What it does
A caption and an alt attribute both sit next to a photograph and both consist of words, which is roughly where the similarity ends. One is visible and read by everybody. The other is invisible and read instead of the picture by the people who cannot see it. Write them as though they were interchangeable and you get a page where one group hears the same sentence twice and the other learns nothing new from either.
This route runs the same engine as the alt-text page with a different ruleset loaded. The checks that only make sense for something spoken by a screen reader are switched off. The checks about whether a line is finished writing stay on, and the one check that matters most on this route — whether the caption and the alt attribute are the same string — is the reason there is a field for the other one.
The order of operations
Open the picture, or do not; a caption can be written from memory and the checker does not need the file. If you do open one, the measurements are the same ones the alt-text page takes — shape, tone, colour, edge density — and they are as useful here for the same reason, which is that they are the facts people stop and squint at rather than facts that require recognition.
Then write, and read the findings as they appear. Put the alt attribute in the second field if the image has one. If the two strings match, the checker will say so, and it will say it as a problem rather than a note, because a duplicated announcement is a cost paid by someone who has no way to skip it.
The same photograph, captioned two ways
As writtenA woman standing on a jetty looking at the sea.
InsteadAnna Whitfield at Mousehole, three days before the harbour wall was closed.
The first sentence is a good alt attribute and a wasted caption: everyone reading it can already see a woman on a jetty. A caption's whole budget should go on what the photograph cannot say for itself.
As writtenImage showing the new bridge, photo by the council.
InsteadThe new bridge carries the coast path over the estuary for the first time since 1996.
Opening by naming the picture spends the first four words on nothing, and the credit belongs in a credit line rather than inside the sentence. What the reader wanted was the fact in the second version.
Two things this route will not do
- It will not draft the caption. A caption's value is almost entirely in the facts that are not visible in the frame — the name, the date, the reason — and those are not in the pixels for anything to find. Even a working describer would only be able to restate the picture, which is precisely the caption nobody needs.
- It will not tell you whether the caption is true. The checker reads the shape of the sentence: whether it is finished, whether it duplicates the alt text, whether it runs past what a reader will sit through. Names, dates and places are facts about the world and this page has no access to the world.
Questions people ask
- If the caption is right there, does the image still need alt text?
- Usually yes, and it should not be the same sentence. A caption is page text: it is announced as part of the document, in reading order, after the image. So a screen reader that meets an image whose alt repeats the caption reads the sentence, then reads it again. The honest options are an alt that says something the caption does not, or an empty alt where the caption genuinely carries everything.
- How long should a caption be?
- Longer than alt text is allowed to be, and shorter than you think. There is no accessibility ceiling because nothing about a caption is an attribute — it is a paragraph in a smaller face. The practical ceiling is attention: a caption is read in the gap between looking at the picture and returning to the article, and two sentences is most of what fits in that gap.
- Should a caption say what is in the picture?
- Only as much as it takes to get to the part that is not. The reader can already see the picture; what they cannot see is the year, the place, the name of the thing, or why it is on this page. A caption that describes the visible and stops has spent its one paragraph restating what was already free.
- Does the checker grade a caption the same way?
- No. The redundant-prefix rule relaxes to a note, because a caption that opens by naming the picture is clumsy rather than broken. The length allowance rises. The screen-reader checks about announcement order drop out entirely. What stays is everything about whether it reads as finished writing, because that is what a caption is.
Where to take this next
- Alt textWrite the alt attribute yourself, with the picture open and the findings live.
- Long descriptionThe paragraph a chart or a diagram needs when one attribute cannot hold it.
- Alt text checkerPaste alt text you already have and read the findings. No picture required.
- What a screen reader announcesThe order NVDA, JAWS and VoiceOver read an image in, and what they add themselves.