This isn’t a new idea. It’s a newly official one.

Sullivan’s framing at the event was commodity versus non-commodity content. Commodity content is factual, widely available, offers no unique perspective, and requires no expertise to produce. Non-commodity content carries your original voice, your particular take, your firsthand experience. The test he offered: only you could have written it.

Kipp Bodnar at HubSpot picked up the framework almost immediately in a Marketing Against the Grain episode, extending it into a six-checkpoint version with a grader tool. That’s where I encountered it, and it’s what I’ll use for the audit below.

If that framing sounds familiar, it should. This newsletter has been calling the same thing the generic photographer problem since March. The argument I’ve been making is that AI tools without voice infrastructure produce output indistinguishable from what any model would generate for any photographer anywhere. Sullivan is making the same argument from the search side. The language is different. The problem is identical.

I find it useful when Google confirms something I’ve been worried about for a while. Less useful when I realize I need to go audit my own site.

The framework

He built a grader around the six checkpoints that scored the top-ranking Google result for countertop material comparisons at 9 out of 100 during a live demo in the episode. A page sitting at number one on Google, scored against this framework, nearly bottom of the scale.

That’s the gap between traditional SEO ranking and AI-era discoverability in one number.

The six checkpoints:

Proprietary evidence. Data or documentation only you could have. Not “professional photographers recommend” but a specific customer’s shoes at 402 miles with 4mm of foam compression on the right heel, photographed in your shop.

First-hand experience. I did, I saw, I built. The word “I” followed by a verb describing something that actually happened.

Specificity. Names, numbers, dates, places. The test Bodnar offers: for every adjective, ask whether you could replace it with a number, name, or date. Fast becomes 11 seconds. Popular becomes 4,200 orders last month. Recent becomes March 14th.

Point of view. A line you’re willing to lose business over. The example from the episode is a countertop installer who won’t put marble in a house with kids, and will walk away from the job. The test: can you start a sentence with “I believe” or “I won’t”? If not, you don’t have a point of view, you have a position statement.

LLM test. Could ChatGPT write this without your input? If yes, it’s commodity. If the piece depends on your being there, on something you documented, on something only you could have observed or decided, it passes.

Information gain. Does the reader learn something not in the top three Google results? Before publishing, search your own headline and read what comes up. If you can’t circle three facts in your piece that aren’t already out there, your information gain is zero.

What happens when you run it against a real site

I ran it against michaelklothphotography.com using Bodnar’s grader, because if I’m going to argue that photographers need to think about this, I should be willing to show what it looks like on my own pages.

The first result stopped me. I submitted the blog URL and got a score of 21 out of 100. Mostly commodity. The grader’s verdict: “Teaser-only excerpt with a compelling hook but zero payoff — no data, no visuals, no firsthand detail delivered.”

Which is accurate. The blog index page surfaces teasers, not full posts. The grader evaluated what it could access, which was a premise and a “Read More” button. That’s not a failing grade on the blog. It’s a correct evaluation of what a blog index page actually is — and a useful illustration of how algorithms encounter your content when they can’t get past the surface.

I ran two actual posts next.

The parking lot/OR post — a session at Friends of PACC that started with a dog escaping in the parking lot and ended in a veterinary surgical suite — scored 69. The grader’s summary: “Strong firsthand storytelling with specific clinical detail, but undersells its own proprietary evidence.” Its evidence for the non-commodity rating: the runaway dog, the disconnected power adapter, the Valley fever radiograph series. Details, it noted, that no LLM or generic photography blog could fabricate. The score is held back by missing images in the text and a few passages that slip into explanatory mode.

Even Odds, the shelter photography retrospective published in March, scored 74. The grader’s headline: “Twenty years of dated metadata, grant-impact footage, and euthanasia statistics the author personally witnessed make this genuinely non-commodity.” It correctly identified the grant-impact footage and the direct causal link between the photography and a funded behavior program as the load-bearing elements. Points lost for passages that drift into reflective generality rather than scene-level specificity.

Three scores: 21, 69, 74. The 21 is a grader limitation — it evaluated a teaser, not a post. The 69 and 74 have a limitation of their own worth understanding: the grader couldn’t read the images. Squarespace serves images through its CDN with captions as separate elements, and the grader treated them as absent from the content. Both posts have images with captions. The grader scored neither of them. The 69 and 74 were earned on text alone. The actual posts are stronger than the scores reflect.

That’s the finding worth sitting with: the content doing non-commodity work scored in the non-commodity range without the grader being able to see roughly half its evidence.

The first thing the audit clarified is that the framework doesn’t apply uniformly across a site. Service pages and blog posts are solving different problems, and conflating them produces a misleading result. That doesn’t mean there isn’t something to learn though.

A service page is navigational and transactional. Someone searching “Tucson headshot photographer” is trying to answer three questions: do you do the thing I need, what does it cost, and how do I book you. That’s a commodity search intent, and the right answer to it is a page that’s clear, credentialed, and easy to navigate. Weaving in non-commodity storytelling where someone is trying to check a box is friction, not value. The service pages on the site are doing exactly what service pages should do.

The framework applies where someone is researching rather than ready to book. Blog posts. Informational queries. “How much do professional photos of a dog cost.” “Do headshots matter for LinkedIn.” That’s where the non-commodity requirement is real and consequential, because that’s where Google and LLMs are deciding whether your site demonstrates actual expertise or just answers the question adequately.

So the audit finding isn’t “the service pages fail the framework.” It’s that the framework applies differently depending on what a page is trying to do, and the blog is where it matters most.

What the blog is doing right.

The “Even Odds” post has the specifics the framework calls for: the 2015 training pilot, the 14,000 fewer euthanasias figure, the rainstorm session where what the shelter actually needed wasn’t photography. First-hand experience documented at the level Sullivan is describing.

The dog photography pricing post passes the information gain test because it’s honest market data from someone who’s worked with thousands of dogs, positioned to help readers decide whether to book me or someone else. That’s a perspective no aggregator site can replicate.

The Friends of PACC post — the one that scored 69 — wasn’t written with this framework in mind. It was written because the session happened and was worth documenting. The score reflects that. When the underlying experience is real and documented at the right level of specificity, the framework takes care of itself.

Where the gap actually is.

The commodity content on the site didn’t happen because the writing got lazy. It happened because some posts were written to answer search queries rather than to document experience. That’s a structural decision, not a writing failure. “Do headshots matter on LinkedIn” has search volume. “What I learned photographing anxious lawyers” does not. Most photographers make the same call.

The dentist who writes “How to Choose a Dentist in Atlanta” gets buried in AI Overviews. The dentist who writes “Why I Refused to Place an Implant on This 67-Year-Old Smoker” gets cited. The search-volume version of the headshot post answers the question adequately. The experience-documented version is the one that only I could have written.

That’s the tension the framework surfaces. Not a writing problem. A strategy problem. And the corrective isn’t to rewrite every informational post — it’s to understand which content is earning discovery and which is just answering the question everyone else is also answering.

The distinction the framework keeps returning to

Bodnar’s test for point of view is worth sitting with: can you start a sentence with “I believe” or “I won’t”?

The countertop installer says: I won’t put marble in a house with kids. I’ve done too many etched surfaces for families who wanted the look. Quartzite or nothing.

What’s the photography equivalent?

I have positions. I don’t deliver same-day digital galleries because I’ve seen what rushed editing does to a portrait, and I’m not willing to put my name on work that hasn’t had time to breathe. I won’t retouch out features that define someone’s face. I’ve walked away from conversations where what the client wanted and what the portraits needed were too far apart to close.

Those positions exist. Most of them aren’t on the site yet.

That’s the gap the framework identifies, and it’s useful because it’s specific. It’s not “be more authentic” or “show your personality.” It’s: name the line, state it plainly, accept that some readers will disagree with it.

Why this matters more now than it did six months ago

Sullivan’s observation at the Toronto event is worth taking seriously: AI features speed up a reality publishers have already been facing for years. Commodity content was always replaceable. What’s changed is the speed of replacement and the scale of the competition.

A blog post that could have been written by any photographer, about any market, for any reader, competes with every other post written the same way. In an LLM’s training data, those posts blur together. In an AI Overview, the one that gets cited is the one that contains something the others don’t.

The photographers building non-commodity content now aren’t doing something exotic. They’re documenting what they actually do, with the specificity of people who were actually there, from a point of view they’re actually willing to defend. The strategy question isn’t whether to do this. It’s recognizing which content decisions have been optimizing for search volume when they should have been optimizing for irreplaceability.

That’s the work. The framework just makes it easier to see where it’s missing.

The connection between this framework and the persistent identity methodology I write about in the paid tier is direct: what Sullivan is describing from the search side is the same problem the generic photographer concept describes from the AI output side. The mechanism underneath both is the same. The conversation will continue to evolve.

Kristen is the founder of Outcomes Consulting and a former executive director at Pima Animal Care Center. She runs a five-week course called AI for Animal Welfare Professionals - instructor-led, peer-to-peer learning, continuing education credits through CAWA and NACA. The people in her class are shelter directors, communications managers, foster coordinators. People running organizations on tight margins who need AI to actually work, not just theoretically work.

I replied the same day. I’ve been volunteering at shelters since 2005. This was not a cold audience.

A few emails back and forth and I had a 15-minute guest slot in her Tuesday class the following week. The topic was AI and photos - specifically, how I’m integrating AI into shelter photography work and what’s actually worth the time of people who need results, not experiments.

Fifteen minutes is enough to establish whether the overlap is real. It was. Kristen reached out afterward to set up a one-on-one Zoom.

That conversation was something I hadn’t quite expected. I went in thinking it would be a speaker briefing. What I got was closer to a peer exchange, and it went in both directions. She’d been working with Gemini Nano specifically because it edits photos without changing the appearance of the animals - which turns out to be exactly the right instinct, and something I’d been thinking about from the photography side of the table. She’s also vibe coding using Claude, which put us in a very similar place methodologically.

The Venn diagram between teaching photographers to use AI without losing their voice and teaching shelter organizations to use AI without losing theirs turned out to be almost a circle. Voice guides, prompt specificity, knowing when to stop iterating, the difference between output that sounds like someone and output that sounds like no one in particular - these are not photography problems. They’re organizational communication problems that happen to show up in the same form whether you’re writing a headshot inquiry response or an adoption bio.

I sent her a follow-up email with a few things I’d promised during the call. She invited me to speak at her First Fridays collaboration call.

First Fridays is a monthly open session for animal welfare professionals navigating AI adoption - broader than the class, open registration, people at different stages of figuring this out.

The topic Kristen wanted covered was AI and shelter photography: what tools actually help, where they create problems, and why the line between enhancement and misrepresentation matters for adoption outcomes. I had a clear point of view on the argument. What I didn’t have was hands-on experience with the specific tools her audience would be using. I work in Adobe Lightroom Classic and Photoshop for my professional editing - not what a shelter volunteer with a phone reaches for between kennel rounds. So the night before the call I ran a series of Gemini editing experiments on real shelter photos, specifically to understand what the tools could and couldn’t do for that workflow. Some worked cleanly. Some hit hard ceilings. One produced something alarming from a two-word prompt that felt completely innocent.

More on that in the paid section below.

For the deck I used NotebookLM. I fed it my outline and it generated a 15-slide presentation I exported as a PowerPoint. The design was genuinely impressive - the kind of visual polish that would have taken hours to build from scratch. The tradeoff: individual elements in the exported file weren’t editable, and there were a few errors I caught during review that needed correcting before the call. The PDF of the deck is attached to this issue. It carries the visual argument better than a description would.

The script and speaker reference card came out of a working session with Claude. By the time I sat down for the call I had a full script, a one-page bullet point reference card mapped to each slide, and five documented experiments ready if the audience wanted to go deeper.

The short version of the argument: two reactions to AI in photography and communications are common right now, and neither is useful. Uncritical adoption says AI can do everything, use it for all your imagery. Reflexive opposition says AI is fake and undermines authenticity. Both skip the actual question - does this help the animal get adopted? Operational pragmatism is the frame that actually produces useful decisions.

What happens next with Kristen I don’t know yet. My assumption is that she’s building a rotating roster of guest speakers so her monthly audience gets a range of perspectives over time. If there’s a May conversation I’d be glad to be part of it. Twenty years of shelter work, the animal welfare community, and a methodology built around authentic communication don’t often find each other in the same room. When they do it’s worth staying.

If you work in animal welfare or know someone who does, Kristen’s monthly calls are open registration. I’ve linked her organization above.

Pragmatic Shelter Ai 4 3 26

3.51MB ∙ PDF file

Download

Download

[PDF of the presentation deck attached above - free to all subscribers]

Read more

The question I kept coming back to wasn’t whether AI tools are useful for photographers. That argument is settled enough to be boring. The question worth asking is more specific: what does the AI actually know about you right now, and how does it decide who gets cited as an expert when someone asks?

I asked four AI systems - Claude, ChatGPT, Gemini, and Perplexity - the same question: when someone asks you about professional photography, what sources do you treat as authoritative, and why?

The methodology

I used the same prompt across all four. I asked about source categories, specific organizations, how individual photographers’ published content gets weighted against major publications, what signals make a photographer’s website more likely to be treated as authoritative, and whether they had any visibility into how their sourcing behavior might change as AI-mediated search becomes more common.

I also asked each of them to be honest about what they don’t know about their own source weighting.

That last part turned out to matter.

I’m describing this methodology because it’s replicable. You can run the same prompt across the same systems and compare the answers yourself. The spread of responses - from Claude’s frank acknowledgment of its own limits to Perplexity’s thin genericism - tells its own story about what these systems actually know about themselves. The comparison is the research.

Read more

The marketing conversation around this has mostly focused on lead generation tactics. I want to take it somewhere more useful for photographers, because I think the implications go deeper than the tactics discussion suggests.

It starts before the conversation does

The “invisible prompt” framing focuses on context built within a single chat session - the back-and-forth before someone finally asks for a recommendation. That’s real, but it’s only part of the picture.

People who use these tools regularly carry something with them that persists across sessions. ChatGPT, for instance, has built something close to a persistent user profile - enabled by default since mid-2025, pulling from conversation history to build an understanding of your interests, tastes, skill level, and personal context. Claude’s memory architecture works differently, opt-in by design and scoped to specific projects rather than building a unified personal portrait. The tools aren’t identical in how they do this. But the broader reality is that someone who has spent months asking about mid-century modern furniture, donating to animal shelters, and the paintings of David Hockney arrives at the question “who does pet photography in Tucson” already known, in ways that no intake form or pre-session consultation could replicate.

The AI isn’t just filtering recommendations against what someone said in the last half hour. It’s personalizing against an accumulated portrait of a person.

What this actually means for the match

Think about what that portrait contains. Someone with contemporary design sensibilities and a strong preference for punchy color in their art collection is probably not a great fit for a photographer whose work runs toward traditional Rembrandt lighting and muted tones. The AI, if it has the signal to work with, knows that. It’s going to make a different recommendation for that person than it would for someone whose searches skew toward classical portraiture and traditional interiors.

This is new. We’ve always known that not every client is a good fit for every photographer. Experience teaches you to read the inquiry, ask the right questions, feel out whether your aesthetic sensibility and theirs are going to produce something both of you are proud of. Sometimes it’s obvious. Often it isn’t until you’re in the session, or worse, in the reveal.

What these tools are building toward is a matching layer that happens before the inquiry. The client who finds you through an AI recommendation may arrive already pre-filtered in ways you can’t see and didn’t engineer. They found you because what you put into the world aligned with who they already are.

The honest implication

Here’s where the marketing conversation around AI visibility goes sideways: it starts looking for the lever. The optimization play. The way to get recommended the same way photographers learned to game local search rankings.

Pay-to-play AI advertising has arrived, but not in any form that’s relevant to this conversation. ChatGPT launched sponsored placements in early 2026 with a $200,000 minimum commitment, access restricted to a handful of major agency holding companies running campaigns for Apple, McDonald’s, and American Express. Microsoft Copilot and Google’s AI search features have their own ad products at similar enterprise scale. Anthropic has explicitly committed to keeping Claude ad-free, and spent money on Super Bowl advertising making that point directly. Perplexity tested ads and abandoned them entirely, concluding they undermine trust in the responses.

So yes, pay-to-play exists. It exists at a scale that has nothing to do with working photographers, on platforms that a meaningful portion of your potential clients aren’t using for this kind of search. The playing field for the rest of us is still organic. And on the platforms most committed to keeping it that way, organic authority is the only game there is.

What the research actually shows

Here’s the part that the content optimization conversation tends to underemphasize: what others write about you matters more than what you write about yourself.

Research analyzing AI brand citations across major platforms found that roughly 85% come from third-party sources, not from the brand’s own content. Press mentions, directory listings, client reviews, venue blog posts that name you specifically, industry articles that reference your work - these are the dominant signal AI systems use when deciding who to recommend. Your About page matters. Your blog matters. But the mention in a local interior design blog, or the wedding venue that lists preferred photographers, or the review that describes your specific approach to anxious animals - those matter more.

The implication for photographers is practical: earning mentions in the right places is at least as important as publishing the right content on your own site. Probably more important. This doesn’t change what good content strategy looks like, but it adds a dimension that most photographers aren’t thinking about yet.

This is a call to put real work into the world

Which brings me to what I think is the most useful reframe, and the one that matters most for working photographers.

This isn’t primarily a marketing strategy. It’s a case for putting work and thinking into the world that honestly represents who you are as a photographer.

The photographer who has spent twenty years developing a specific approach to anxious animals, and who has written honestly about what that looks like in practice, and whose work reflects a genuine and identifiable aesthetic point of view, is building signal that AI tools can actually use. Not because they engineered it for discoverability. Because it’s true, and it’s specific, and it exists in enough places that the tools can find it and trust it.

The photographer who wants to be everything to every client, whose content is designed to appeal to the broadest possible audience, produces signal that either gets lost in the matching process or - worse - matches them to clients who aren’t actually a good fit. Which helps no one.

Experience teaches most photographers eventually that the clients who are right for your work are the ones worth finding. The invisible prompt, at its best, is a mechanism that accelerates that matching - and the foundation you build today, accurate representation of real work and genuine perspective, will matter more as these tools get better, not less.

When the pay-to-play layer eventually reaches the scale where working photographers need to think about it, the ones who built honest signal into their public presence over the years before that will have something to protect. The ones who waited for an optimization strategy will be starting from scratch.

Put real work out there. Write honestly about what you do and why. Earn the mentions that come from doing the work in public. Let the AI figure out who it’s for.

That part, it turns out, it’s already pretty good at.

Last issue ended with a transparency note: the methodology I described is the one I used to write the issue describing it. A voice guide plus project knowledge, Claude as the primary tool, my thinking and examples throughout.

This issue is about building that guide. Not the concept of it, we covered that. The actual components, where the material comes from, how to feed it to the tool, and what to do when it still doesn’t sound right. By the end you should have enough to build a working first version.

Working is the key word. It will not be perfect. We will get to that.

Read more

Part two is where the kinks get worked out. Or at least, where you find out how many there are.

The PhotoShelter API Problem

The original plan included pulling images directly from my PhotoShelter galleries into the visualizer automatically. When the API documentation finally arrived, it became clear that wasn’t going to work the way I’d imagined. The API was designed to import images into WordPress for blog posts, one at a time. Not a feed. Not a dynamic gallery pull.

The workaround is a local image library that mirrors the store’s fine art print listings, either pulled from PhotoShelter manually or uploaded directly. Less automated than planned. That’s a limitation of PhotoShelter’s API design, not of the app itself.

For client-specific previews, the solution is more manual still. Images will be uploaded and tagged with client-specific identifiers, and clients will get a version of the page that only surfaces their tagged images. Clunkier than envisioned. Workable enough to run a trial.

The Iteration Loop

Once the image library question was sorted, the focus shifted to getting the “Send to Michael” functionality working - the piece where a client submits their visualization to me with the print dimensions baked in.

The basic flow: client places a print on their wall photo, the app saves the image to their downloads folder, a contact form appears, they fill it out and submit, I get an email with their name, contact information, and the print specifications. Straightforward in concept.

Getting the Contact Form 7 ID wired correctly was the easy part. One URL in the browser address bar, one number to plug in.

What followed was a long afternoon of screenshot-and-iterate. Make a change in WordPress or the file manager, share a screenshot, get a diagnosis, make another change, share another screenshot. This is the actual texture of vibe coding, and it’s worth describing honestly: it’s not “describe what you want and it appears.” It requires you to stay oriented through a lot of small adjustments, and it requires enough working knowledge to recognize when the conversation has gone off track.

Which it did. For most of the afternoon.

The CORS Problem That Wasn’t

The tool spent several hours diagnosing a CORS error. CORS - Cross-Origin Resource Sharing - is a real thing, a browser security mechanism that can block canvas exports when images are loaded from other domains. The symptoms looked plausible. The fixes were technically reasonable. Headers were added to WordPress. Image loading code was updated. Debug logging was added to catch errors that weren’t appearing.

None of it was the problem.

The download was working the entire time. The image was landing in the downloads folder on every test. The CORS diagnosis was a confident wrong answer, pattern-matched against plausible explanations rather than the actual evidence. By the time that became clear, the afternoon was gone.

The real issue is simpler and still unresolved: after the download fires, the contact form is supposed to appear. It doesn’t. The code that should trigger it is there. It’s just not working. That’s where things stand.

What This Is Actually Teaching Me

The learning was the stated goal going in, and the learning is happening, even when the tool is wrong.

A few things I now understand better than I did a week ago: WordPress serves as both the CMS and the API layer here, and those two roles create friction when you’re building something outside the normal template flow. The visualizer is a standalone HTML file living in a theme folder, which means it doesn’t automatically inherit WordPress’s authentication tokens or other values the rest of the site takes for granted. Each of those has to be deliberately wired.

The more useful lesson is about the limits of the collaboration itself. The tool can’t actually see your screen. Every diagnosis is an inference from whatever you can describe or screenshot. When the symptoms are ambiguous - and a download that works while a form doesn’t appear produces ambiguous symptoms - the tool will generate confident explanations that are plausible but wrong. Recognizing that pattern, and being willing to stop and reframe rather than keep patching, is a skill the tool can’t supply.

Nine months of working closely with these tools has given me enough of a working vocabulary to follow the logic and recognize when the conversation has gone circular. That’s not the same as knowing how to fix it. It’s enough to keep moving, and to know when to stop for the day.

Where Things Stand

The frame and mat color selectors work. The wall size calculation is correct enough. The image placement and resize work. The download works. What doesn’t work yet is the contact form appearing after the download - the piece that makes this a two-way tool rather than a visualization toy.

I’ll finish it. The workable tool as the cherry on top. That’s still the goal.

That’s not a disclaimer. It’s the point.

I started doing something unusual for a photographer with an MFA and no engineering background. I’ve been building software.

Not tinkering with plugins. Not configuring a booking system. Actually building a web application from scratch, using AI tools to do the technical work while I do something else: knowing what I want, evaluating what I get, and refusing to accept less.

The app is a room visualizer for my photography clients. The concept is straightforward. A client uploads a photo of the wall where they want to hang artwork - a fine art print, a pet portrait, whatever they’ve been considering. The app either accepts a manual measurement or walks them through marking reference points in the photo (a standard door, a wall outlet) to calculate approximate dimensions. They browse my Photoshelter gallery, choose a print, select mat and frame options, and the app renders an augmented reality mockup of that image on their wall. The mockup is downloadable. It also gets emailed to me automatically.

If you’ve ever tried to sell fine art prints and watched a client hesitate because they couldn’t visualize the scale, you understand why I wanted to build this. The design consultation conversation changes completely when someone already has a mockup in hand.

Where the idea to build this way came from

I didn’t invent the workflow I used today. I borrowed it from Dan Roth, editor at LinkedIn, who described it on the How I AI podcast.

Roth builds iOS apps on weekends using Claude Code. He’s not an engineer. He’s spent 15 years inside a tech company, watched how engineering teams actually work, and translated that into a two-agent setup. One agent builds. One reviews. He breaks ties.

His builder is named Bob (yes, Bob the Builder). His reviewer is Ray - a persona prompted to behave like a senior software engineer obsessed with security, architecture, and not rubber-stamping everything the builder does. The human-in-the-loop isn’t optional. It’s the design.

Roth spent some time figuring out what to call his own role in this setup. He ruled out PM (too rigid, requires keeping the whole app in your head). He ruled out architect (an architect actually knows real details). He landed here:

“All I am is a really picky customer. I’m like walking through this house and I’m telling the architect, no, I want this room blue. I know you don’t think it’s a good idea. I’m telling you, this is what I want.”

That framing clicked for me immediately. I knew what I wanted to build. I had clear opinions about how it should work and what the client experience needed to feel like. What I lacked was the technical vocabulary to build it myself. The picky customer role is a real role. It turns out to be enough. Or at least I think it’s enough (part 2 to come).

I adapted the setup for my situation. My builder is Dag. My reviewer is Norbert. I am still the picky client.

What “agentic” actually means when you’re running a photography business

There’s a version of the agentic agent pitch that implies you set something in motion and come back to a finished product. That’s not what happened today.

The process took most of a day. Some of that was because I was doing other things in parallel - sessions to edit, emails to answer, the actual photography business continuing to exist. That’s the appeal of the setup: Dag works while you’re occupied elsewhere.

But this particular workflow has human handoff points built in by design. Dag builds to a milestone and stops. Norbert reviews and either approves or requires changes. The notes between them pass through me. When other work pulls your attention, it’s easy to miss the moment when the build is waiting on a review, or when Norbert has signed off and Dag is ready to proceed.

This isn’t a flaw. Norbert exists precisely because unsupervised building is how you get code that looks finished but has real problems underneath. The handoff friction is intentional. Reading what Norbert flags, understanding why Dag made a particular technical decision, following the back-and-forth - that’s where the learning happens. I’m not trying to learn to code. But I am trying to understand enough of what’s being built to evaluate it. That’s a different skill, and it compounds.

The honest description of agentic work for a solo business owner isn’t “it runs while you sleep.” It’s “it works in chunks between your attention, and your attention at the handoff points still matters.”

The platform constraint that turned into better architecture

My original plan was to embed the room visualizer directly into my Squarespace site. That didn’t survive contact with reality.

Photoshelter - where I host both my fine art galleries and client proofing - only supports API keys for WordPress integrations. Squarespace isn’t on the list. The app needed Photoshelter access to pull gallery content, so Squarespace was out.

Unfortunately, we only provide API Key for WordPress for the Photographer's account

I have a hosting account with InMotion Hosting that I use for my custom domain. The fix was straightforward: spin up a dedicated WordPress install, point a subdomain at it via an A record on the Squarespace-hosted domain, and put the app there. The subdomain is visualizer.michaelklothphotography.com. It’s branded. It’s separate. It resolves.

What I didn’t fully appreciate until I was already committed to it: this is probably the right architecture regardless of the Photoshelter constraint. Squarespace is already slower than I’d like. A room visualizer with image upload, canvas rendering, and API calls would compound that load on every page that shares the domain. Isolating the app on a separate install protects the main site’s page speed, which matters for SEO and for the client experience on michaelklothphotography.com.

The workaround turned out to be the right answer. That happens more often than the planning-first approach would suggest.

One thing worth naming explicitly

The workflow I used today didn’t come from a photography forum. It didn’t come from a PPA webinar or a photography educator’s course. It came from a podcast about building with AI tools, hosted by and for people who are figuring this out in real time across industries.

I’m not saying that to be critical of photography education. I’m saying it because the photographers who are moving fastest on this aren’t necessarily the ones with the deepest photography networks. They’re the ones paying attention to how people in other fields are solving adjacent problems.

The How I AI podcast is one place to look. There are others. The instinct to look outside the industry for methodology is itself worth developing.

Part 2 is forthcoming, and I genuinely don’t know what it will say.

Today I will configure the WordPress install properly and drop the code in. Both Dag and Norbert signed off on the final build. That’s a good sign. It is not a guarantee.

If it works, Part 2 is about what a room visualizer actually does for the print sales conversation, and whether a day of parallel attention was a reasonable investment.

If it doesn’t, Part 2 is about that. Which, honestly, might be more useful.

If you want to hear the Dan Roth episode I referenced, I linked to “How I AI - Dan Roth” above. His newsletter is called Forward Deployed Editor.

Hey!! 😊📸 So SO excited 4 ur session tmrw!! It is gonna b so MAGICAL and honestly so transformative 🌟✨ like I literally cannot believe we get 2 do this!!! The light at ur location is gonna b PERFECT and we r gonna create such amazing memories together!!! 🚀🙌📷📷📷 lol see u soon!!

“Thank you for letting me know about Buddy’s mobility problems. I will work within his comfort zone but it helps me to prepare for the session.”

Cant wait!!!!! 🐾💫🎉

...anyway. AI is amazing for emails!!

Okay. I’ll stop.

That one sentence about Buddy is the only part of that confirmation I actually wrote. Everything else is what happens when you hand a prompt to an AI tool with no context and accept whatever comes back. If you’ve been reading this newsletter from the beginning, you probably felt something shift when you hit that sentence. That shift is the whole point of this issue.

Using AI for correspondence and other business copy is genuinely a great entry point, for photographers and pretty much anyone else running a small business. The time savings on routine communication are real and the barrier to starting is low. But as with most things worth doing, the more work you put into setting it up properly, the better your results will be. The difference between that email up there and something you would actually send a client is not the AI tool. It’s the infrastructure you build around it.

Before we get into what that infrastructure looks like, it’s worth spending a moment on the tools themselves, because they are not interchangeable and the choice matters.

A Quick Orientation to the Main Players

The landscape as of early 2026 has a few dominant tools, each with a distinct character.

ChatGPT (OpenAI) is the most widely recognized name and has a massive user base. The current model is GPT-5.4, and it is capable across a wide range of writing tasks. Its broad training makes it strong for general business writing and it handles long-form content well. A free tier exists with usage limits; paid plans (Plus, Pro) unlock the more capable models and higher usage limits.

Claude (Anthropic) tends to produce writing that reads more naturally and has stronger instincts around tone and nuance. For work where voice consistency matters (which, as you’ll see, is the entire argument of this issue) it is worth serious consideration. It also handles long documents and complex context well, which becomes important once you are working with detailed style guides. A free tier is available; paid plans (Pro, Max) unlock higher usage limits and more capable models.

Gemini (Google) comes in two meaningfully different versions and it is worth understanding the distinction. The free chatbot at gemini.google.com is capable for general tasks. The Workspace-integrated version, built into Gmail, Docs, Drive, and Sheets, requires a paid Google Workspace subscription and is where the real productivity integration lives. If your business already runs on Google, that integration can be genuinely useful.

Microsoft Copilot is worth mentioning specifically because it is included with Microsoft 365 business subscriptions, which many photographers and small businesses already pay for. If you have an active Office subscription, you may already have access to Copilot inside Word, Outlook, and other apps without realizing it. It is worth checking before paying for a separate tool.

A Note on Free vs. Paid, and Why Privacy Should Be Your First Question

The two most visible differences between free and paid tiers across most tools are usage limits (paid plans give you more capacity before you hit a wall and have to wait) and model quality (paid plans generally access the more capable, current models). Both matter for photographers doing real business work. But there is a third difference that matters more than most people realize, and it is the one least likely to appear in the feature comparison tables: what happens to your data.

Across nearly all of these platforms, free tiers and some paid consumer tiers default to using your conversations to train future models. That means the inquiry you paste in, the client details you reference, the pricing strategy you think through out loud, all of it can become training data. Paid business tiers, by contrast, typically prohibit this by contract. The details vary by platform and change frequently, but the pattern is consistent: paying for a consumer plan buys you features, not necessarily privacy. Paying for a business tier buys you both.

This matters for photographers specifically because our work involves client information. Buddy’s mobility problems are not mine to share with a model’s training dataset.

Which One Do I Actually Use?

Mostly Claude, and I’ll tell you why, because I have spent real time with all of them and the reasons are specific rather than arbitrary.

The voice work is the biggest factor. I maintain detailed writing guides and project knowledge that I use across sessions, and Claude handles that context in a way that makes it feel more like a working partner than a prompt-and-response machine. I have run the same voice style guide through Gemini and gotten noticeably more generic results. That gap matters when the whole point is output that sounds like you.

The second reason is reliability on facts about my own business. This sounds like a low bar but it is not. Gemini, with some regularity, will invent details when generating business content: a street address that doesn’t exist, a staff member I’ve never had. Each of those is obvious when it surfaces, which is exactly the problem. If a tool is confidently fabricating things I can easily verify, I have no way of knowing what else it’s gotten wrong that isn’t as visible. That uncertainty is too expensive when I’m using the output in client-facing work.

There is more to say about Claude specifically, including some tools that are worth their own dedicated issues down the road. For now, I will just say: try them all. My experience is my experience. The right tool for your workflow is the one that gets you the output you can actually use.

The Part Everyone Skips

Here is what most photographers do when they discover AI tools are useful for writing: they start prompting. They type “write a session confirmation email for a pet photography client” and they get something back and they either use it, tweak it a little, and move on.

The results are fine. Serviceable. Occasionally they produce the word “magical.”

What they do not do, and what almost no one does at first, is spend time teaching the tool how they actually write. That gap is exactly where the generic output lives.

Training an AI model on your voice is not a technical process. You do not need to fine-tune anything or understand how large language models work. What you need is documentation: a written record of how you communicate, what words you use, what constructions you avoid, and what your writing sounds like when it is working. A writing guide, built specifically to give an AI tool the context it needs to produce output that sounds like you rather than like everyone else who used the same prompt today.

The reason this matters more than most photographers realize is that AI tools are, at their core, pattern-matching systems. Without context, they match to the average of everything they have ever been trained on. “Session confirmation email” drawn from the average of all professional correspondence produces something enthusiastic, slightly generic, and apparently quite fond of emojis. With context, with a detailed brief about your voice, your tone, your specific prohibitions, and your communication style, the same tool produces something that could plausibly have come from you.

The good news is that you set this up once and then refine it. You do not need to have it perfect before you start. Build a first version, use it for a few weeks, and pay attention to where the output still sounds off. That is your edit list. The guide improves naturally as you use it, and it pays forward into every piece of writing you produce after that.

That one sentence about Buddy came from knowing specifically how I communicate with clients: directly, with useful information, no performance of warmth, no unnecessary punctuation. The AI did not produce that sentence. I did. But with the right documentation in place, it would have.

Building that documentation, what it includes, how to construct it, and how to actually put it to work in your workflow, is what we are covering next issue. What I want to leave you with here is the framing: the time you spend on this is not setup time. It is the work. Everything downstream of it gets easier and better in proportion to how seriously you take it.

The session confirmation email is a trivial example. But the same principle applies to your inquiry responses, your blog posts, your LinkedIn content, your client follow-ups, your pricing conversations. All of it. The voice infrastructure you build once does that work across everything, which means the return compounds in a way that no individual prompt ever will.

Before Next Issue: Some Homework

You have two weeks before we get into building the actual documentation. That is enough time to do something useful with each of the tools above, even briefly.

Pick one or two you have not spent real time with and run the same task through each of them. A session confirmation email works fine. A draft inquiry response works too. The specific task matters less than using something you actually know well enough to evaluate the result.

What you are looking for is not which output is best. You are looking for where each tool feels clunky, where it misses something obvious, where it produces language you would never use. Those friction points are going to be useful context when you sit down to build your own voice guide, because a good guide is partly a list of exactly what to correct for.

Come back with a feel for the tools. The methodology will make a lot more sense when it’s solving problems you’ve already run into.

One more thing before I go. I told you I only wrote that one sentence in the confirmation email. That’s true of this entire post. I told Claude what I wanted to say, made a few corrections along the way, and what you just read is genuinely my voice, my examples, my thinking throughout. The methodology I’m describing is the methodology I used to write the issue describing it. That’s not a disclaimer. That’s the point.

The photographers who figure this out early will look, to everyone who did not, like they have some kind of magical AI setup.

They will just have done the unglamorous part first.

I spent eight years as a research specialist in cancer biology. University of Wisconsin first, bladder and prostate cancer, then the University of Virginia, breast cancer. The work was methodical and collaborative and genuinely important, and I was good at it. We published. I understood how to read data, how to question a result, how to hold a hypothesis loosely until the evidence said otherwise.

I also understood, by the end of those eight years, that it wasn’t what I was supposed to be doing.

That’s the part of the story I’ll leave mostly undeveloped here, because it deserves more space than an introductory issue can give it. What matters for now is that I left the bench with something I didn’t know I was carrying: a deep comfort with tools that are more powerful than they first appear, and a trained instinct to ask what the data actually shows rather than what you want it to show. That instinct turned out to be useful in ways I couldn’t have predicted.

The Pivot

I picked up a camera. Then I picked up another. I started volunteering at a local humane society - this was 2005 - less than a year before I opened my photography business. I couldn’t have told you at the time why the shelter work felt so connected to the photography work. It just did.

What I could tell you is that photographing shelter animals is an education in patience, in reading a subject, in earning trust quickly from someone who has no particular reason to give it to you. An anxious dog in a shelter run doesn’t care about your equipment. A frightened cat doesn’t care about your MFA. You either understand how to create safety in front of the camera or you don’t, and the animal will tell you which one is true in about four seconds.

I built a portfolio from that work. That portfolio became my graduate school submission. I got into the Academy of Art University in San Francisco, completed my MFA in Photography, and wrote a thesis project centered on shelter animals. The thesis became two books, published by Merrell Publishers: Shelter Cats in 2010, Shelter Puppies in 2011.

By the time those books were out, I had also built a photography business that moved with me across three states. Kentucky first, where it started in 2006. Then Washington, where I taught as an adjunct professor of fine arts at Washington State University. Then Tucson, in 2011, where I’ve been ever since.

Adding the Suit and Tie Work

At some point along the way, I started photographing executives. It seemed like an odd addition to a career built on animals, and more than one person said so.

Here’s what I noticed: it wasn’t odd at all. The same skill that lets me photograph a dog who doesn’t want to be there also lets me photograph a CEO who doesn’t want to be there. Reading a subject. Moving slowly enough to build comfort. Understanding what authentic looks like as opposed to performed. Knowing when to make a joke and when to just let the silence work. A nervous executive and a shelter dog are not the same thing, obviously. But the underlying skill isn’t as different as it looks from the outside.

That overlap is real, and it’s been good for the business. I’ll write about it more specifically in a later issue, because the mechanics of it - the actual techniques that transfer - are worth a detailed conversation.

On Being Told the Industry Is Ruined

I opened my photography business in January 2006. Digital SLRs were new. Film photographers were predicting the end of professional photography, because now anyone could take a hundred exposures without paying for film and something would come out usable. The barriers were collapsing. The industry was ruined.

It wasn’t ruined.

Then the iPhone arrived in 2007. Suddenly everyone had a decent camera in their pocket at all times. The barriers collapsed further. The industry was ruined again, and this time surely it would stick.

It didn’t.

What actually happened, both times, was that photography bifurcated. The lowest tier of the market, the work that was always just documentation with no particular craft or intention behind it, got absorbed by the new technology. The work that required judgment, relationship, artistic vision, or technical expertise that went beyond point-and-shoot - that work found its level. The market sorted itself. The photographers who stayed in business were the ones who understood what technology could and couldn’t replace.

I’ve watched this cycle twice. Now it’s happening a third time, and the technology is more disruptive than a new camera sensor or a smartphone. The conversation this time is louder, faster, and considerably more anxious.

Nine Months Ago

Last summer, I started using AI tools in a serious and systematic way inside my photography business. Not to generate images. To run the business: research, content strategy, drafting, workflow documentation, SEO. I’d dipped a toe in earlier, but six months ago I committed.

It changed things. It continues to change things. Not in the way the anxious corner of the internet describes, where AI arrives like a demolition crew and levels everything in its path. More like acquiring a genuinely capable collaborator who works very fast, has read more than any human ever could, gets some things exactly right, gets other things confidently wrong, and requires real editorial judgment to use well.

That last part is important. The judgment doesn’t go away. It moves to a different place in the process.

How Fast This Is Actually Moving

I want to close with a single story, because it illustrates something that pure argument can’t quite capture.

In November 2025, an Austrian software developer named Peter Steinberger published a side project on GitHub. He called it Clawdbot - a weekend experiment in AI agents, the kind of tool that doesn’t just answer questions but actually does things: reads your email, schedules your calendar, posts to Slack, orders dinner. He described it as a playground project.

By late January 2026 - roughly ten weeks later - it had nearly 200,000 GitHub stars and two million users a week. By mid-February, Sam Altman had confirmed that Steinberger was joining OpenAI to build the next generation of personal agents, with both Meta and Microsoft having also come calling. The project had gone from weekend experiment to one of the most-watched open-source projects in the world in about ninety days.

That’s not a normal product arc. That’s not even a fast startup arc. That’s a measurement of the current velocity of this technology - how quickly something that didn’t exist can become something that reshapes a market.

The photographers who position themselves well for what’s coming are the ones paying attention right now, while the conversation is still unsettled and the methodology is still being built. Not because AI is going to replace photography - I’ve watched two technologies that were going to replace photography fail to do so. But because the photographers who understood digital, and then understood mobile, ended up with durable businesses. The ones who waited to see what would happen ended up catching up at a disadvantage.

You can wait for the steam train to pass and decide later if it was worth boarding. That’s a reasonable choice if you’re close enough to retirement that it doesn’t much matter. The rest of us need to figure out how to climb aboard while it’s moving.

That’s what this newsletter is for. Let’s get started.

Michael Kloth is an MFA-trained, CPP-certified photographer based in Tucson, Arizona, specializing in executive headshots and fine art pet photography. He has volunteered as a photographer for Woodford Humane Society (Kentucky), the Benton-Franklin Humane Society and the Tri-Cities Animal Shelter (Washington), Pima Animal Care Center and the Humane Society of Southern Arizona (Arizona) for a combined 21 years.

Subscribe now

Meet Chica and Birdy. Chica experiences life at full intensity and maintains her dignity about it. Birdy requests affection primarily as a tactical maneuver. It might be genuine affection, but it could just as easily end in roughhousing or Birdy claiming Chica’s spot on the couch.

I’m an MFA-trained photographer. I used AI to put superhero costumes on my dogs. This newsletter is about what happens when those two things exist in the same brain.

This isn’t a newsletter about making silly photos with AI. It’s about where these tools actually deliver value in running a photography business - and how to navigate a toolset that keeps changing out from under you. I make no guarantees that what works for me will work for you. But I’m an art-school-trained, certified professional photographer with 20 years of experience, and I think the lessons I’ve learned are worth sharing. Follow along and I’ll show you what I’ve found. New issues drop every other week, with notes filling in the gaps.

Subscribe now