In November 2025, Coca-Cola released its second consecutive AI-generated Christmas ad: animated animals watching red delivery trucks roll through a snowy town. The comments arrived within hours. Viewers called it soulless, lifeless, digital slop. Social sentiment tracking showed positive reactions falling from 23.8% before the campaign to 10.2% after. McDonald’s Netherlands pulled a Christmas ad the same season. Valentino yanked an AI campaign off Instagram after viewers said it looked cheap.
AI slop is low-quality digital content produced in bulk by generative AI, usually with little human review and little reason to exist beyond filling a feed. The lesson everyone took from that winter was simple: audiences can spot it, and they’ll punish you for it.
Then researchers actually tested that assumption.
In May 2026, Ipsos and Syracuse University’s S.I. Newhouse School showed 20 ads to 3,000 US consumers, pairing human-made ads produced before 2021 with AI-generated versions built from the same creative brief. Only 13% of viewers who watched an AI ad were confident it was AI. Roughly the same share of viewers who watched the human-made ads also suspected AI. People were guessing.
The AI ads still lost. Badly. And that gap between what audiences can detect and what actually works is the part almost every article about the AI slop backlash gets wrong.
Last updated: August 2026
Quick answers
What is AI slop?
AI slop is low-quality digital content generated at scale by AI tools with minimal human review. The term was coined by developer Simon Willison in May 2024 and Merriam-Webster named “slop” its 2025 Word of the Year. It covers AI images with warped anatomy, filler blog posts, fake reviews, and ads assembled from a prompt rather than a point of view.
Why are brands getting backlash for AI content?
Brands get backlash when AI is used to remove human craft and cut costs rather than to add something. Coca-Cola, McDonald’s Netherlands, and Valentino all faced public criticism in late 2025 for AI campaigns viewers described as soulless. Gartner found 50% of US consumers now prefer brands that keep generative AI out of consumer-facing content.
Can customers actually tell if content is AI-generated?
Mostly no. In the Ipsos and Syracuse Newhouse study, only 13% of viewers were confident an AI ad was AI-made, and the same share misidentified human-made ads as synthetic. Emplifi found 87% of consumers assume brand content is at least partly AI, while only 13% feel very confident telling the difference.
What is AI slop?
AI slop is low-quality digital content generated in quantity by AI tools, published with little editorial judgment and little purpose beyond occupying space. British developer Simon Willison put the word into circulation in May 2024, describing the flood of unwanted machine output washing across the web. Merriam-Webster made “slop” its 2025 Word of the Year, and the American Dialect Society picked the same term.
The category is broader than bad images. It includes the AI-written blog posts stuffed with keywords and no information, the product reviews generated by the hundred, the YouTube videos assembled to farm watch time, and the ad creative built from a prompt instead of an idea. What unites them is not the tool. It’s the absence of anyone making decisions.
That distinction matters commercially, because it’s also how Google frames the problem. The company’s scaled content abuse policy targets content produced in bulk primarily to manipulate rankings, and it applies whether a human, a machine, or both did the producing. Google has issued manual actions under that policy since mid-2025, deindexing sites that published high volumes of unreviewed AI pages. Publishing AI-assisted work isn’t the violation. Publishing volume without judgment is.
Why are brands getting backlash for AI content?
Brands get punished when AI visibly replaces human craft to save money, and get ignored when AI quietly supports work a human still owns. That’s the pattern across every 2025 and 2026 blowup worth studying.
Coca-Cola’s second AI holiday spot drew the loudest reaction, with viewers describing inconsistent animation that switched between realism and cartoon awkwardness. McDonald’s Netherlands aired an AI-generated Christmas advert full of surreal winter scenes and pulled it after critics called it bleak. Valentino posted an AI campaign to Instagram and removed it when the response turned hostile.
None of those brands were caught by a detector. They were caught by the work being obviously worse than what audiences expected from a company of that size. The AI wasn’t the offense. The visible cost-cutting was.
Underneath the individual incidents sits a wider trust problem. Gartner surveyed 1,539 US consumers and found that 68% frequently wonder whether the content they see online is genuine, and 50% said they would rather buy from brands that avoid generative AI in consumer-facing content. A separate Emplifi study with Alchemer, covering more than 1,600 consumers in the US and UK, found 87% assume brand content is at least partly AI-made. Suspicion is now the default setting, which means brands are being judged against it whether or not they use the tools. The same dynamic is reshaping how founders think about the AI tools they pay for every month.
The detection myth
The most repeated claim in AI slop coverage, that customers can spot AI content, does not survive contact with the data. Two independent studies landed on the same number: 13%.
In the Ipsos and Syracuse Newhouse research, 13% of viewers who saw an AI-generated ad were at least somewhat confident it was AI. The identical share of viewers who saw human-made ads produced before 2021, when the tools didn’t exist, also suspected AI. Emplifi’s consumer survey found only 13% feel very confident they can tell AI-generated content from human work.
Detection software isn’t better. The Federal Trade Commission brought a case against Workado, formerly Content at Scale AI, over its claim that its AI text detector was more than 98% accurate across content from tools like ChatGPT and Claude. FTC testing put real-world accuracy at roughly 53%, barely better than a coin flip, because the model had been trained largely on academic text. The consent order became final in August 2025 and requires the company to stop the unsubstantiated claims and retain evidence for future ones.
So the audit question founders keep asking, “will people know we used AI,” has a boring answer. Probably not. Which is exactly why it’s the wrong question.
So why does AI content still lose?
AI content underperforms because it’s worse at the commercial job, not because audiences identify it and retaliate. The Ipsos and Syracuse Newhouse study is the cleanest evidence available on this, and its design is what makes it useful: existing human-made ads from before 2021 were paired with fully AI-generated counterparts built from the same strategic brief, then tested on 3,000 consumers.
Human-made ads scored 14% stronger on short-term effectiveness and 17% stronger on long-term effectiveness, both measured with sales-validated metrics Ipsos developed. Against category benchmarks, human ads outperformed by an average of 11 points. The AI versions came in 5 points below benchmark. Same brief, same message, 16-point swing in measured commercial impact.
That result reframes the whole backlash. If audiences can’t reliably tell the difference but respond differently anyway, the thing they’re reacting to isn’t provenance. It’s specificity, tension, surprise, a sense that a person made a choice. AI defaults to the average of what it has seen, and average creative has always underperformed. The tool just makes average cheaper to produce at volume.
The economics make this worse rather than better. When drafting costs collapse, the temptation is to publish more, and more average work drags the whole brand down. Founders tracking what they actually spend on AI tokens often find the cheapest output is the most expensive mistake.
AI-assisted vs AI-authored
The usable distinction isn’t AI versus no AI. It’s whether a human owns the judgment or the machine does. AI-assisted work uses the tool for speed on the parts that don’t require a point of view. AI-authored work hands over the parts that do.
| Approach | Who makes the calls | Where it breaks | Search risk | Best for |
|---|---|---|---|---|
| AI-authored | The model picks angle, structure, and examples | Generic claims, invented specifics, no verifiable sources | High: exposed to Google scaled content abuse actions | Almost nothing customer-facing |
| AI-assisted | Human sets angle and supplies facts, AI drafts and edits | Voice drift if the human pass is rushed | Low when each page is reviewed and sourced | Blog posts, email, product copy, first-draft ad concepts |
| Human-authored | Human writes, shoots, or designs end to end | Slow and expensive per unit | Lowest | Brand films, founder POV, anything trust-sensitive |
LEGO offers a clean illustration of the line. The company uses AI in internal development, but tagged its “Everyone Wants a Piece” World Cup film with #HonestlyItsNotAI to confirm the footage of Messi, Ronaldo, Mbappé, and Vinícius Jr. was real. Backend efficiency stayed private. Public representation stayed human. That’s the split most small businesses should copy, and it’s roughly the calculus behind MrBeast’s AI-native production bet too.
How to avoid AI slop in your marketing
Run every piece of customer-facing content through three tests before it ships. None of them ask whether AI touched the work. All of them ask whether a person did anything a machine couldn’t.
The specificity test. Count the concrete, checkable facts. Named companies, exact figures, dates, prices, real customer situations. A page about pricing strategy that says “consider your market position” fails. A page that says “we raised our starter tier from $19 to $29 in March and lost four customers out of 210” passes. This is the difference between Cadbury naming the chocolate on set and a competitor claiming it values craftsmanship. If you can delete your brand name and the copy would fit a competitor, it’s slop regardless of who wrote it.
The stakes test. Does the piece take a position someone could disagree with? Average content hedges because averaging is what prediction does. A recommendation with a tradeoff attached, a tactic you tried that failed, a claim you’d defend in a meeting, all of these signal that a person made a call. Pieces that could have been written by anyone about anything convert like it.
The source test. Can a reader verify your central claim by clicking one link? Primary sources beat aggregators: the SEC filing, the survey the company published, the founder’s own post. Google’s own guidance on helpful content asks essentially this question, and it’s also what makes a page citable by Perplexity, ChatGPT, and Google AI Overviews rather than replaceable by them.
For small operators, that last point is worth sitting with. If a piece of your content can be summarized without loss by a chatbot, it will be. Content that survives is content holding something the model doesn’t have: your numbers, your customers, your specific mistakes. Independent publishers working out the economics of paid newsletters ran into this earlier than most.
A practical cadence for a team of one to five: draft with AI, then spend the time you saved adding what only you have. Pull two real numbers from your own dashboard. Name the customer situation that prompted the piece. Cut every sentence that would be true for any company in your category. Most of that work takes 20 minutes and it’s the entire difference between assisted and slop.
What the no-AI brands are actually doing
A cluster of consumer brands now treat “made by humans” as a label, and the ones doing it well are narrower than the headlines suggest. Aerie, owned by American Eagle, pledged in October 2025 never to use AI-generated bodies or people in its content, extending a no-retouching promise it made back in 2014. The Instagram post announcing it became the brand’s most-liked ever at more than 40,000 likes. Aerie’s Q4 2025 sales rose 23% year over year, though the pledge was one of several changes that quarter and the brand has not published an isolated attribution.
Dove got there first. In April 2024 it committed never to use AI to create or distort images of women in its advertising, as part of a “Keep Beauty Real” campaign backed by a global study of 33,000 respondents across 20 countries. Coterie, a baby products company, told The Wall Street Journal it would keep AI images out of its social marketing entirely, with CEO Jess Jacobs saying it would never replace “the human moments that define our brand.” Coterie reports subscriber retention around 98% month over month.
Others made the craft itself the message. Cadbury shot its 2025 “It Could Only Be Cadbury Dairy Milk” film with real chocolate on set, with global brand vice president Guilherme Ferreira citing the choice to use “experts in their craft.” Polaroid ran billboards near Apple Stores and Google offices reading “AI can’t generate sand between your toes.” Blue Diamond’s Almond Breeze launched “The Pitch” with the Jonas Brothers in January 2026 under the title “No AI Needed.”
The pattern is narrow and deliberate. Every one of these brands still uses AI operationally. Dove and Aerie both do. What they pledged is specific: how the brand depicts people in creative work. That’s defensible. A blanket “we never use AI” claim from a company running an AI chatbot on its support page is not, and the gap will get expensive. New York’s AI disclosure law, the first in the US, takes effect in June 2026, at which point the difference between a pledge and a campaign slogan starts carrying legal weight.
Should a small business label its work no AI?
Only if the claim is narrow, true, and enforceable across everything you publish. For most small businesses the answer is no, and the reason is that the label is a liability without a moat.
The brands winning with no-AI positioning sit in categories where physical authenticity is the product: intimates, beauty, baby care, food, analog photography. Gartner’s data explains why. When 68% of consumers already question whether anything they see is genuine, a disclosure only helps if your customer’s purchase decision actually turns on human authorship. For a B2B software company or a local services business, it rarely does.
The stronger play for a small operator is the thing the label is a proxy for. You have access to specifics that a billion-dollar brand’s legal review would never approve: real numbers, named customers, opinions with edges on them. That’s a genuine advantage over larger competitors, and it doesn’t require a pledge. It requires showing up with information nobody else has, which is the same standard that separates faceless AI content operations that last from the ones that get deindexed in a quarter.
If you do publish a commitment, follow the Aerie and Dove template rather than the campaign-slogan version. Name the specific thing you won’t do, apply it to a defined surface, and leave your operational use of AI out of it. Vague pledges collapse under a single screenshot. Founders weighing which parts of the stack to automate at all face a related question when they evaluate which AI models to build on, and the honest answer in both cases is that the tool matters less than what you do with the output.
The AI slop backlash will keep generating headlines through 2026, and most of them will tell you customers are catching brands in the act. The measurement says otherwise. Customers aren’t catching anyone. They’re just responding, accurately, to work that nobody bothered to make good, and the fix has been available the entire time. Put a person in charge of the judgment, and let the machine handle the typing. Teams that skip that step tend to discover the cost later, in the same slow way creator burnout shows up on a P&L, and in the same way internal revolts over AI mandates taught Meta something expensive.
Sources: Syracuse University Newhouse School, Gartner, Federal Trade Commission, Emplifi, Digiday.
