No Portrait Here Is AI-Generated, But Each One Identifies Itself to AI.
Updated: Oct 6
By Hart Getzen — executive portrait photographer, producer, and publisher. Austin-based, working nationwide. PPA member; work held in the collection of the New York Historical.
Every frame is an original photograph. The portrait is half the job. The other half is giving the systems now deciding how people are found the structured context to read who is in it, who made it, and how it was made.
In the last Insight, I promised three things: what can be claimed versus what’s permanently fixed, structured markup, and a look at what a Person schema block contains. They’re all here, as parts of something larger.
First: no portrait here is AI-generated. Every frame is made with a camera, by me, with a real person in front of the lens. AI-ready means AI-readable, never AI-generated.
I came to this through my own record: twenty years across motion picture and large-format production, then photography and publishing. Several disciplines, one person, and machines reading it as several people. The full story is in the September edition. This one is about what I built because of it.
Why does a real photograph need a record?
Machines don’t infer; they match. A person looking at your portrait brings context: they know your company, they’ve heard your name, they recognize the room. A search engine or an AI system brings only what it can read, and what it can read is text and data. Structured context is what they match against.
Consider a portrait saved as IMG_4471.jpg. It might be the best photograph ever made of you. But the file carries no creator, no date, no description, and no statement of who is in it or where it came from. To a machine, it’s a grid of pixels named by a camera’s counter. It can’t be connected to you, credited to its maker, or told apart from an image something generated last Tuesday. However good the photograph, it is invisible as evidence.
That gap is what the AI-Ready Record fills:
The AI-Ready Record is everything that travels with a portrait besides the pixels: a consistent, machine-readable record of who is in it (identity), and embedded documentation of who made it, when, where, on what terms, and the declaration that it is an original capture, not AI-generated and not licensed for AI training (attribution). In AI terms, it’s structured context: information organized so machines can read it directly instead of guessing.
Put plainly, identity is the record of who is in the photograph. Attribution is the documentation of who made it, and on what terms.
The photograph shows who you are. The AI-Ready Record makes the detailed context readable.
It has two halves, and this piece takes them in order: identity first, then attribution.
Part one: Identity — Identify Yourself
Identity is the half that lives outside the image file: on your website, your profiles, and the databases that already describe you. The workflow I use to define identity is machine-drafted, human-finished: machine-drafted where precision matters, human-finished where judgment does.
What can be claimed, and what’s permanently fixed?
Before you correct your online record, find out which parts of it you’re allowed to touch. Some listings can be claimed and corrected. Others can’t, no matter how politely you ask.
The claimable ones are the obvious places to start: your own site, your LinkedIn profile, professional directories such as a Google Business Profile, and the bio and photo on a credits database’s professional tier. Claim each one, and make it say exactly what the others say.
The fixed ones take more thought. Disambiguation numerals are the clearest case. When two people in a database share a name exactly, the database adds a notation, “(I)” or “(II),” to keep their records apart. The director featured in the September edition carries one. It is effectively permanent. You can’t remove it; you can only give machines enough consistent signals to connect it to the right person.
A few other fixed entries are worth knowing before you lose a month to them:
• Permanent identifiers. The ID number a database assigns to a person doesn’t change, even when everything else on the page does.
• Archived coverage. Old articles, festival programs, and conference listings describe you as you were, and nobody is going back to update them.
• Aggregator profiles. Pages assembled from scraped data often have no correction path at all.
• Former employers’ pages. They can still rank well and still carry your old title, and you don’t control them.
The principle is simple: fix what you can, and outweigh what you can’t. Claim and correct every listing that allows it. Then make your own authored account the strongest, most consistent, most clearly attributed source available, so the fixed entries point toward it instead of competing with it. A full platform-by-platform guide is a future edition.
Canonical name, titles, and bio
Choose one styling of your name, with or without the middle initial, with or without the suffix, and one set of titles. Then use them identically everywhere. To a person, “Hart Getzen” and “H. Getzen” are obviously the same. To a machine, they are two candidates. Write one canonical bio and let every profile, caption, and introduction draw from it. Consistency is dull work. It is also the most valuable kind.
What is structured markup?
Structured markup is a short block of code, written in a shared vocabulary called schema.org, that tells search engines and AI systems exactly what a page is about. Without it, a machine reads your About page and guesses. With it, the page states its subject directly: this is a person, this is his name, these are his titles, these profiles are his.
Schema.org was created by the major search engines, and the AI systems that draw on search inherit what it says. That makes it one of the most direct ways to tell machines who a page is about.
Where it lives: in the code of your About page, invisible to visitors. Most site builders, including Wix and Squarespace, have a place to paste it. For clients, it goes on their own site, because the authored account should live on a page they control.
The Person schema block, line by line
Here is mine, shown as a working example rather than a template to copy and paste. Structured markup is exact: a misplaced comma or bracket can break the whole block, and writing one correctly for a scattered, multi-platform career is exactly the kind of detailed work that's harder than it looks.
<script type="application/ld+json">
{
"@context": "https://schema.org",
"@type": "Person",
"name": "Hart Getzen",
"jobTitle": "Executive Portrait Photographer, Producer, and Publisher",
"image": "https//static.wixstatic.com/media/f6417a b8bc32ddb5674497baa750870fc7410b~mv2.jpg",
"sameAs": [
],
"worksFor": {
"@type": "Organization",
"name": "Hart Getzen Photography",
},
"affiliation": {
"@type": "Organization",
"name": "Professional Photographers of America"
},
"alumniOf": [
{
"@type": "CollegeOrUniversity",
"name": "Vassar College"
},
{
"@type": "CollegeOrUniversity",
"name": "London Film School"
}
],
"knowsAbout": [
"Executive portrait photography",
"Portrait series",
"Image libraries",
"Structured data for images",
"Large-format film production",
"IP authorship",
"Immersive motion simulation",
"AI applications"
]
}
</script>
What each line does:
• @context and @type tell the machine which vocabulary you’re using, and that the block describes a person.
• name is your canonical name, styled exactly as it appears everywhere else.
• jobTitle is your canonical set of titles, in the same order you use in every bio.
• url is the page you control that carries your authored account, usually your About page.
• image is your canonical portrait: the same file you use on LinkedIn and in directories. This is where the face meets the record.
• sameAs is the most important line. Each address is a profile that is genuinely yours. This is how a numbered database entry gets connected to you instead of floating free. List only profiles you’ve claimed.
• worksFor ties you to your company, so the person and the business read as connected.
• affiliation names the professional organizations you belong to.
• alumniOf names the schools you attended; list more than one as an array, as it is here.
• knowsAbout lists the subjects you want to be associated with, phrased the way people actually search for them.
Now the honest part. The block is the easy part. Writing it correctly for a scattered, multi-platform career is harder: deciding which profiles belong in sameAs, which title leads, how to state a career that crosses disciplines so it reads as range rather than as several people. Keeping it consistent with every bio, caption, and listing, year after year, is the hardest part of all.
Part two: Attribution — who made the image, and on what terms
The image inside itself is mise en abyme. Van Eyck painted a mirror reflecting the room, and likely himself, then signed the wall above it: “Jan van Eyck was here.” Velázquez painted himself at the easel in Las Meninas. Both put the maker inside the work. This frame does it live, in camera, with no compositing, and the AI-Ready Record does it in the file: who made it, and how, written where machines can read it.
Attribution is the half that lives inside the image file and right alongside it. It’s what keeps a photograph identifiable as itself wherever it travels.
Image labeling
Every photograph in a Series gets a descriptive filename, alt text, and a caption. The filename replaces IMG_4471.jpg with something a machine can read, such as hart-getzen-executive-portrait-austin.jpg. Alt text describes the image accurately, for screen readers and for search. The caption names the person, role, and organization the same way every time.
Accuracy is the whole point. Alt text that describes a room the subject wasn’t in, or a title they don’t hold, teaches machines the wrong thing, and does it at scale.
Embedded attribution
Inside the file itself, I write the creator, capture date, location, and rights; the declaration that the image is an original capture, not AI-generated; and terms stating it is not licensed for AI training. These go into both the print file and the web file of every image. They follow the IPTC photo metadata standard, the one news agencies and publishers already use, which includes fields for an image’s source and for data-mining terms.
The honest caveat: some platforms strip embedded metadata on upload. That’s why attribution never works alone. The filename, caption, alt text, and your structured markup all carry the same facts, so when one is lost, the others still speak.
Deployment map
The last piece is a map of which image goes where: the square crop for profiles, the vertical for press, the wide frame for a site banner. A Series is composed with each job in mind; the map makes sure each job gets the right frame. The Allie Gallion Series in the July edition shows each crop and the job it does.
What’s included with my work, and what’s extra?
There are three tiers, and they match the Commissions page:
• Every commission: each file is delivered with authorship, copyright, and licensing information embedded.
• The AI-Ready Record (add-on): the original-capture declaration and no-AI-training terms in both print and web files, plus captions and alt text, a controlled keyword vocabulary, consistent file naming, multi-format crops, and an indexed folder structure with a delivery sheet. Included with every image library, or as requested.
• Record work (quoted separately): the profile audit, canonical name and bio, and Person schema from Part one. Every career is scattered differently, so this is scoped to yours.
Depending on what’s commissioned, a single AI-Ready Record file carries somewhere between eleven and sixty-one data fields, alongside three to eight additional external files: captions, alt text, a delivery sheet, a keyword list, the schema. That’s a lot of structured context for one photograph to carry.
Current pricing is on the Commissions page.
What it means for you: when someone asks an AI who you are, it has an authoritative, consistent source to draw on: structured context, clearly attributed, with your name, your work, and your actual face. I can’t promise what an AI will say, and no one honestly can. The promise is the quality of what it draws from.
Closing
Platforms and AI systems will change; schema.org, embedded metadata standards, and consistent naming won’t. Build on those, and the work doesn’t need rebuilding every time a system updates.
The four frames in this piece were made in the room where this work happens, and each one carries its own AI-Ready Record. That room is still a darkroom. It never went away; what it produces changed. It once developed film into a print. Now it develops the image and the record that travels with it.
Photographers still work in a darkroom. We just develop more than the image now.
Frequently asked questions
Is the AI-Ready Record included with every commission?
Partly. Every file carries authorship, copyright, and licensing information. The full AI-Ready Record is an add-on, included with every image library, or as requested.
Do you ever use AI to generate or alter portraits?
Never to generate. Sometimes for masking and removing artifacts. Those tools adjust the photograph; they never add anything that wasn’t there.
What does “not licensed for AI training” actually do?
It declares your terms in a machine-readable form that travels with the file. It doesn’t technically block anything; it records the terms where anyone handling the file can read them. Readable is not the same as trainable: the AI-Ready Record invites AI systems to read the photograph and declines to let them learn from it.
Does embedded metadata matter for print?
Yes, though it protects the file, not the page. The printed image carries only its visible credit. But the print file is the full-resolution master, it often ends up online or in a publisher’s licensed archive, and the embedded terms go with it.
Is this the same as Content Credentials (C2PA)?
No. Content Credentials use cryptographic signatures to make tampering detectable. The AI-Ready Record uses standard embedded metadata and structured data. It doesn’t cryptographically certify. It documents and declares, in a form AI systems can read.
Can I do this myself?
Partly. Claiming your own profiles is something you can do yourself, and the audit principles above apply to anyone. The schema block is shown to illustrate what a Person schema contains, not as a template to copy: a single syntax error can break the markup. The schema block above is a real start, and so is claiming every profile that’s yours. Consistency across every platform, over time, is the hard part.
How long does it take to see changes in AI answers?
Weeks to months, depending on how often a system refreshes its sources. There is no fixed timeline and no guarantee.
Add the AI-Ready Record to any Portrait Series; it’s included with every image library. Start with a brief inquiry.
Hart Getzen builds executive portrait systems, institutional image libraries, AI-ready image libraries, and printed brand editions for founders, leaders, artists, and institutions. To discuss a commission, get in touch.
Comments