Captions and Alt Text for Your Photos

An AI tool can write a first draft of alt text and captions. It can also get the picture wrong. So look at the photo and read the draft side by side, every time, before anything goes on your site.

Two different jobs

A caption sits under a picture and is for everyone. It tells the story: who, where, why it matters.

Alt text is a short description hidden in the page's code. It's for people who can't see the image, and for software. Google's documentation calls alt text the most important attribute for giving more metadata about an image.¹ Wikipedia adds that screen readers and text-based web browsers read the alt attribute in place of the image.²

A screen reader is software that reads a page aloud to a person who can't see it. So alt text is what that person hears instead of your photo.

What good alt text does

Wikipedia says alt text conveys the meaning and intent of the image, not a literal description of it.² On a portfolio, the meaning is often the subject and the setting.

Google advises you to "focus on creating useful, information-rich content", and gives a clear example: "Dalmatian puppy playing fetch".¹ It marks a string of stuffed keywords as a poor choice.

• Say what the picture shows, plainly and briefly.
• Match the page. A wedding page may need the moment, not the camera angle.
• Skip "image of" and "photo of". The reader already knows.
• Don't pack in search words.
• Leave decorative images with an empty description. Wikipedia refers to "images that convey no information, but are purely decorative" for that case.²

What a tool can and can't see

A tool that reads images can name objects, scenes and colours. It can't know who the people are, which venue it was, or why the moment matters. Only you know that.

It may also describe things that aren't there. The OAIC names inaccurate output as a known risk with generative AI.³ A drafted line might say "smiling guests" when nobody smiles, or name a place that's wrong.

A check for every draft

• Is every object named in the text visible in the photo?
• Are the people's relationships, names and ages stated only if you know them?
• Is the place correct?
• Does the text avoid guesses about feelings, health or background?
• Is it short enough to listen to?
• Would it make sense to someone who can't see the picture?

Any no means an edit. Sometimes it means deleting the whole line and writing it again from scratch, which is fine. It takes a minute.

Privacy before you upload

Uploading a client's photo to a public AI tool shares that person's image with another company. The OAIC points out that many photographs of individuals "contain sensitive information".³ Its advice for Australian organisations is to avoid entering personal information into public generative AI tools.

Get your client's permission first, and read the maker's privacy page. Elsewhere, look to your own regulator. Or describe the photo to the tool in words and leave the image out.

Writing the caption

Give the tool your own notes: names you have permission to use, the place, the date, and why the moment matters. Ask for one or two sentences. Tell it not to add facts.

"Using only these notes, write a one-sentence caption. Don't add names, places or feelings I haven't given."

Then rewrite it so it sounds like you. Captions should add what the eye can't see, such as who was there and what happened next.

Questions before paying

• Does the tool let you review every description before it's saved to your site?
• Can it work from your notes rather than the image alone?
• What does the maker say about keeping uploaded images? Read their page and note the date.
• Does a free plan let you test on photos that have no people in them?