The caption decides what the photograph means

A photograph is ambiguous until something fixes it

A picture of a man standing in an empty workshop at night can be a portrait of failure, of dedication, of solitude, or of a man who has locked himself out. The image contains all of these. What the reader takes from it is decided by the words next to it.

This gives the caption two different jobs, and conflating them is why most captions are wasted.

1. Fixing the meaning. The image has several possible readings and the caption selects one. “Anil, closing up after a 14-hour day” closes off failure and solitude and leaves dedication.
2. Adding what the image cannot contain. A photograph cannot show a date, a name, a price, a consequence or a cause. “The same lathe, four months after the workshop flooded” adds a fact that no arrangement of pixels could have carried.

The second is where captions earn their space. The first is necessary but cheap.

The wasted caption restates

A photograph of a lathe. Caption: “A lathe.”

This is extremely common and it is a real cost, not merely a missed opportunity. The reader’s eye goes to the caption (captions are among the most-read text on any page, because they are short and attached to something interesting) and it returns nothing. You have spent your most reliable piece of attention on a label.

The test is the same one used for images: cover the picture, read the caption; cover the caption, look at the picture. If the caption tells you nothing new, rewrite it.

Alt text is a different job again

Alt text is not a caption. A caption is for everybody. Alt text is read aloud by a screen reader to somebody who cannot see the image at all, and its job is to convey what a sighted reader gets from looking.

The differences in practice:

• Describe, do not editorialise. “Man in a workshop at night, standing beside a lathe, coat still on”, not “a moving portrait of dedication”.
• Do not begin with “image of” or “photo of”. The screen reader has already announced that it is an image; you are making the listener sit through it twice.
• Length follows function. A decorative image takes empty alt text (alt=""), which tells the screen reader to skip it: far better than making somebody listen to a description of a background texture. An image carrying information needs as much text as the information takes, which for a chart can be a paragraph.
• If the caption already describes it, the alt text can be short or empty, because otherwise the listener hears the same thing twice.

Every major platform now has an alt-text field, usually two taps away, and almost nobody fills it. It costs fifteen seconds.

Numbers, names and dates belong in the caption

A reliable habit for documentary and commercial work alike: every caption contains at least one checkable thing: a name, a date, a place, a quantity. This does two things. It forces the caption past restatement, and it puts the specifics somewhere they survive, because a caption is what gets quoted, indexed and remembered after the image itself has been separated from the article.

The order people read in

On a page with a headline, an image, a caption and body text, readers do not go top to bottom. A common path is image, then caption, then headline, then the first line of body text, then out. That is one reason captions are worth more than their size suggests: for a meaningful share of your audience, the caption is not a supporting detail, it is the second thing they read and sometimes the last.

Write them as though that is true, because for some readers it is.

The limitation worth naming

Fixing an image’s meaning with words is a power that can be used dishonestly, and it does not require lying. A true caption placed against a true photograph can produce a false impression, because the caption directs the reader towards one reading and away from another that is equally supported by the picture.

A protest photograph cropped to a confrontation, captioned truthfully with the date and place, will read as a riot; the same photograph captioned with the size of the crowd will read as a large peaceful gathering with an incident in it. Nothing in either caption is false.

There is no technique that resolves this, because there is no neutral caption. What you can do is ask the question directly: what would somebody who was there say about this pairing? If the answer is that they would feel misrepresented, you have your answer, regardless of whether each individual word could be defended.