Usually because nothing on the page explains what they are. A search engine cannot look at a photograph and know it shows your product. It relies on the alt text, the filename, the words around the image, and the page it sits on. If all four are vague, the image has no way to rank.
Image search gets treated as a leftover. Teams optimise the copy, the titles, and the internal links, then upload photos called IMG_4471 with empty alt attributes. For businesses selling something physical, that is a whole traffic channel left switched off.
The work is small and mostly clerical. What makes it worth doing now is that the same signals feeding Google Images also feed the systems deciding whether your page is worth summarising at all.
Image optimization is about weight: compressing files, choosing formats, and serving the right size so pages load fast. Image SEO is about meaning: telling search engines what each picture shows and why it belongs on that page. You need both, and doing one well does nothing for the other.
Confusing the two is common. A team runs everything through a compression tool, watches the page score improve, and assumes image SEO is handled. The files are smaller and just as anonymous as before. Nothing about compression tells anyone what is in the picture.
The two do meet in one place, which is that a slow image can fail before it is ever indexed. The HTTP Archive Web Almanac 2025 puts median image bytes on a mobile home page at 911 KB, which is a lot of weight to carry. We worked through that side of the problem in our guide to optimizing images for web performance.
For the rest of this article we are talking about meaning, not weight. Assume the files are already reasonable and the question is whether anything can understand them.
Because it is the only place you describe the image in words attached to the image itself. Google's Google Images documentation is unambiguous, calling alt text the most important attribute when it comes to providing more metadata for an image. Everything else is context. Alt text is the description.
Google also warns about the obvious abuse. Its guidance is to focus on useful, information-rich content that uses keywords appropriately and is in context, rather than stuffing keywords in. Alt text reading like a list of search terms helps nothing and reads badly to anyone using a screen reader.
The test we use is whether the sentence would make sense read aloud to someone who cannot see the page. That single question fixes most bad alt text, because it forces you to describe what is actually there instead of what you wish the page ranked for.
Purely decorative images are the exception. A background texture or a divider does not need describing, and an empty alt attribute is the correct choice so a screen reader skips it. The mistake is leaving alt empty on images that carry real information. We looked at where machines can help with this in our piece on whether you should let AI write your image alt text.
A little, and they are almost free to get right. Google's documentation says the filename can give Google very light clues about the subject matter of the image, and contrasts a descriptive name with something like IMG00023.JPG. Light clues are still clues, and the cost of naming a file properly is seconds.
Note the word light. Renaming files is not going to rescue an image with no alt text on a page about something else. It is a small supporting signal that helps when everything else already lines up, and treating it as a lever is how people end up with filenames stuffed full of hyphens and keywords.
The practical version is a naming convention applied at upload. Product name, variant, view. Two or three real words separated by hyphens, lower case, no dates or camera codes. Once it is a habit it costs nothing and it makes your own asset library searchable, which is worth more day to day than the SEO benefit.
Google's documentation lists seven. It states that Google Search supports images referenced in the src attribute of img in the following file formats: BMP, GIF, JPEG, PNG, WebP, SVG, and AVIF. If your image is in that list and reachable, format is not what is holding it back.
That makes format a performance decision rather than an indexing one. AVIF and WebP produce smaller files than JPEG at similar quality, which helps your page speed and therefore your rankings indirectly, but choosing one over another does not make an image more findable in image search.
Responsive delivery has one requirement worth remembering. Google's guidance is to use the picture element or srcset attributes and to always specify a fallback URL using the src attribute. Skip the src fallback and you can end up with an image that renders in browsers and is invisible to anything simpler.
Because that is how a search engine works out whether the image belongs there. Google's guidance is to make sure images are placed near relevant text and on pages that are relevant to the image subject matter. An image of a kitchen renovation on a page about accounting software has no supporting context to rank on.
This is why galleries underperform. Twenty images stacked with no words between them give a crawler nothing to attach each one to, so the whole set gets judged on the page as a whole. Breaking a gallery into sections with a line of real text above each one changes what can be understood.
Captions do more work than most people expect. They sit visibly next to the image, they get read by humans, and they add the descriptive text that makes the surrounding context specific. Unlike alt text, a caption is also a chance to say something interesting rather than merely accurate.
Our position is that if an image needs no caption and no nearby text explaining it, it is probably decoration. Decoration is fine, but it should not be expected to earn search traffic, and it should not be counted as content.
An image sitemap lists your images so Google can find ones it might otherwise miss. It is most useful when images load through JavaScript, sit behind a gallery script, or live on a separate domain. Google explicitly allows URLs from other domains in the image loc elements of image sitemaps, which is what makes CDN hosting workable.
That cross-domain allowance is the detail worth knowing. Plenty of sites serve images from a CDN on a different hostname and quietly assume those images are being credited to somebody else. Listing them in your own image sitemap is how you connect them back to your pages.
For a small site where every image sits in plain HTML next to relevant text, an image sitemap adds little. Google can already find those. The value climbs with complexity, which is the same logic that applies to sitemaps generally, covered in our guide to getting your XML sitemap right.
Add licensing metadata. Google's image license documentation says that alongside contentUrl you must include one of creator, creditText, copyrightNotice, or license, and that you must include the license property for your image to be eligible to be shown with the Licensable badge.
There are two ways to supply it. One is structured data on the page. The other is IPTC photo metadata embedded in the file itself, where Google states you must include the Web Statement of Rights field for your image to be eligible for the licensable badge. Either route works, and the embedded version travels with the file.
Google also recommends pairing the license with acquireLicensePage, described as a URL to a page where the user can find information on how to license that image. That is the part that turns the badge into a business outcome rather than a decoration, because it gives an interested buyer somewhere to go.
This is mostly relevant if you sell or license photography. For a typical business site it is not worth the effort, and we would not add it just to have more markup on the page.
Less transparently than Google Search does, and we are not going to pretend otherwise. There is no published specification telling you how ChatGPT, Perplexity, or Google's AI surfaces select and display images. What we do know is that the text signals around an image are what any language model can actually read.
That points to a practical conclusion. A model parsing your page sees the alt text, the caption, the heading above the image, and the paragraph beside it. It does not see the photograph in the way a person does. Every one of the signals in this article is therefore doing double duty.
Google does document one control worth knowing here. The robots meta tag supports max-image-preview, which sets the maximum size of an image preview for your page, and noimageindex, which tells Google not to index the images on a page. Those are the documented levers. Everything beyond them is guesswork.
Our honest advice is to ignore anyone with a confident framework for ranking images in AI answers. Describe your images accurately, put them next to text that explains them, and you have done what is actually knowable today.
Audit alt text on your top twenty pages. That is where the traffic already is, and empty or useless alt attributes are the most common problem we find. Write a real sentence for every image that carries information, and leave alt empty on anything purely decorative.
After that, fix the context. Move images next to the text that explains them, add captions where they help a reader, and split any gallery that is currently a wall of pictures with nothing between them. Filenames come third, applied going forward rather than as a mass rename.
If your images are a real part of what you sell and none of this is happening, we are happy to take a look and tell you where the gaps are. Reach out at phoenix.studio and point us at a page you think should be doing better.
Tell us where you want to go. We'll tell you how we'd get you there.