needed?

If you want to try ours, it’s called Tablingo.