Guide · 7 min read · Updated September 2026

Spotting AI-generated photos in rentals, marketplaces and dating profiles

Most writing about AI images is about politics and news. The version that is more likely to cost you personally is duller: an apartment that does not exist, a product that will never ship, a person who is not on the other end of the conversation. These fakes are easier to catch than political ones, because the fraud has to make a specific, checkable claim about a real place, object or person — and specificity is where generated imagery falls apart.

By Ira Peoples

Rental listings: the interior that does not obey architecture

Generated interiors look beautiful and make no structural sense. Follow the walls: do they meet at plausible angles, and does the room's shape stay consistent between photos? Count the windows in each shot and check whether the light coming through them agrees with where the building faces. Look at where the floorboards run and whether their direction changes between images of what should be the same room.

Then look for the boring things generators omit because photographers crop them out: light switches, power outlets, radiators, door handles, thermostats, smoke alarms, the edge of a skirting board. A fully furnished apartment with no visible outlets anywhere is not a minimalist triumph. Check the view from the window too — generated views tend to be a pleasant blur rather than an identifiable street.

Rental listings: the checks that settle it

Reverse image search every photo in the listing. Stolen real photographs from an estate agent elsewhere are still more common than generated ones, and the search takes a minute. Then put the address into street-level map imagery and confirm the building exterior matches, along with the number of floors and window arrangement.

Finally, ask for something a fraudster cannot produce: a short live video walkthrough, filmed now, with a specific detail you nominate — open the fridge, show the view from the second bedroom, pan to the front door and read the number. A generated listing collapses at that request, and so does a stolen one. The standard warning signs still apply: below-market rent, a landlord who is conveniently abroad, and pressure to pay a deposit before viewing.

Marketplace and product photos: too clean to be owned

A private seller's photos look like a private seller took them: on a carpet or a bed, at an awkward angle, with a slightly wrong white balance and part of a room behind. Studio-grade lighting on a second-hand item is a mismatch worth noticing.

On the object itself, check text and logos at magnification. Brand marks are where generators fail hardest — letterforms subtly wrong, a logo whose proportions do not match the real one, model numbers that are not real model numbers. Look at buttons, ports, cables and seams: generated hardware frequently has a port too many, connectors that do not correspond to any standard, or a cable that changes thickness along its length.

Marketplace: ask for the proof-of-possession photo

The single most effective request in online selling is a photo of the item next to a handwritten note with your name and today's date on it, taken from an angle you specify. This defeats stolen photos and generated ones simultaneously, and legitimate sellers are used to it.

Be wary when a seller's images are all at the same angle, all at the same time of day, and all missing wear on an item described as used. Real second-hand goods have scuffs, and honest sellers photograph them.

Dating profiles: the pattern matters more than the pixels

Generated portraits have improved enormously, so stop hunting for melted ears and start looking at the set. A convincing person has photos across years, seasons, lighting conditions and companions. A generated identity usually has three or four images that share a lighting setup, an identical camera distance, and a suspicious absence of other people — because group photos with consistent faces remain hard to fabricate.

Within a single image, the giveaways are now in the accessories rather than the face: the arms of glasses passing behind an ear at an impossible angle, jewellery that changes design between photos, earrings that do not match each other, text on clothing that is not quite language, and hands interacting with objects.

Dating profiles: the checks that settle it

Reverse image search each photo, and search cropped faces separately. A stolen identity — far more common than a generated one — shows up as an active social presence under a different name.

Then ask for a live video call, or a photo doing something specific and unlikely right now: holding up three fingers, or a picture taken outdoors in the next few minutes. Someone who offers a long chain of reasons why video is impossible while the emotional intensity escalates is following a script. If money enters the conversation at any point before you have met, the answer is no regardless of how the photos check out.

The short version

In listings, look for what generators omit: outlets, switches, radiators, wear and identifiable views.
Check text and logos at magnification — brand marks are where generated objects fail hardest.
Reverse image search everything; stolen real photos remain more common than generated ones.
Ask for something only a genuine owner can produce: a live walkthrough, or the item beside a dated handwritten note.
For profiles, judge the whole set — consistent lighting, no group photos, no time span — not one face.