Philosophical: Does Google know when a photo isn't what your meta data says it is? And could you be downgraded for that?

[BradsDeals]

Not something I've ever heard discussed before, probably still a bit too esoteric for present day, but I've always been one to be guided by where I see Google headed rather than trying to game the system as it exists now. So think about it:

• Most stock and public domain photos are used repeatedly throughout the internet.
• Google's reverse image search proves that Google can recognize when the same photo is used across dozens of sites.
• Many of those photos will have alt and/or title text that Google also has crawled. If not it has the content of the page on which the photo exists to consider for context.

So if Google has a TON of clues about what a photo is likely to be about, and can in theory aggregate those clues about a single photo from the dozens of sites using it, how might Google treat a site that mislabels it, old school "one of these things is not like the others" style?

Would a single site hosting that photo be bolstered by the additional context that the known repeated photo brings in, essentially from other sites?

If 10 sites about widgets are using the same widget photo, but the 11th uses an entirely new, never before published photo, would the 11th site then be rated better for bringing something new to the table? (I think this would be almost certainly true, drives home the importance of creating your own graphics content.)

Anyway, like I said, all theoretical and philosophical and probably not currently in play, especially since an image can be used in so many different contexts, but it's New Years and things are slow and my brain is running, so I'm curious what other folks might think about that as the future of image optimization.

[donford]

Hi Rebecca,

I can see this happening in the future for sure, if not already. The new Google reCaptcha already kind of does this, "Select the pictures with tacos", which is kind of like Google saying hey we already know which of these pictures have tacos lets see if you do. They could of course expand the reCaptcha to help identify more pictures if they wanted to.Though that may diverge from the original purpose of captcha which was designed to tackle 2 problems. OCR readers having trouble with certain words / scripts in books, and spammers.

Nice thoughts,

Don

[Lumina]

Certainly an interesting question. It's becoming more and more evident that image recognition software (more specifically, subject recognition) is gaining traction within big names including Facebook and Google. The software (still in development) can recognize subjects, objects, settings, etc. - to the point where they can "name" an image based on these factors. Which, of course, is extremely relevant to this conversation.

That said, I disagree with the notion that incongruities between an image name, alt-text, or title and the recognized subject of that image will have any factor at any point in time. I have two main points on why I suspect this will never become practice:

1. Naming an image based directly on its contents has never been a suggestible convention. Historically, naming an image has been more about the "message" or intended use of that image than about its direct, visual content. To push content creators to start doing this would be overly heavy-handed (yes, even for Google).
2. The web would be utterly polluted by images with the exact same name, all over the place. As you'd brought up stock photography and its proliferation across the web, I'd counter that this is exactly why it won't happen. The amount of images by this convention that would be named "man in suit at laptop" alone is staggering. More to the point, Google and other curators prefer specificity; so much so that it would be impossible for them to accurately define more than the visual assets - which often don't make up the bulk of a pictures meaning.

TL;DR version: Do I think what you're suggesting is possible? Absolutely. Do I think it will happen? No; this would go against naming conventions and Google's own desire for specificity.

[BradsDeals]

I agree with you about naming convention. I'm thinking more about alt text, title attributes, on-page context.

But I think it would be difficult to figure out if an image is being used in an unusual way. Say you have a photo of a Hawaiian sunset. What are you using that for? Maybe a travel site. Maybe a page of inspirational quotes. Maybe a church. Maybe a massage therapist. Maybe a Hawaii-themed restaurant in Oslo. Maybe a funeral home. The appropriate context could vary so much that it would be a tall order.

[BradsDeals]

Good call on the reCaptcha stuff, I hadn't even thought about that. Google is teaching its algo image recognition by asking real humans "so, what exactly is this?" in a sort of backhanded way. And what would that do with that?

I do see a case to make for unique images being more highly valued. If duplicate content is devalued, and images are content, well... ¯_(ツ)_/¯

[EGOL]

I believe that popularity in image search has an impact upon rankings in websearch.  So, if you have produced a unique image that is more popular, then you will benefit from it.  But, if your unique image is not popular then the effect will be neutral.

[DonnaDuncan]

Thought provoking discussion Rebecca!

I'm with you in thinking there is potential for Google to start using misleadingly labeled images in it's ranking algorithm. Alt tags in particular. They're supposed to be used, in part, to help visually impaired search engines and people understand what's being shown on the page. If they don't do that, if they're just stuffed with keywords, they lessen the value of the page. In that context "Hawaiian sunset" has more value that "church", "travel site" or "inspirational quote", even if dozens or hundreds or thousands of other sites use the same descriptor.

I also agree with Egol's opinion that unique content derives value from its perceived popularity; its ability to earn repeat and lengthy visits as well as exposure, links, and shares.

I consider it a best practice to use unique images accurately named and described (using alt tags) with a brief and accurate description of the image that incorporates keywords. Not easy or even possible all of the time, but a good target to aim for.