Caption Builder

A caption is rarely read the way it was written. The writer composes it top to bottom, as one piece. The reader meets a fragment: a picture, a line or two of text, and a fold. Whatever sits above that fold is the whole advertisement for whatever sits below it. If the opening does not earn the tap, the careful middle and the thoughtful ending might as well not exist.

This is not a guess about attention spans. It is a description of how long text is displayed. The US Federal Trade Commission, explaining where a sponsorship disclosure has to sit, describes a stream in which longer descriptions are truncated so that only the first two or three lines show, and a reader has to click “more” to see the rest. The agency was writing about advertising law, but it gave the clearest public statement of the layout problem every caption writer faces.

The opening line is the whole advertisement

Treat the first line as a headline that also has to be a sentence. It should say what the post is, in words the reader already uses. A hook that opens with a handle, a run of tags or a string of emoji spends the most valuable space on the page on things that are not the message. Plain-language guidance from the US government makes the same point about all public writing: one idea per sentence, shorter words, the active voice. Its archived page on short sentences puts it bluntly: long, complicated sentences often mean the writer is not sure what they want to say.

Capitals are a tempting shortcut for urgency. The same plain-language archive warns that putting everything in capital letters makes text harder to read, and that online it reads as shouting. A hook in capitals is louder and slower, which is the wrong trade at the one point where speed matters.

The cut is not a fixed number of characters. It falls wherever the second or third line runs out of room, and that depends on the width of the screen, the size of the type and the shapes of the letters. A line of narrow letters holds more characters than a line of wide ones. Chinese is written without spaces between words, so a line can end between almost any two characters, while English waits for a space.

Here is a worked example. The hook “Our studio opens its doors to all this Saturday” is 47 characters long. Measured in a common system font at 15 pixels, it is about 312 px wide. In a column 280 px wide it wraps onto a second line. Put a one-word disclosure on the line after the hook, and with two lines showing, that disclosure lands on line three, after the cut. Widen the column to 340 px and the hook fits on one line, so the disclosure moves up to line two and both show.

Nothing about the words changed. Only the width did, and with it the answer to whether a reader sees the disclosure without opening the post.

That is why a character count is the wrong tool for this job. The preview here measures the rendered width of each candidate line with the browser’s own text measurement, which returns the width of a string in the current font. It breaks lines the way the CSS text rules describe: after a run of spaces, with spaces at the end of a line taking no room, and inside a word only when the word is too long for the line. Counting is done in graphemes, the characters a reader perceives, so a family emoji built from five code points counts once.

We checked that approach against the browser itself. For 400 random captions, mixing words, hyphens, slashes, links, emoji and Chinese and Japanese text, we computed the cut at three widths and at one, two and three visible lines, then laid the same text out in a page element and read off where each line actually ended. The result: 3,600 of 3,600 cut points matched. A version that cut by character count instead, using the average width of a letter, matched 1,199 of them.

Structure that survives being cut off

If only the first lines are certain to be read, the order of the parts matters more than their polish. Put the point first. Put the one thing a reader needs to know before they decide, such as a date, a price condition or a warning, where it will show. Leave the texture, the backstory and the credits for later, where a reader who wants them will find them.

A caption with a clear structure also survives a reader who skims. A hook, a short body in short paragraphs, one call to action, then credits and tags at the end, is easy to scan from either end. A caption that asks for a save, a share, a comment and a visit asks for nothing in particular. One clear next step beats four competing ones, and the tool flags a call to action that runs to more than one sentence.

Tags belong at the end, not in the middle of a sentence. The US government’s Section 508 guidance on social media notes that screen readers may read a multi-word tag as one long word unless each word starts with a capital, so a tag dropped mid-sentence can turn a readable line into noise for someone listening. The same guidance says screen readers describe every emoji one by one, so several in a row slow the post down, and it suggests placing emoji at natural stopping points rather than between sentences.

Where a disclosure has to sit to count

The FTC does not dictate the words of a disclosure, but it is specific about placement. Its guide for people who post endorsements says to place a disclosure where it is hard to miss, and lists the places where it is likely to be missed: only on a profile page, at the end of a post, or anywhere that requires a person to click “more”. It also says not to mix a disclosure into a group of tags or links.

Its question-and-answer guide goes further on the truncated view. If a post makes an endorsement at the beginning of the description, any required disclosure should be visible without clicking “more”. Starting a post with “Ad:” or “Advertisement” would likely be effective, and “#ad”, “Ad:” and “ad” at the beginning of a post are probably all equal. A disclosure in the middle or at the end is easier to miss, especially at the end of a long post or among other tags.

On wording, the agency names simple terms such as “advertisement”, “ad” and “sponsored” as often enough, and warns against vague terms such as “sp”, “spon” or “collab”, or stand-alone words like “thanks” or “ambassador”. The tool offers the plain words, accepts your own, and flags the vague ones. It then checks the disclosure against the measured cut at every width you set, because “before the more link” is a question about layout, not about where the words appear in your draft.

None of this is legal advice, and none of it is a guarantee. It is the regulator’s published guidance, read on the date at the top of this page, applied as a placement rule you can see.

Alt text is part of the post

The picture is usually the first thing anyone notices, and for a reader using a screen reader, alt text is the picture. Section 508 guidance on alternative text says it should communicate the same information as the image, describe what the image conveys rather than how it looks, and not repeat what the text already says. Its example of unhelpful alt text opens with “Black and white photo of”, which is why the tool asks you to say what the picture shows instead of that it is a picture.

The W3C decision tree for alt text starts with two questions worth asking of every post. What is the image for? If it is pure decoration, its alt text should be empty. Is there text in it? If the words in the picture appear nowhere else, they belong in the alt text, word for word. The W3C tips add that alt text should be the most concise description possible, and that anything needing more than a short phrase or sentence belongs elsewhere.

That last point matters for captions. A caption and its alt text work as a pair, and the pair is only useful if each says something the other does not. The caption gives the story and the context. The alt text gives what the picture shows that the caption leaves out. The tool measures the longest run of words the two share and flags one of six words or more.

The line-break workaround and what it costs

Blank lines between paragraphs make a caption easier to read. The spacer trick is a workaround for posting boxes that drop empty lines: put one visible character on each otherwise empty line, so the line is no longer empty and survives. A common choice is U+2800, which the Unicode Character Database names BRAILLE PATTERN BLANK and classes as a symbol, not a space.

The cost is that the character is real. It is counted against your length, it is copied with the post, and a screen reader meets it. Section 508 guidance notes that screen readers may struggle to identify special characters. Try plain line breaks first, and if they survive where you post, use them. If you do use a spacer, the tool lists every character it inserted, by code point, so nothing invisible slips into the post.

What the preview cannot tell you

The preview is an approximation, labelled as one. It measures text in your device’s system font, at the widths and size you set. The place you post may use another font, other margins and its own rules about where to cut. A service may also cut by characters rather than by lines. The widths, the line count and every length budget in the tool are therefore your settings, not anyone’s published limits.

The reading time is a rough figure too. It divides the words by a rate you choose. The default of 213 words per minute is the reading-aloud rate from an early study by Huey, as reported in the encyclopedia entry on words per minute, which gives 355 words per minute for silent reading in the same tests. Readers vary widely, and the figure says nothing about whether anyone will read.

Used that way, as a check on structure rather than a promise of results, the preview answers one question well: when the post is cut short, what is left?