The thesis: the problem is not quality, it is uniformity
With generative AI, the average quality of texts and images produced for brand communication has risen, but campaigns and messages tend to look increasingly alike. This is the central thesis of the analysis published by Agenda Digitale, according to which the “AI slop” phenomenon — the automated, serial production of content — should not be read only as a problem of low quality, but as one of uniformity: when models, prompts and briefs resemble one another, content becomes interchangeable, and production savings end up consuming attention, recognizability and budget.
The author of the analysis — Piermario Tedeschi, founder and managing director of the agency Digital Angels, a lecturer at Luiss and a board member of IAB Italia — makes clear that he speaks not as a nostalgic but as someone who uses AI daily at an agency. His argument is that slop, in brand communication, has “a more polite face” than the junk that clogs social media: it is the well-written post, with clean visuals and a headline that complies with the brief, yet after a few minutes no one remembers whose it was. “Those who produce it, agencies included, are selling off their own professionalism,” he writes.
Why content comes out “in series”: models, briefs and approval committees
The mechanism described is simple: companies and agencies often query the same few models with similar prompts, starting from even more similar briefs, and those models by design tend toward the most probable responses, that is, the most expected and “normal” ones. The result is a mediocrity that passes every review, because marketing departments, approval committees, legal departments and pre-tests reward what resembles what has already worked.
In support, the article cites two references: the WFA and LIONS Clients & Creativity 2026 research, according to which 58% of senior marketers fear that AI will make creativity all the same, and IAB Spain's Observatory on Brands in Social Media, published in September, according to which in one year Spanish brands reduced the content they published by 30% while recording a 29% increase in engagement. The author warns that the figure alone does not prove a causal link between publishing less and gaining more engagement, but it indicates that those who cut almost a third of their posts benefited, contradicting the promise with which generative AI entered marketing departments: to produce more at the same cost.
The consequences: double cost, new KPIs and tools chosen based on the audience
The analysis extends the phenomenon to offices as well: Stanford and BetterUp Labs identified it in the workplace and Harvard Business Review called it “workslop,” for documents and presentations that look like finished work and offload the effort onto those who receive them. According to the cited data, 41% of the workers surveyed have received at least one such item, and remedying it takes an average of almost two hours. In agencies, the author writes, it can be recognized in extremely long briefs that do not address the substance of the requests, and sometimes it originates with the agencies themselves.
On the economic front, the thesis is that a confusable idea is paid for twice: in production and in planning, because a message that no one recognizes requires more steps to stay in mind, and the initial saving turns into impressions bought to chase a recollection that does not arrive. Hence the position on KPIs: cost per asset would be the metric that rewards slop, whereas recall and the effect on sales should be measured, valuing ideas as ideas separate from production.
The article does not call for abandoning AI. It cites two ads made for the Consiglio Nazionale degli Ingegneri [National Council of Engineers] and presented in September at the national congress: “Ingegno Italiano” (Italian Ingenuity), made with visible generative AI but a human concept, and “Domande” (Questions), shot as a traditional production, with the choice of tool dictated by the target audience. A paper presented at the 2025 SIM Conference on TikTok Smart+, co-authored with Maria Vernuccio and Mihaela Giannelli, indicated that it is advisable to leave mechanical work to the algorithm — budget, pacing, creative rotation — while deciding what to say to one's target remains a human choice. The cited data remain to be verified against the original sources, as does the actual applicability of the Spanish model to other markets.
