Can a Website With Only Original Photos Rank on Google Without Text?
I’ll admit I’ve half-believed this one. Google’s models can look at a photo and tell you it shows a Corinthian column, a few tourists, late afternoon light, and a gelato sign in Italian. Gemini will write a better paragraph about your picture than most people write in their captions. So why would a site full of original photography need prose at all? If the machine can read the image, the image is the content.
It’s partly true. Which part depends on which Google you mean.
What Google actually sees in a photo now
Google’s own image documentation says it works out what a picture is about from the alt text, from computer vision, and from the page around it. The vision part is real. The 2024 leak of Google’s internal API docs pointed to stored image attributes for text read off the picture (OCR), how many faces are in the frame and how much space the biggest one takes up, and which colours dominate. Landmark recognition is good. Object detection is good. Lens runs on the same kind of understanding and handles billions of visual searches a month.
Ten years ago an image was a black box with a filename. It isn’t anymore.
Where the image-only site wins
Google Images and Lens are where the idea holds up, and originality matters there in a direct way. When the same photo sits on dozens of sites, Google treats them as copies and picks one page to send people to. Mueller has said this plainly: reuse a common image and ranking in image search gets harder. An original photo has no competition inside that cluster. Google’s “About this image” panel also shows when a picture was first seen, which quietly favours whoever published first.
Lens visual matches work the same way. Someone points a phone at a church in Sicily, Lens looks for similar images, and a sharp original shot of that church can be the one it returns. Discover likes big, strong images too.
So a photographer with two thousand original shots of small European towns can pull real image traffic with very little writing. That’s not a fantasy.
Where it falls apart
Regular web search ranks pages against queries, and most queries are questions. Is the town worth a day trip. When does the market run. Can you park near the old gate. Vision tells Google what’s in the frame. It can’t tell Google what the photographer knows: that the square floods every November, that the restaurant on the left closed last spring, that this is the only angle without scaffolding this year. A page with no words answers one query well (“what does this place look like”) and almost nothing else.
Mueller has also said good images on their own won’t lift a site in web search. That was before the latest vision models, but nothing Google has said since suggests it changed.
Then there’s quality. A page of twenty uncaptioned photos looks thin to Google’s quality systems. There’s no sign of who took them, where, or when. The odd thing is that original photos are one of the strongest first-hand experience signals a site can have. They only count as that if something on the page ties them to a person and a place, and that something is usually a sentence.
Links follow the same logic. People link to pages they can point at and explain. Photographers do get links through credits and embeds, but a bare gallery earns fewer than a page that tells you something.
AI search adds two more problems. AI Overviews and AI Mode build written answers out of pages that have writing in them; your photo might get shown beside the answer while someone else’s text gets the credit. And if a model can describe your image well enough, the searcher may never need to open your page at all.
Last one: scrapers. An image-only page is trivial to copy. A scraper that wraps your photo in four hundred words of filler has more for Google to rank than you do. Google usually picks the original. Usually.
“No text” is never really no text
Even a minimalist gallery ships words. The filename is text. So is the URL, the title tag, the alt attribute, and the IPTC metadata embedded in the file. The real question is how little you can get away with, and the answer is: less than a blog post, more than zero.
Name files for what they show, so cefalu-cathedral-dusk.jpg beats DSC_4471.jpg. Write alt text that says what the picture is for, not a list of keywords. Add a caption of a line or two with the place, the date, and one thing the camera can’t show. Fill in the IPTC creator and copyright fields, since Google Images can display that credit next to your photo. Use real <img> tags with the source in the HTML, because Google doesn’t index CSS background images as images, and heavy JavaScript loading can hide pictures from the crawler. For big galleries, an image sitemap helps Google find everything.
That’s maybe fifteen minutes of work per page. It’s the difference between ranking in image search only and ranking in both.
So is it true?
For image search, mostly yes, and original photos are a genuine edge there. For the main results page, no. Google can see the picture fine now. What it still can’t get from pixels is the stuff only the photographer knows, and that’s what most people are searching for.
Write the caption.