It used to be very easy to spot an AI-generated image. The fingers were wrong, the text was gibberish, faces in the background looked vaguely haunted and any object with a repeating pattern tended to fall apart if you looked at it for more than a few seconds. That has changed quite quickly. The latest image generators are much better at producing people, products and environments that can easily pass while someone is scrolling through Instagram or glancing at an ad on their phone.
What has not disappeared is the underlying weirdness. It has simply moved into smaller details. Instead of a person having seven fingers, you might notice that she is technically holding a cup but not in a way any real hand would hold one. A row of supposedly identical chairs looks consistent until you realise that each chair is slightly different. A hotel mirror reflects something that is not actually in the room. The logo on a product looks familiar from a distance, but one of the letters is wrong when you zoom in.
That is why spotting AI imagery now is less about finding one spectacular mistake and more about asking whether the image holds together once you stop looking at it as a whole. AI is very good at producing the overall impression of a scene. It knows what a luxury café, skincare campaign, fashion shoot or minimalist apartment is supposed to look like. What it is still less reliable at is making every individual part of that scene obey the same logic.
Text is still one of the easiest places to look
Text used to be the most obvious giveaway because image generators simply could not handle it. That is no longer entirely true. Large headlines can now look surprisingly convincing, which means the more useful clues tend to appear in the small stuff: product labels, signage in the background, dates, prices, menu boards, disclaimers and logos.
You might see a word that begins correctly and then turns into nonsense halfway through, or the same letter drawn differently two lines later. Numbers can be oddly shaped. A logo may be almost right but have incorrect spacing, a missing stroke or a subtly altered symbol. None of this is dramatic enough to jump out during a quick scroll, but it often becomes obvious as soon as you enlarge the image.
This is particularly useful for posters and social graphics because those formats rely on exact typography. A designer may make a bad font choice, but they do not normally invent a slightly different version of the same letter every time it appears.
Repeating objects often stop repeating properly
One of the most reliable things to inspect is anything that should be identical. Look at rows of windows, bottles, chairs, plates, tiles, shelves, ceiling lights or railings. AI often gets the general pattern right while changing the individual objects inside it.
A café scene may appear to have six matching chairs, but one has a different backrest, another has an extra leg and a third seems to merge into the table beside it. A skincare campaign may show several bottles of the same product, except the label shifts slightly between bottles and the cap changes shape. In a building exterior, window spacing may become increasingly strange as your eye moves across the image.
This becomes even easier to spot when a brand uses several AI-generated visuals featuring the same product. Compare them side by side and continuity problems start to appear. The bottle is slightly taller in one image, the logo moves in another and the packaging text has somehow rewritten itself. Each individual image may look plausible, but the series does not agree with itself.
Hands are no longer the joke people think they are
The old advice to count fingers is increasingly outdated. AI-generated hands have improved a lot, and many now look perfectly normal. The better test is to look at what the hand is actually doing.
A person may have five fingers and still be holding a fork in an impossible grip. A phone may be pressed against a hand without actually being held. A handbag strap might disappear between two fingers. Someone may appear to be writing, except the pen is not touching the paper or is facing the wrong way.
The same applies to the rest of the image. Instead of asking whether a person looks anatomically correct, ask whether the interaction makes sense. Is the waiter actually carrying the tray? Is someone really sitting on that chair? Is the cyclist’s foot connected to the pedal? AI images become much easier to spot once you stop admiring the composition and start checking the logistics.
Eyes can look convincing individually but wrong together
Faces are far better than they used to be, which means obvious facial distortion is no longer something you can rely on. Eyes, however, are still worth checking carefully because both eyes have to agree with each other and with the lighting in the rest of the image.
Look at the catchlights first. In a real photograph, reflections in both eyes should generally correspond to the same light source. In generated portraits, one eye may show a large rectangular reflection while the other has a tiny circular one. The irises may also have slightly different patterns or levels of sharpness, particularly in very polished beauty portraits.
There can also be a subtler problem with gaze. Both eyes look normal on their own, but they are not quite looking at the same point. It is not dramatic enough to look like a squint; the face simply has a slightly unfocused quality that is difficult to explain at first glance.
This is one reason AI portraits can look attractive but strangely lifeless. The facial features are technically present, yet the tiny coordination cues we are used to reading in real faces are not completely consistent.
Teeth and mouths are another good place to zoom in
Teeth have become much better too, but mouths still contain a lot of small repeating structures that need to line up properly.
In some generated portraits, individual teeth merge together, change size halfway across the smile or become unnaturally uniform. A person may appear to have a perfect row of white teeth until you notice that the tooth boundaries do not make sense. Lips can have a similar issue, especially at the corners of the mouth, where they may blend into skin or produce an odd crease.
Another tell is the relationship between the mouth and what someone is doing. A person eating may have a utensil positioned in front of the lips without actually entering the mouth. Someone drinking through a straw may have the straw touching the wrong part of the face. These are small details, but they are often more useful than checking the face for some spectacular deformity.
Ears, earrings and glasses often disagree with one another
Humans are not perfectly symmetrical, so two ears do not need to be identical. What does matter is whether the accessories attached to them make sense. Earrings can change shape from one side of the head to the other, vanish behind hair without reappearing or connect to an earlobe at an impossible point.
Glasses are even more useful because the frame has to maintain consistent geometry across the face. One arm may disappear into the hair, the bridge may not sit correctly on the nose or the lens edges may change thickness. In profile shots, the far side of the frame can become especially strange.
Jewellery in general is worth inspecting. Necklaces can merge into skin, pendants shift position without a chain supporting them and bracelets sometimes become part of a sleeve. None of these mistakes is spectacular on its own, but together they can make a portrait feel subtly artificial.
Hair is better than ever, but the edges can still betray it
Long hair gives image generators plenty of opportunities to hide mistakes because real hair is already complicated. A few stray strands do not prove anything.
The useful clues tend to appear where hair meets something else. A strand passes through an earring rather than around it. Hair merges into the collar of a shirt. A ponytail begins in a place that does not quite correspond to the person’s head. Wisps on one side of the face are sharply defined while equally distant strands on the other side dissolve into the background.
Braids and complicated hairstyles can be even more revealing because the pattern is supposed to continue logically. A braid may start correctly and then turn into a rope-like texture, or individual sections weave together in a way that could not physically happen.
This tends to be easier to see in beauty and fashion images because the hair is usually rendered with a very polished, almost shampoo-ad level of detail. The overall effect is beautiful, but the structure underneath it does not always hold up.
Clothing seams, buttons and straps are easy to overlook
Clothes are another area where the image can look completely normal from a distance while the construction makes very little sense.
Look at seams. Does the seam on a jacket continue where it should? Does a shirt collar actually connect around the neck? Are buttons aligned with buttonholes? Does a belt pass behind and in front of the correct parts of the clothing?
Straps are particularly useful. Handbag straps, bra straps, apron ties and backpack straps often disappear into clothing or reappear somewhere they could not possibly have travelled. A shoulder bag may look fine until you realise the strap is not actually connected to the bag.
Patterns can also give things away. Checks, stripes and prints are supposed to follow the shape of the garment. In generated images, a pattern may suddenly change scale, break at a seam or drift in a different direction around the body.
Again, none of this automatically proves an image is AI-generated. Real clothes fold, twist and photograph strangely. The point is to look for several small inconsistencies that all seem to be happening at once.
Skin can look a little too uniformly human
This is one of the least reliable signs on its own, but it becomes useful when combined with other clues.
AI portraits often have skin that is not exactly smooth, because generators have learned that obviously airbrushed skin looks fake. Instead, they may add a very controlled version of realism: a few freckles, some pores, a faint blush, one or two tiny blemishes and perhaps a little under-eye texture.
The problem is that these imperfections can feel distributed rather than lived-in. The face has exactly enough texture to signal “real person” without the messy variation you usually get in actual skin.
Commercial retouching can obviously produce the same effect, so this should never be treated as evidence by itself. It is more useful when the skin looks slightly too perfect and the eyes, hair, earrings or background contain other inconsistencies.
Body proportions can be correct while posture is not
Modern generators are much better at full-body images, but you can still find problems when you look at weight distribution and posture rather than simple anatomy.
Someone leaning against a wall may not actually be touching it. A seated person’s hips and legs may not correspond properly to the chair. A model appears to put her weight on one leg while the position of the foot suggests otherwise. A person can also have perfectly normal limbs that connect to the torso at slightly implausible angles.
Exercise and sports images are particularly revealing because the body has to obey very specific physical relationships. A runner’s legs may be in a position that could not occur during the stride shown. Someone lifting a weight may not actually be supporting its weight. A yoga pose can look impressive while the joint positioning would be extremely difficult or impossible for a real person.
The clue is less “this body part looks weird” and more “could a person actually be standing like this?”
Background people are often where the image starts to unravel
Image generators tend to spend their effort on the main subject, which means the people nobody is supposed to look at can be surprisingly revealing. In a restaurant image, the woman in the foreground might look completely believable while someone three tables behind her has an oddly smooth face, a half-formed hand or an arm that seems to belong to another diner.
Crowds are especially difficult because there are so many relationships to maintain at once. People have to sit on chairs properly, hold objects, interact with one another, wear coherent clothing and exist in the same physical space. You may not get the spectacular mangled faces that were common a few years ago, but you still see subtler failures: people merging into each other, unfinished facial features, clothing that blends together or limbs that do not quite attach where they should.
These are exactly the kinds of things that disappear at normal social-media size, which is why obviously generated ads can still make it through several rounds of approval without anyone noticing.
The same person can become a different person from one image to the next
This is particularly useful when you are looking at a carousel, campaign or set of supposedly related photographs.
A generated model may look consistent at first, but compare the details. The ears are a slightly different shape. A mole disappears. The distance between the eyebrows changes. Teeth are arranged differently. The nose becomes narrower. An earring swaps sides.
This matters increasingly because brands are using generated “models” across several posts and campaigns. A human viewer may not consciously notice that the face has changed, but the lack of continuity can still make the campaign feel synthetic.
The easiest way to check is to ignore the overall face and compare the boring details. Ears, teeth, jewellery, hairline and small skin marks tend to be more useful than asking whether the person generally looks like the same person.
Posters and social graphics often have a different problem: they imitate design without really having a design system
This is one of the more interesting tells because it has less to do with anatomy and more to do with judgement. AI-generated posters often contain all the things we associate with contemporary graphic design: tiny numbers, arrows, crosses, decorative labels, little pieces of micro-copy, pseudo-technical markings, floating words such as “DISCOVER” and the occasional mysterious “01 / 06” in a corner.
The problem is that a lot of these elements do not appear to mean anything. They are there because posters often contain them, not because they help explain the product or guide the viewer through the layout. Once you ask why the arrow is there, what the number refers to or what the tiny caption is supposed to communicate, there is often no good answer.
The underlying layout can have the same problem. Three cards that are meant to be identical have slightly different padding. The icons do not quite belong to the same visual family. Margins drift from one section to another and typography changes for no obvious reason. A human designer can make all of these mistakes too, of course, but the giveaway is usually the accumulation of choices that do not seem to have any rationale behind them.
Some images do not contain a single obvious mistake and still look unmistakably synthetic
This is harder to pin down, because polished commercial photography exists and not every attractive image is AI. Still, there is a certain style of generated lifestyle photography that has become very recognisable.
The lighting is perfect, the skin is flawless but not too flawless, the furniture is arranged beautifully, the coffee cup is sitting exactly where a coffee cup ought to sit and there is almost always some tasteful greenery somewhere in the frame. Even the wrinkles in the linen look art-directed. Nothing is awkward, inconvenient or accidentally ugly.
Real commercial shoots are obviously styled, sometimes heavily, but they still have to deal with reality. Fabric behaves badly, reflections appear in places nobody wanted them, objects cast shadows at inconvenient angles and rooms contain little imperfections that are difficult to remove completely. Generated images can feel strange because all of those annoyances have been optimised away at the same time.
That is often why someone can look at an image and say “this feels AI” without being able to point to a single broken finger or mangled sign. The scene looks less like a photograph of a real place and more like an average of every photograph ever tagged “warm minimalist café interior”.
Why this matters more for businesses than it does for random internet images
The problem for brands is not simply that customers may notice AI. It is what they infer from noticing it. Plenty of people genuinely do not care, and there are situations where AI-generated visuals make complete sense. But there is also a growing group of consumers who associate obvious AI creative with cheapness, laziness or a lack of care.
That reaction becomes more important when a business is selling something where authenticity is part of the product. If a bakery shows an AI-generated cake, customers may reasonably wonder whether the cake actually exists. If a hotel uses a generated room, people may start questioning whether the property really looks that way. If a clothing brand generates models wearing its clothes, customers lose one of the main things they need from the image in the first place: some idea of how the garment actually looks and fits.
The same risk applies to businesses built around taste or craft. A florist, interior designer, photographer, beauty brand or handmade business may spend years trying to convince customers that quality and attention to detail matter, only to undermine that message with an obviously synthetic ad created because it was quicker than organising a proper shoot.
Consumers do not have to hate AI for this to become a branding problem
There is a tendency to frame the discussion as though everyone either loves AI or is morally opposed to it. Most customer reactions are probably much less dramatic than that.
Someone sees an obviously generated advertisement and simply thinks the business looks a bit cheap. They scroll past. They trust it slightly less. The restaurant feels a little dodgier, the skincare company feels more dropshippy or the property agent’s post looks like it was produced in ten minutes. Nobody announces a boycott; the brand just loses a little credibility.
At the stronger end of the spectrum, there are also consumers who actively object to companies replacing photographers, illustrators, artists and designers with generative AI. Brands have already discovered that people will screenshot bad AI ads, point out the errors and share them for everyone else to inspect. Once that happens, the image stops being an advertising asset and becomes content about how little care appeared to go into the advertising.
That is why the usual defence — “but the image only cost us $20” — is not especially useful. The production cost tells you what you paid for the asset. It tells you nothing about what the asset may cost the brand if it makes the company look cheap or careless.
This does not mean businesses should never use AI images
AI can be extremely useful in the creative process. It is good for testing rough ideas, building mood boards, exploring visual directions, producing internal mock-ups or helping a team decide whether a concept is worth spending money on. There is also nothing inherently wrong with using AI as the final medium when the artificiality is part of the concept.
What is less convincing is using it as a substitute for something the customer actually needs to see accurately. A conceptual image on a software landing page is very different from a restaurant generating food it does not serve or a hotel generating a room that does not exist.
The useful distinction is not “AI good” versus “AI bad”. It is whether the image is doing a job that requires authenticity, consistency and judgement. If it is, there is still a strong case for using a photographer, illustrator, designer or another proper creative.
Proper creative does not necessarily mean an enormous production budget
Businesses sometimes talk about AI as though the alternative is automatically a five-figure shoot. It usually is not.
A café can photograph the food it actually serves. A florist can show arrangements it has actually made. A founder can appear in the company’s own content. A designer can build a reusable set of templates so the brand does not need to reinvent its visual identity every week.
There is also a value to creative judgement that is easy to miss when everything is measured by speed. A good designer notices that the logo is slightly wrong, that the visual hierarchy is unclear or that the decorative arrow is pulling attention away from the CTA. A photographer understands how to make the product look good without making it look like a completely different product. An illustrator can keep a character consistent across an entire campaign instead of subtly redesigning it every time.
AI can produce something very quickly. That is useful, but speed and quality are not the same measurement.
How to check an AI-generated image before publishing it
If a business is going to use generated visuals, the minimum standard should be to inspect them properly before they go live. Zoom in and read every bit of text that is supposed to be readable. Check the logo, prices and product packaging. Compare anything that appears more than once. Look at hands, eyes, teeth, earrings, clothing seams, reflections, background people and the points where objects touch each other.
For a poster or social graphic, go one step further and ask why every element exists. If there is a tiny number, arrow, label or decorative word, does it actually contribute anything? Are the margins consistent? Are repeated cards really identical? Does the design look coherent because someone designed it, or does it merely contain enough fashionable design elements to create that impression from a distance?
Most importantly, look at the finished image as a customer rather than as the person who knows how quickly it was produced. Does it make the business look more credible? Does it represent the product accurately? Would you still be happy with it if someone zoomed in, screenshotted the weirdest detail and posted it elsewhere?
That last question is probably more useful than any AI-detection tool.
Frequently asked questions
What are the most obvious signs of an AI-generated image?
The strongest clues are usually inconsistencies rather than one dramatic mistake. Small text may be wrong, repeated objects may change shape, logos may be slightly inaccurate and people’s interactions with objects may not make physical sense. With portraits, eyes, teeth, earrings, hair, clothing seams and continuity across several images can also be revealing.
What are the biggest signs of AI in human faces?
Look for eyes that do not quite agree with each other, inconsistent catchlights, teeth that merge or repeat unnaturally, ears and earrings that do not match, hair blending into clothing and small facial details that change between images. None of these proves an image is AI on its own, but several appearing together can be a strong clue.
Are strange hands still a reliable way to spot AI?
Not as reliably as before. Modern image generators are much better at producing hands, so it is more useful to look at how the hand interacts with another object. A hand can look anatomically normal while still holding a cup, phone or piece of cutlery in a way that makes no sense.
How can you spot AI-generated posters and social media graphics?
Look closely at typography, logos, repeated icons, spacing and decorative elements. AI-generated graphics sometimes contain lots of visual details that resemble contemporary design but have no real purpose, such as unexplained arrows, random numbers, fake labels or tiny text that turns out not to say anything.
Why do some AI-generated images look fake even when nothing is obviously wrong?
Sometimes the clue is not an error but an unusual level of perfection. Lighting, skin, furniture, styling and composition may all look ideal at the same time, with very little of the randomness or inconvenience that appears in real scenes. That does not prove an image is AI, but it can make the image feel synthetic.
Can AI-generated advertising put customers off?
Yes, although reactions vary. Some consumers do not care, while others associate obvious AI creative with cheapness or a lack of effort. Negative reactions are more likely when the image is supposed to show a real product, place or experience that customers expect to receive.
Should businesses avoid AI images altogether?
Not necessarily. AI can be very useful for concepts, mock-ups and early-stage creative work. The issue is whether the final image represents something customers need to see accurately or whether the brand is relying on generated visuals in a way that makes the work feel cheaper or less trustworthy.
The real question is not whether customers can detect AI
AI-generated images will keep improving, and some of today’s most obvious tells will eventually disappear. That is why it makes little sense for a business to base its creative strategy on the hope that customers will not notice.
The better question is whether the work is actually good. Does it represent the product properly? Does it feel considered? Does the design make sense? Does it make the business look more credible, or does it look like the fastest possible way to avoid commissioning creative work?
Most businesses do not have a shortage of content anymore. AI has made producing more of it easier than ever. What they still need is content that looks like someone cared enough to make it properly.