Can advertising mechanisms designed for humans work on agents?
This explores whether the persuasion and attention-capture tactics built for human audiences — emotional appeals, social proof, click-bait, attention competition — still move autonomous agents acting on a user's behalf, or whether agents need an entirely different incentive layer.
This explores whether ad mechanisms tuned for human attention transfer to agents, and the corpus suggests the honest answer is: the *goal* of advertising survives, but most of its *machinery* doesn't. The clearest signal is that the human attention economy is expected to give way to an agent attention economy — once users delegate goals to autonomous agents, services stop competing for clicks and start competing to be *selected* by an agent, which spawns agent-optimized discovery, ranking, and recommendation infrastructure that mirrors human ad ecosystems in shape but not in mechanism Will agents compete for attention just like users do?. So advertising-as-competition persists; advertising-as-emotional-capture is the part that breaks.
Why it breaks comes into focus when you look at how persuasion actually lands differently. Human advertising leans heavily on emotion and social proof, but LLMs persuade — and are persuaded — through logical appeals and quantitative framing in nearly every exchange, which makes their reasoning feel objective rather than manipulable in the human way Do LLMs persuade users more often than humans do?. A banner that works on a person by triggering FOMO or aspiration has little to grip onto in a system that weighs arguments. And the persuasion that *does* work on models is uneven: it varies by model family and even by whether the claim is true or false, so there's no single lever an advertiser can pull across all agents Do large language models persuade better than humans?.
The deeper reason is what agents optimize for. When people evaluate a dialogue partner, perceived *competence* dominates their impression far more than likability or human-likeness How do users mentally model dialogue agent partners? — and agents pushed one step further, selecting on hard functional signals. Phone agents, for instance, treat task success, privacy compliance, and reuse of a user's *saved preferences* as distinct capabilities, meaning an agent is acting on a stored profile of what its user actually wants, not on whatever message just grabbed its attention Do phone agents succeed at all three critical tasks equally?. That's a buyer with a shopping list, not a browser to be tempted.
But the surprising twist — the thing you might not expect — is that agents are not fully immune, because they partly *inherit* human response patterns. LLM personas reproduce roughly three-quarters of published human experimental effects, tracking the strength of the original evidence Can AI personas reliably replicate human experiment results?, and recommendation simulators deliberately build in emotion modules so their agents act on taste and mood, not just utility — enough to reproduce filter-bubble dynamics and fatigue Can LLM agents realistically simulate filter bubble effects in recommendations?. So the residue of human-style susceptibility is real. The likely outcome the corpus points to is a layered one: emotional and social-proof tactics fade, functional and logical signals dominate, and the contest moves to a new battleground — being chosen by ranking and discovery systems agents trust, which is exactly where the next round of optimization (and manipulation) will concentrate.
Sources 7 notes
Research shows that as users delegate goals to autonomous agents, services must compete for agent selection rather than clicks. This drives agent-optimized discovery mechanisms, ranking systems, and recommendation infrastructure mirroring human-facing ad ecosystems.
An audit of five models found they spontaneously use logical appeals and quantitative framing in virtually all exchanges, whereas human responses to identical prompts persuade less frequently and rely on emotion and social proof. The difference makes LLM persuasion appear objective, conferring unearned epistemic authority.
Claude beats incentivized humans at both truthful and deceptive persuasion, while DeepSeek only beats them when arguing for falsehoods. The persuasion mechanism appears content-independent, suggesting model family itself acts as a contextual moderator.
The Partner Modelling Questionnaire reveals that perceived competence dominates user impressions (49% of variance), followed by human-likeness (32%) and communicative flexibility (19%). This three-factor structure reflects how people evaluate dialogue partners against both functional and social standards.
MyPhoneBench demonstrates that task success, privacy-compliant completion, and saved-preference reuse are statistically distinct capabilities with no model dominating all three. Success-only rankings do not predict privacy or preference performance.
Viewpoints AI reproduced 84 of 111 main effects from Journal of Marketing experiments with replication success strongly correlated to original p-value strength. Marginal effects showed unreliable performance with both false positives and negatives.
Agent4Rec demonstrates that 1,000 LLM-empowered agents initialized from real behavioral datasets can model both taste-driven and emotion-driven actions, enabling researchers to study filter bubble effects and user fatigue through causal interventions without expensive human experiments.