DeAnn Wiley found the house on Zillow. A Detroit listing, photographed with the kind of warm afternoon light that makes even vinyl siding look dignified. Green lawn, clean windows, a front porch that suggested Sunday mornings with coffee and a newspaper. She pulled it up on Google Maps street view, because she is a careful person, and discovered what was actually sitting at that address: boarded windows, a lawn that was more dirt than grass, and a facade that looked like it had been abandoned sometime during the Obama administration.
Her post on X went viral, not because the deception was sophisticated but because it was easy.
Forty-five seconds in ChatGPT is how long New York realtor Jason Haber told PetaPixel it takes to turn an actual listing photo into something the property never was and possibly never could be. Traditional virtual staging, the kind where a human operator places digital furniture into empty rooms using specialized software, costs roughly $500 per room, which meant that deception at scale required a budget. AI staging costs nothing, and when the cost of deception collapses to zero the volume of deception does exactly what you would expect.
Staging Is Not the Problem. Fabrication Is.
Real estate photography has always trafficked in aspiration, and nobody pretends otherwise. Wide-angle lenses make rooms look larger than they are, twilight shots bathe even the most ordinary facade in golden warmth, and professional staging places furniture that costs more than most buyers' annual car payments into living rooms that were empty an hour ago. Nobody pretends this is documentary photography, and nobody should.
But staging has a physical constraint that AI removes entirely. You can place a $4,000 Restoration Hardware sofa in a living room, but you cannot place a living room that does not exist. You can light a kitchen to minimize the dated countertops, but you cannot replace the countertops with Calacatta marble that the seller never installed. A staged photo alters mood while an AI-altered photo alters reality, and the distance between those two acts is the distance between marketing and fraud that happens to look like marketing.
A 2025 study in the International Journal of Information Management confirmed what anyone who has been catfished on a dating app already knows intuitively: consumers perceive AI-generated real estate images as less authentic, less professional, and more misleading than traditional photography, and those perceptions directly reduce purchase intent. For high-involvement decisions, the kind where you are committing $600,000 and thirty years of mortgage payments, trust is not a nice-to-have but the entire mechanism by which a listing converts to a showing.
California Moved. Forty-Eight States Have Not.
California Assembly Bill 723, effective 2026, requires that any digitally altered image used in a real estate listing or advertisement include two things: a clear disclosure statement that the image has been digitally altered, and a link, URL, or QR code to the original unedited photograph. "Digitally altered" means exactly what it should mean: modified with photo editing software or AI to add, remove, or change visual elements including fixtures, furniture, appliances, paint colors, landscaping, flooring, facades, or exterior views. Noncompliance exposes agents and sellers to liability under consumer protection law and disciplinary action by the California Department of Real Estate, per analysis from Bornstein Law.
New York has not passed equivalent legislation, but Secretary of State Walter T. Mosley issued a formal "trend alert" warning homebuyers about AI-generated listing images, citing Real Property Law §441-c (prohibition on dishonest and misleading advertisements), General Business Law §§349 and 350 (deceptive acts and false advertising, carrying substantial monetary penalties), and 19 NYCRR §175.25(c)(9) (all advertisements must include honest and accurate depiction). New York's approach: we do not need a new law because the existing ones already prohibit this.
Everyone else has said nothing, which means that buyers in 48 states navigate a marketplace where AI-altered listing photos carry no specific disclosure obligation and enforcement depends entirely on how aggressively a local MLS or state attorney general chooses to apply consumer protection statutes written decades before generative AI existed.
What You Can Actually Do About It
If you are buying a home in 2026, you need a verification protocol that did not exist two years ago, because the listing photo can no longer be taken at face value and your agent may not know the difference between a staged room and a fabricated one.
Cross-reference with satellite and street view. Google Maps street view, Apple Maps Look Around, and satellite imagery provide time-stamped exterior views that AI-altered listing photos cannot retroactively modify. If the listing shows a lush garden and the most recent satellite capture shows bare dirt, you have your answer. This takes ninety seconds and catches the most egregious fabrications.
Examine lighting inconsistencies. AI compositing frequently produces shadows that fall in different directions within the same image, reflections that do not match surrounding objects, or lighting temperature shifts between the foreground subject and the background environment. A human photographer working with natural light produces consistent shadows because there is one sun. An AI generator pulling elements from different source images produces inconsistent shadows because it is stitching together fragments that were never in the same room at the same time.
Request the unedited originals. In California, AB 723 makes this a legal requirement. Outside California, there is no statutory obligation, but any agent who refuses to provide original, unedited photographs should raise the same alarm as a seller who refuses a home inspection. You are not asking for anything unreasonable. You are asking for evidence that the property you are considering spending half a million dollars on actually looks like its photographs.
Check the metadata. Original photographs contain EXIF data with camera model, timestamp, GPS coordinates, and exposure settings. AI-generated or heavily edited images frequently strip or overwrite this metadata. If a listing photo has no EXIF data at all, that is not proof of manipulation, because platforms like Zillow and Redfin sometimes strip metadata on upload, but it is a data point worth noting alongside your other observations.
If You Are an Agent, Read This Twice
National Association of Realtors guidelines already prohibit misleading images, but NAR's enforcement mechanism is the local MLS, and MLS organizations vary wildly in their capacity and willingness to police AI-altered content. Some have adopted explicit AI disclosure requirements, but many have not updated their policies since before generative AI existed as a consumer tool, and enforcement where it occurs at all is reactive: a complaint triggers a review, and nobody is proactively scanning 2.3 million active listings for AI artifacts.
This means that agent liability currently depends on geography more than behavior. An agent in Sacramento who uses ChatGPT to add landscaping to a listing photo faces specific statutory exposure under AB 723. An identical agent doing identical work in Phoenix faces whatever a court decides to apply from Arizona's general consumer protection statutes, which were not written with AI image generation in mind and have never been tested in this context.
Twenty percent of homebuilders now use AI for marketing materials, according to a July 2025 NAHB/Wells Fargo survey, with strong interest in expanding that usage further. As generative tools get cheaper and more accessible, the volume of AI-altered imagery in real estate marketing will increase regardless of what any industry association recommends. Regulation will follow the disasters, not prevent them.
Where This Gets Architecturally Interesting
Consider what it means for the relationship between physical space and its representation when the representation is no longer constrained by the physical space. A photograph of a building used to be evidence. Imperfect evidence, certainly, subject to the photographer's choices about angle, lighting, and timing, but still fundamentally tethered to something real. AI severs that tether completely, and a listing photo can now depict a property that exists nowhere except in the latent space of a diffusion model while the buyer scrolling through Zillow at 11 PM on a Tuesday has no mechanism to distinguish the real from the generated without actively investigating each image.
This is not a technology problem but a trust architecture problem, because every real estate transaction rests on an information asymmetry that the seller knows more about the property than the buyer, and the entire regulatory apparatus of disclosures, inspections, and appraisals exists to narrow that asymmetry to a manageable level. AI-altered listing photos blow the asymmetry wide open again, because the buyer cannot trust the first piece of information they encounter about any property. Forty-four percent of consumers in the Cotality survey said they would pay someone to verify AI-generated decisions, which is a remarkable statistic: nearly half the market is willing to spend money to confirm that the information they are receiving is real.
Novel Finding: Two-State Regulatory Analysis
Combining California's AB 723 text with the New York Department of State's trend alert reveals an underexamined regulatory divergence. California created a new, specific disclosure obligation with defined compliance mechanics (disclosure statement plus link to original image). New York asserted that existing consumer protection law already covers AI-altered listing photos and declined to legislate new requirements, and both approaches have structural weaknesses that will matter when enforcement arrives.
California's specificity creates a compliance checklist that agents can follow, which makes enforcement straightforward but may fail to anticipate AI capabilities that evolve faster than legislative language. "Digitally altered" as defined in AB 723 covers current image editing and generation techniques, but a sufficiently advanced model that generates photorealistic images without "altering" an existing photograph might argue it falls outside the statute's scope.
New York's reliance on existing law avoids that definitional trap but creates enforcement uncertainty, because applying General Business Law §349 to AI-generated listing photos requires a court to analogize from precedent that predates the technology. Neither approach has been tested in litigation as of May 2026.
Limitations
This analysis covers regulatory frameworks in California and New York only. Forty-eight other states and thousands of local MLS organizations have their own rules, and we did not audit them. Cotality's survey methodology is not publicly available beyond the summary statistics reported in trade press, which means the 30-to-16 percent trust decline cannot be independently validated from published sources. Consumer detection techniques described above (satellite cross-reference, EXIF inspection, shadow analysis) are practical heuristics, not forensic methods, and will become less reliable as generative AI improves. AB 723 has not yet produced enforcement actions or case law, so its practical impact on agent behavior remains theoretical. We did not interview agents, buyers, or MLS administrators for this article; all claims are sourced from published reports, regulatory filings, and industry surveys.