AI slop ads are cheap to make and expensive to run. In a 2026 Ipsos test of 20 ads, AI-made versions were 14% weaker on short-term sales impact and 17% weaker on long-term brand impact than human-made ones on the same brief, even though most viewers could not tell which was which. Separately, Gartner found that half of US consumers prefer brands that avoid GenAI in consumer-facing content. The cost of AI creative is not production; it is the trust and memory it fails to build.
- 14% and 17%: how much stronger human-made ads scored than fully AI-made versions on short- and long-term effect (Ipsos, 2026).
- Only 25% of viewers of an AI ad were confident it was AI-made; the gap shows up anyway.
- 50% of US consumers prefer brands that avoid GenAI in consumer-facing content (Gartner, 2026).
- ASCI's 2026 guidelines sort AI ad content into banned, labelled and no-label tiers.
- The saving in production can return as a larger loss in media efficiency.
I use AI in production every week, and I have seen it cut cycle times sharply in teams I have run. So this is not an argument against the tools.
It is an argument about where the cost moves when the tools make making things nearly free. The cost moves from the production budget to the media budget and the brand, where it is harder to see and slower to show up.
What counts as AI slop in advertising?
AI slop is generated content that is credible enough to publish and generic enough to forget. It is not the same as AI-assisted work:
- Not slop: a storyboard drafted with an image model and then shot, cast and edited by people.
- Usually slop: a spot generated from a reverse-engineered brief and published without anyone deciding what it should make a viewer feel.
Why the word spread
The phrase spread because the volume did. On Hacker News in July 2026, a founder asked how to market at all "in the age of slop", describing a market where products are trivially easy to build and every channel is full.
The replies are worth reading: the advice that held up was about specific forums, real conversations and targeting, not about producing more. The same pressure applies to ads. When a hundred variants cost the same as one, the feed fills with variants, and the scarce thing becomes an ad someone remembers.
In mid-November 2024, Coca-Cola released a set of holiday spots generated entirely with AI, an homage to its 1990s Holidays Are Coming trucks. Within days the reaction on social media turned. HuffPost UK, writing on 18 November, collected about 20 posts from the week: viewers pointed to distorted hands, eyes and fingernails, a dog that did not move, and a bottle the characters never quite touched. Others objected to AI's environmental cost and its threat to creative jobs, and some said the films simply looked like a copy of Coke's old Christmas ads.
A Coca-Cola spokesperson told The Independent the campaign showed the brand's commitment to innovation. The company made another AI version the next year, and the backlash returned with it.
HuffPost UK: Why Coca-Cola is facing a backlash over its AI-generated "Holidays Are Coming" ad (Nov 2024)

Can people tell an AI-made ad from a human-made one?

Mostly no, and that is what makes the effectiveness gap interesting. In the Ipsos study with Syracuse University, only 25% of people who saw an AI-made ad were at least somewhat confident it was AI, and 40% of all viewers were unsure either way. Yet the same viewers rated the human-made ads as more eye-catching, more imaginative and more entertaining.
The team took 10 real 30-second ads from major brands, had Google's Gemini reverse-engineer each creative brief and write a shot list, and fed that into OpenAI's Sora 2 with no human post-production. All 20 ads were tested with 3,000 US consumers in Ipsos Creative Spark, which shows ads in a distracted, feed-like setting and predicts sales and brand effects. Human-made ads scored 14% higher on the short-term Creative Effect Index and 17% higher on the long-term Equity Effect Index. The gaps were biggest on being entertaining, unique and talked about. The limits: 10 pairs is a small set, the AI ads had no human editing, and the measures are Ipsos's own validated models rather than in-market sales.
Read the report (free PDF)See the data as a table
| Human-made ads (%) | AI-made ads (%) | |
|---|---|---|
| Eye-catching | 56 | 52 |
| Entertaining | 55 | 48 |
| Unique | 52 | 47 |
| Imaginative | 49 | 46 |
| People will talk about it | 40 | 37 |
Source: Ipsos and Syracuse University, AI Ads Are Good Enough (May 2026), n=3,000 US adults
Read the chart carefully. The gaps are three to seven points, not thirty. That is the uncomfortable finding: AI ads are not bad.
They are "good enough", and good enough is exactly the level that does not get remembered.
Ipsos also notes that ads which break category conventions are over 20% more likely to command brand attention. A model trained on existing ads is built to reproduce conventions, so it is weakest at the one thing that earns attention cheaply.
Why do consumers trust AI-made ads less?
Consumers read effort as a signal of quality, and AI removes the visible effort. When people learn or suspect an ad was generated, they infer the brand cared less, and they carry that inference to the product. Three independent sources show the same direction.
In a between-subjects online experiment with 220 participants, people saw the same ad for a novel sustainable food product, either with or without a note that it used generative AI. Disclosure lowered purchase intention, and the effect ran entirely through two beliefs: that the brand had put in less effort, and that the food would taste worse. The authors frame it with the labour illusion, the tendency to value things more when we see work behind them. The limit: one product category, one online sample, and stated rather than actual purchases.
Read the paper (free)NIM surveyed 1,000 people each in the US, the UK and Germany and ran two experiments. Only 20% trusted AI itself and 21% trusted AI companies; just 25% thought they could recognise AI-generated content. When an identical product image was labelled as AI-generated instead of a photo, it was rated more negatively, especially on emotional measures, and fewer people wanted to click to learn more. In a second test in Germany, six identical ads labelled "AI-generated" were seen as less natural and less useful than the same ads labelled "human-generated". The limit: NIM does not report effect sizes or per-group samples for the experiments, and calls the effect small but meaningful.
Read the article (free)What the surveys say
The survey data points the same way. Gartner's March 2026 consumer survey found 50% of US consumers would rather do business with brands that do not use GenAI in consumer-facing content and advertising, with 68% saying they frequently wonder whether the content they see is real. A Harris Poll reported by Marketing Brew in June 2026 put the numbers higher on sentiment: 73% said they would be less likely to trust an ad they suspected was made with AI.
See the data as a table
| Share of respondents (%) | |
|---|---|
| Say AI makes ads feel less authentic (Harris) | 78 |
| Less likely to trust an ad they suspect is AI-made (Harris) | 73 |
| Frequently wonder if content is real (Gartner) | 68 |
| Less likely to buy from brands using AI ads (Harris) | 63 |
| Prefer brands that avoid GenAI content (Gartner) | 50 |
Source: Harris Poll via Marketing Brew (Jun 2026); Gartner via DestinationCRM (Mar–Apr 2026) · Stated attitudes from two different US surveys with different question wording; not directly comparable.
Stated attitudes overstate behaviour. People say they will punish AI ads, then scroll past and buy anyway. That is why the Ipsos effectiveness numbers matter more than the attitude numbers: they measure what the ads do, not what people say about AI.
What did the Coca-Cola AI holiday ads actually show?
The Coca-Cola case shows that backlash and effectiveness can point in different directions, which is why it is often misread. Coca-Cola released AI-generated versions of its "Holidays Are Coming" film in 2024 and again in November 2025. The 2025 version was made by the AI studios Silverside and Secret Level; reporting by Implicator put Silverside's team at five AI specialists, against a typical crew of 50 or more, with more than 70,000 clips generated and about a month of production. About 100 people worked across Coca-Cola, WPP and the two studios, and the campaign ran in 140 countries.

Backlash online, decent scores in testing
Online, the reaction was harsh both years. Critics pointed to glitches such as truck wheels that seemed to glide, and the animator Alex Hirsch attacked the 2024 film over its effect on artists' jobs. Yet the effectiveness firm System1 found the 2024 AI ads tested well with general consumers. Both things can be true. The loud minority who notice and object are often journalists, creators and industry people with large followings, which is the group Ipsos calls the "informed minority with the reach to make it matter".
| Evidence | Sample | What it found | What it does not show |
|---|---|---|---|
| Ipsos and Syracuse, 2026 | 3,000 US adults, 20 ads | Human ads 14% stronger short term, 17% long term | In-market sales; AI work with human editing |
| Maurer and Buchbauer, EMAC 2025 | 220 participants | AI disclosure cut purchase intent via perceived effort and taste | Categories other than food |
| NIM, Transparency Without Trust | 3,000 survey; two experiments | AI labels lower ratings and click intent | Effect sizes |
| Gartner, March 2026 | US consumers | 50% prefer brands avoiding GenAI content | Actual switching behaviour |
| Coca-Cola holiday films, 2024–25 | Public reaction; System1 testing | Loud online backlash, decent mainstream test scores | Long-term brand effect |
Does it matter in India?
It matters more each quarter, because India now has a formal labelling regime. The Advertising Standards Council of India issued draft guidelines on AI-generated advertising content in May 2026 and final guidelines in September 2026, effective three months after publication. They sort uses into three tiers:
- Banned whatever the label: deepfakes, fake endorsements and using someone's likeness without consent.
- Label required: synthetic influencers, fabricated settings and demonstrations of products that do not yet exist, with a label such as "Video created using AI".
- No label needed: colour correction, minor retouching, obviously fantastical effects and AI-written ad copy.
The research above suggests the label itself will cost some persuasion, which creates a temptation to stay just inside the no-label tier. That is the wrong lesson. If the work only performs while people don't know how it was made, the problem is the work. For how Indian brands have used AI well, read my story on Cadbury's "Not just a Cadbury ad", where generative AI was the point of the idea rather than a cheaper way to fill a slot.
Why does cheap creative raise media costs?
Cheap creative raises media costs because forgettable ads need more impressions to do the same work. An ad's job is to build or refresh a memory structure that makes the brand come to mind when the buyer is ready. If a creative is 14% weaker at short-term response and 17% weaker at building the brand, the media plan has to buy the difference back in frequency, reach or discount. Media is usually the bigger line in the budget, so a saving in production can become a larger loss in media efficiency.
Why more variants do not fix it
Volume makes it worse. The usual AI pitch is that you can now test 50 variants instead of five. But variants made from the same reverse-engineered brief tend to share the same conventions, so you are testing small differences in a narrow band. On Meta, where Andromeda-era delivery rewards genuinely different creative concepts, near-duplicates can also compete with each other for the same auction. My guide to creative fatigue on Meta covers how often to refresh, and the honest answer is that concept diversity matters more than variant count.
There is a quieter cost too. In MIT economists Noy and Zhang's 2023 ChatGPT writing experiment, 68% of treated participants said they submitted the model's first output without editing it. That is a reasonable habit for a memo. It is a costly habit for an ad, because the first output is, by construction, the most probable one.
What does human direction actually add to AI-made ads?
Human direction adds the decisions a model cannot make for you: which insight is true for this audience, what the brand should feel like, and which convention to break. The machine can build what you specify very fast. It cannot tell you what is worth specifying.
- Start from a buyer insight, not a brief summary. Reverse-engineering a brief from past ads, as the Ipsos AI arm did, guarantees an average ad. Write down what the buyer believes, fears or wants that your category ignores.
- Decide the distinctive assets before generating anything. Colours, a character, a sound, a line. AI output drifts toward the generic unless these are fixed inputs.
- Use AI where effort is invisible. Storyboards, animatics, cut-downs, resizing, translation and versioning save time without changing what the viewer sees as craft.
- Keep humans on faces, hands, food and product. These are where viewers notice "uncanny" errors and where the effort signal is strongest.
- Test effect, not production speed. Judge each concept on attention, brand linkage and response against your human-made control, with the same media weight.
This is how I think about it in my own work, which I describe as human-directed and AI-built; there is more on that on my work page. The production pipeline itself, from prompt to finished cut, is in my explainer on how AI video ads are made, and the weekly operating rhythm is in the AI ad creative pipeline guide. Whether one person can run all of this is a separate question, which I take up in the one-person marketing team.
What's new
- ASCI issues final guidelines for labelling AI-generated content in advertising, effective three months after publication MediaNews4U, Sep 2026
- Harris Poll at Cannes Lions finds AI fatigue and lower trust in AI-made ads Marketing Brew, Jun 2026
- Ipsos and Syracuse test 10 human ads against fully AI-made versions Ipsos, May 2026
- Gartner: half of US consumers prefer brands that avoid GenAI in their content DestinationCRM, Apr 2026
Questions people ask
Do AI-generated ads perform worse than human-made ads?
On current evidence, somewhat. The largest controlled comparison so far, Ipsos and Syracuse in 2026, found fully AI-made ads 14% weaker on short-term effect and 17% weaker on long-term brand effect than human-made ads on the same brief. AI-assisted work with human direction was not tested and may close much of the gap.
Should brands disclose AI use in ads?
In India, ASCI's 2026 guidelines require a label when synthetic content could mislead, such as synthetic influencers or demos of products that do not yet exist. Research shows labels slightly reduce persuasion, but undisclosed AI that is later exposed carries a larger trust risk, and 79% of US respondents told Ipsos companies should have to disclose AI use.
How do you market in the age of AI slop?
Compete on what is scarce: specific insight, distinctive brand assets, real people and proof, and presence in the communities where your buyers already ask questions. Use AI to make the production of good ideas cheaper, not to replace the ideas.
Is AI-generated content bad for brand trust?
It can be when people notice it or are told about it. Experiments by NIM and by Maurer and Buchbauer found AI labels lowered ratings and purchase intent, mostly because people infer the brand put in less effort. Unlabelled AI that viewers cannot detect shows a smaller but still measurable effectiveness gap.
Sources
- Ipsos and Syracuse University: AI Ads Are Good Enough, and That's the Problem (2026)
- Maurer and Buchbauer: The Taste of Effort, Proceedings of the European Marketing Academy (2025)
- NIM: Transparency Without Trust (NIM Insights Vol. 7)
- Gartner: Marketing survey finds 50% of consumers prefer brands that avoid using GenAI in consumer-facing content (2026)
- DestinationCRM: Consumers prefer brands that avoid using GenAI in content (2026)
- Marketing Brew: Harris Poll on AI fatigue and trust in AI-generated ads (2026)
- HuffPost UK: Why Coca-Cola is facing a backlash over its AI-generated "Holidays Are Coming" ad (Nov 2024, source for the story)
- Implicator: Coca-Cola debuts AI holiday ads after 2024 backlash (2025)
- MediaNews4U: ASCI issues guidelines for labelling AI-generated content in advertising (2026)
- Noy and Zhang: Experimental Evidence on the Productivity Effects of Generative Artificial Intelligence (2023)
- Hacker News: Ask HN: How do you do marketing in the age of slop? (2026)
