On Instagram, AI Slop Costs You Customers Before It Costs You Reach

It can, but mostly through customer trust rather than reach. Instagram does not penalise most AI-generated images, yet shoppers increasingly distrust or disengage from brands that post them, especially when an image misrepresents the product. Using AI to improve your own real photos and videos carries far less risk.

Last week Just Eat apologised after its AI menu images turned a tiger prawn dish into a bowl of tiger cubs. It is a funny story until you picture the Northern Ireland café whose bacon and egg soda bread appeared online as a glass of cola. Most small businesses will never post anything that strange, but plenty are posting images that are wrong, or simply forgettable, in quieter ways. This piece looks at who judges that content, and where AI belongs for a business with a real counter, chair or stockroom.

Slop has a definition now

Merriam-Webster chose slop as its 2025 Word of the Year, defining it as low-quality digital content, usually churned out in volume with AI.¹ The editors noted the word carried more mockery than fear, which matters for a brand. People do not feel threatened by an AI-generated latte with three handles. They laugh at it, screenshot it and send it to a friend, and the business that posted it becomes the punchline.

For a small business, the dictionary definition is a little too narrow. Slop on a brand account is any image, video or caption that could have been posted by anyone, or that shows something the business does not sell. A cake that was never baked in your kitchen counts. So does a model wearing a dress whose drape and fit were invented by software, and a caption so generic it would work equally well for the salon across town.

Polish has very little to do with it. A beautifully rendered AI image of a dish you do not serve is still slop for your purposes, because the thing it promises does not exist. A slightly underexposed phone photo of the dish you do serve is not, and shoppers appear to draw the line in roughly the same place.

Instagram will not filter it

It would be convenient if the algorithm buried slop on your behalf, but it does very little of that directly. Instagram's ranking work over the last two years has centred on originality, which Instagram head Adam Mosseri names as one of its ranking principles.² In April 2026 the rule extended from Reels to photos and carousels, so accounts that mostly post other people's material, judged across roughly a month of output, stop being shown to people who do not follow them.³ Around the same time, Instagram said 75% of its US recommendations were coming from posts its systems treat as original.⁴

Those rules target reposting, and a brand that generates its own AI images is, technically, posting content it created. The other big change, at the end of August 2026, introduced an "AI-generated profile" label and limited reach for accounts featuring an AI-generated person that do not use it.⁵ Engadget noted that Meta has done little to discourage AI content generally and that the label covers one particularly misleading kind of account.⁵

That leaves the judgement with the people scrolling. The signals Mosseri says matter most are watch time, likes per reach and sends per reach, with sends weighing slightly more when Instagram decides whether to show a post to non-followers.² Every one of those is a human reaction. If nobody lingers on a glossy generated image or forwards it to a friend, the algorithm will register that, but only because people got there first.

Customers are keeping score

Sprout Social surveyed 2,250 social media users in the US, UK and Australia in February, and 56% said they saw AI slop often or very often, while 66% said they had become more selective about what they engage with than a year earlier.⁶ Half of the Gen Z respondents had unfollowed, muted or blocked a brand or creator because its content felt like AI slop. Clutch's August survey of 601 consumers found 53% were less likely to buy from brands they knew used AI in their social content.⁷

In the Clutch survey, 64% said they would trust a brand or creator less if AI-generated content misrepresented a product, while 42% viewed AI positively for educational content.⁷ People are not rejecting every use of the technology. They are reacting to being shown something untrue, which is precisely the risk a restaurant, salon or clothing label runs when it generates pictures of its own products.

Shoppers are also noticing more than brands tend to assume. An April report on Retail Technology Show research cited Klaviyo data showing half of shoppers had spotted low-quality AI material in brand social campaigns over the previous six months, and 48% said authentic, human-led content was more likely to get them to buy.⁸ An owner who assumes nobody can tell is betting against a large group of people who say they can.

Most slop looks fine

Tiger cubs make headlines, but most AI content on small business accounts fails far more quietly. Picture a generated flat white on a marble counter, a caption that opens with a question and ends with three emojis, and a request to tag a friend. None of it is wrong, and none of it belongs to anyone in particular. Followers do not complain about posts like that; they scroll past, and over a few weeks the account teaches them that its posts are safe to skip.

Volume makes this worse. In a Gartner survey of 1,006 US consumers, published in September, 65% said brands were producing too much AI-generated marketing content and 57% said it had made them less trusting of brand messaging overall.⁹ A tool that lets a business publish seven interchangeable posts a week instead of three real ones is feeding the exact pattern those people are reacting against.

Sprout Social also asked what people want from brands instead. Educational posts about products or services came top at 40%, followed by community-focused content at 27% and behind-the-scenes content at 19%.⁶ All three depend on things a generator does not have, such as what went into today's special, how a stylist rescued a colour that went brassy elsewhere, or how a linen shirt really creases after a day's wear. That detail makes an account recognisable, and it makes a post worth sending to someone.

The menu photo test

Food shows the cost of misrepresentation most clearly, because picture and plate get compared every time an order arrives. Just Eat had been trialling AI-generated images for UK restaurants that had no photos of their own, and last week customers started sharing the results.¹⁰ A jacket potato appeared as a winter jacket covered in baked potatoes. Tiger prawns became tiger cubs swimming in sauce, and that café's bacon and egg soda bread was rendered as a glass of cola with an egg and a rasher of bacon floating in it.

The images were labelled as AI-generated, and Just Eat said every one had been checked by a person before it went live.¹⁰ Neither safeguard helped. A label tells people how a picture was made and says nothing about whether it is accurate, and a reviewer who has never eaten at a café has no way of knowing what its soda bread should look like. Just Eat apologised, removed the images and paused the rollout.

Small businesses should still label generated imagery, and nearly nine in ten people in the Clutch survey want them to.⁷ Instagram now makes a label compulsory for one category of account, profiles built around an AI-generated person.⁵ Disclosure does make your use of AI known, though, and Clutch's 53% figure measures people who knew a brand was using it. A business that labels honestly while still posting invented products is moving itself into that group.

Delivery platforms have been here before. In 2024 Zomato's chief executive, Deepinder Goyal, said AI-generated dish images were leading to more refunds and lower customer ratings, and the platform started removing them and offered restaurants real food photography instead.¹¹ A misleading photo turns into refund requests, one-star reviews and customers who quietly order elsewhere next time.

The same logic holds outside food. An AI-generated balayage on a salon's grid sets an expectation that a real colourist then has to meet under real lighting. A fashion brand that shows a generated model in its knitwear is making a claim about how that jumper hangs on a body, and its returns will report on whether the claim was true.

So where does AI belong?

Staying away from AI altogether is the wrong conclusion to draw. For a small business, AI is much safer when it works on material the business already owns rather than inventing the product itself. Most owners have hundreds of photos and clips in a camera roll, and the bottleneck has always been time: choosing the right one, finding the three seconds of video that will stop someone scrolling, and writing a caption that sounds like the business. AI is genuinely good at that work, and none of it requires making anything up.

Mosseri made a related point in his end-of-year posts. He argued that the polished, square-photo Instagram feed is finished, that imperfection is becoming a sign something is real, and that the question for creators is shifting towards whether they can make something only they could have made.¹² "Flattering imagery is cheap to produce and boring to consume," he wrote. For a café or a barber, that is encouraging, because the thing only they can make is already in the camera roll.

A simple check before posting catches most slop, whether it is invented or merely interchangeable. The questions are plain, because the people judging your posts are not reading them as marketers. Run every post past these, whatever tool produced it:

The last question is the one the Just Eat trial shows up. Human review only protects a brand when the human knows what the dish, the cut or the product should look like, which usually means the owner or someone who works the floor. That is why tools built around a business's own media make more sense than tools built to fill a feed. Asteris.ai's Instagram tools for restaurants and cafés start from a kitchen's own dish photos and footage, look for the strongest frames and hooks, and leave the final approval with someone who knows the menu.

Post the dish you actually serve

The businesses most exposed to AI slop are the ones with the least time, and they are also the ones being sold tools that promise a full feed without anyone picking up a phone. A small café or independent salon trades heavily on the expectation that the thing on the screen is the thing in the shop. Every invented or interchangeable post spends a little of that expectation, and on Instagram it is far harder to win back than it was to spend.

A better use of AI is getting more out of what you already make. Your real photos and clips are the one thing a competitor, a content farm or an image generator cannot copy, and the direction Instagram has taken this year points the same way.¹²⁴ Spend the time AI saves on taking slightly better pictures of the real thing, and let the software handle the cropping, the hook and the schedule. Then have someone who knows what a good plate looks like press approve.

Sources