For most of photography’s history, a photo either worked or it didn’t. A human looked at it, felt something, and either trusted the brand or scrolled past it. That was the whole test.
Brand photography now has to pass two separate tests: whether a human trusts it at a glance, and whether an AI system can extract what the image actually shows from its file name, alt text, and surrounding page copy. Most brand photography — including genuinely excellent photography — only passes the first test, because alt text, descriptive file names, and accurate captions are treated as an afterthought rather than part of the shoot itself.
That first test still matters. But it is no longer the only one running.
A growing share of the discovery that used to happen through a Google results page now happens through an AI answer — a chatbot summarizing “best options near me,” a shopping assistant comparing vendors, a recommendation engine deciding which three businesses to surface out of thirty. These systems do not “see” a photo the way a person does. They read around it. And most brand photography was never built to be read.
This is a gap I run into constantly in client work: beautifully shot, well-lit, on-brand images that are functionally mute to the systems now doing a meaningful share of the recommending. Here’s what’s actually happening, and the checklist I use to close the gap without touching the art direction.
What Are the Two Tests Every Brand Photo Has to Pass Now?
Every brand image now has to pass a human glance test and a machine-read test — and most photography only prepares for the first.
Test one: the human glance. Does this feel real? Does it match the story the brand is telling? Does it make someone stop scrolling for half a second longer? This is the test photographers have always shot for, and it hasn’t changed.
Test two: the machine read. Can a crawler or an AI model extract what this image is, of what, in what context — without a human ever looking at it? File names, alt text (the short text description read by screen readers and search crawlers when an image itself can’t be displayed or seen), surrounding copy, structured data, and page context all feed this. Google’s own documentation on image indexing is explicit that it “extracts information about the subject matter of the image from the content of the page, including captions and image titles,” alongside alt text and computer vision analysis, according to Google Search Central’s Google Images best practices.
Most brand shoots optimize hard for test one and skip test two entirely. The images get delivered, dropped into a CMS as DSC04821-final-v2.jpg, and published with an empty alt field because nobody on the creative side thought that was their job. It isn’t glamorous work, and it happens after the “real” creative decisions are already made — which is exactly why it gets skipped.
Why Is This a Photography Problem and Not Just a Tech Problem?
A machine-readable caption needs specific, truthful detail — the setting, the people, the product, the moment — and that detail can only be captured during the shoot itself, which means the fix has to start on set with the photographer or creative director, not after the fact in a CMS.
The instinct is to hand this off entirely to a developer or an SEO tool at the end of the process. That’s a mistake. You cannot manufacture specific detail after the fact from a file of unlabeled images — you either captured it in the shoot notes or you’re guessing. A photographer or creative director who knows, while shooting, that image four is going to need to communicate “small-batch skincare production, hands-on quality check, natural light” can leave the frame and the shot list built for that sentence. One who doesn’t know will hand over a folder of gorgeous, undocumented files, and whoever inherits the CMS work will write generic alt text that helps nobody.
In other words: the brands showing up well in AI-driven discovery aren’t necessarily the ones with better photographers. They’re the ones where photography and information architecture were planned as one job instead of two separate ones handed off in sequence.
How Do You Make a Shoot Machine-Legible Without Making It Ugly?
Making a shoot machine-legible takes six habits: writing the caption before the shutter, naming files for meaning, treating alt text as a short honest caption rather than a keyword dump, keeping surrounding copy specific, marking the hero image explicitly, and auditing quarterly.
This is the process I use — nothing exotic, just discipline most shoots skip.
1. Write the caption before the shutter, not after. For every planned shot, write one specific sentence describing subject, setting, and context before shooting it. If you can’t write that sentence in advance, the shot is probably too vague to be useful later either.
2. Name files for meaning, not for the camera. bakery-owner-flour-hands-morning-prep.jpg beats IMG_4821.jpg for the same reason Google says it does in their own documentation: descriptive filenames are a direct, low-effort signal of subject matter.
3. Treat alt text as a second, shorter version of the caption — never a keyword dump. Alt text that reads like a list of search terms actively works against you; specific, accurate description works with both accessibility tools and AI systems.
4. Keep the surrounding page copy honest and specific. An image sitting next to vague marketing copy gets read as vague. An image sitting next to a paragraph naming the real service, real location, and real context gets read accurately.
5. Mark the hero image explicitly. Use primaryImageOfPage or og:image (structured tags that tell a crawler which single image represents the page) so the system isn’t guessing which image represents the page — guessing is where good photos get skipped over in favor of a worse one that happened to be first in the file order.
6. Audit quarterly, not once. Recommendation engines change what they weigh. A one-time image audit ages the same way a one-time SEO audit does — fine for a moment, stale within two quarters.
None of this requires different lighting, different composition, or a different creative brief. It requires deciding, before the shoot, that the photography and the metadata are one deliverable — which is the same discipline shift that shows up across brand work at Acromatico, where the photography and the visibility work are treated as a single loop instead of two departments that hand off files to each other.
Why Doesn’t Anyone Notice When This Is Working?
Nobody on the client side directly sees the win, because a human doesn’t experience “the AI assistant recommended you instead of your competitor” — they just experience getting a call from someone who says “an AI tool suggested you.” The cause is invisible even when the effect is obvious, which is exactly why the metadata work gets deprioritized by teams optimizing for what’s easy to measure in a client presentation — engagement on a post, likes on a photo — over what’s actually shifting: whether the brand exists at all inside the systems now doing a meaningful share of the recommending.
Does This Replace Good Photography?
No. Machine-readable metadata doesn’t replace good photography — a technically perfect but forgettable image still won’t get recommended, because human trust remains the floor. But a stunning photo that no system can parse is now doing only half its job. The brands quietly pulling ahead aren’t shooting differently; they’re finishing the job they used to consider finished at delivery.
The fix isn’t a redesign. It’s discipline applied at the exact moment most teams stop paying attention — the caption, the file name, the alt tag, the five minutes after the shoot that everyone is in a hurry to skip.
Key Takeaways
- Brand photography now has to pass two tests: a human glance test and a machine-read test that extracts subject, setting, and context without a human present.
- Google’s own documentation confirms it reads captions, image titles, alt text, and surrounding page copy to determine what an image depicts.
- The fix has to start on set — a machine-readable caption needs detail that can only be captured during the shoot, not reconstructed afterward from unlabeled files.
- Six habits close the gap: pre-shutter captions, meaningful file names, honest alt text, specific surrounding copy, an explicitly marked hero image, and quarterly audits.
- The payoff is largely invisible to the client, since the effect shows up as an inbound call referencing “an AI tool” rather than a trackable click.
Frequently Asked Questions
What is the “machine read” test for a brand photo? It’s whether a crawler or AI system can determine what an image shows — subject, setting, and context — purely from its file name, alt text, and surrounding page copy, without a human ever viewing the image itself.
Why can’t alt text be written after the shoot is finished? Because accurate alt text requires specific, truthful detail about the setting, people, and moment that only exists if it was captured in shoot notes during the shoot. Writing it afterward from unlabeled files usually produces generic, guessed descriptions instead.
What does Google’s documentation say about how it reads images? Google Search Central states that it extracts information about an image’s subject matter from the surrounding page content, including captions and image titles, in addition to alt text and computer vision analysis of the image itself.
Does better photography alone improve AI recommendation visibility? No. A technically perfect but unlabeled photo of a forgettable brand still won’t be recommended, and a well-labeled but poorly shot photo won’t build human trust. Both the creative quality and the machine-readable metadata have to be right together.
How often should a brand audit its image metadata? Quarterly. Recommendation engines change what signals they weigh over time, so a one-time metadata audit becomes stale within a couple of quarters, similar to how a one-time SEO audit ages.