AI Content Tools and the Science of Persuasion

Jasper, Copy.ai, and their competitors are remarkably good at peripheral route persuasion. That may be precisely the problem.

Anika van der Berg · March 15, 2025

Over the past two years, AI content generation tools have become standard equipment in marketing departments. Jasper (formerly Jarvis) claims over 100,000 paying customers. Copy.ai reports similar adoption. Writer, Anyword, Writesonic, and a growing roster of competitors are building substantial businesses around the premise that AI can generate marketing copy that is as persuasive as—or more persuasive than—human-written copy.

But "persuasive" is not a unitary concept. Persuasion researchers have known since the early 1980s that there are fundamentally different routes to attitude change, and that these routes produce qualitatively different outcomes. The question is not simply whether AI-generated copy persuades, but how it persuades—and whether the type of persuasion it produces is the type that marketers should want.

The Elaboration Likelihood Model: Two Routes to Persuasion

Petty and Cacioppo's Elaboration Likelihood Model (ELM), introduced in 1986, remains the most influential framework for understanding persuasion processes.¹ The model distinguishes between two routes to attitude change:

The central route operates when the recipient is motivated and able to think carefully about the message content. Persuasion occurs through careful evaluation of the quality of the arguments presented. Attitudes changed via the central route tend to be strong, persistent, resistant to counter-persuasion, and predictive of behavior.

The peripheral route operates when the recipient lacks motivation or ability to process the message carefully. Persuasion occurs through heuristic cues: source attractiveness, number of arguments (regardless of quality), emotional tone, social proof signals, and stylistic fluency. Attitudes changed via the peripheral route tend to be weak, temporary, susceptible to counter-persuasion, and poorly predictive of behavior.

This distinction is not merely academic. It has direct, measurable consequences for marketing outcomes. A consumer persuaded via the central route—who has carefully considered the product's genuine merits—is more likely to follow through on a purchase, less likely to experience buyer's remorse, more likely to become a repeat customer, and more likely to recommend the product. A consumer persuaded via the peripheral route—who was swayed by the emotional resonance of the copy, the social proof signals, or the sheer fluency of the writing—is more likely to abandon the purchase funnel, more likely to return the product, and less likely to develop brand loyalty.

What AI Content Tools Optimize For

AI content tools are, almost without exception, optimized for peripheral route persuasion. This is not a deliberate design choice but an emergent property of how they are trained and evaluated.

Large language models generate text by predicting the most probable next token given the preceding context. When fine-tuned on high-performing marketing copy, they learn the statistical patterns of text that has historically produced engagement: emotional language, active voice, short sentences, power words, social proof phrases, urgency cues, and stylistic fluency. These are precisely the peripheral cues that the ELM identifies as shortcuts to persuasion in low-elaboration conditions.

Consider what Jasper produces when asked to write a product description for, say, a project management tool. The output typically includes: emotional benefit language ("reclaim your time," "work without chaos"), social proof signals ("trusted by thousands of teams"), urgency framing ("start your free trial today"), and high stylistic fluency (clear, rhythmic prose with varied sentence structure). What it rarely produces is: specific evidence of efficacy, detailed comparative analysis against alternatives, honest discussion of limitations, or technical argumentation that would support central route processing.

I ran an informal experiment last year, generating 50 product descriptions using Jasper and Copy.ai across five product categories (SaaS tools, consumer electronics, health supplements, financial services, and educational courses). I then coded each output for the presence of central route elements (specific claims, evidence, logical arguments, comparative data) and peripheral route elements (emotional language, social proof cues, urgency signals, source cues, stylistic fluency).

The results were striking, though I offer them with the caveat that this was an informal analysis, not a controlled study. Peripheral route elements were present in 100% of outputs. Central route elements were present in fewer than 15%. The AI tools consistently produced copy that was emotionally engaging, stylistically polished, and substantively empty.

The Elaboration Likelihood Mismatch

The ELM predicts that peripheral route persuasion is most effective when the audience is in a low-elaboration state—when they are not motivated or able to think carefully about the message. This describes some marketing contexts well: impulse purchases, low-involvement product categories, casual social media browsing.

But it does not describe all marketing contexts, and the mismatch between AI-generated copy and audience elaboration state can be counterproductive. When a prospective buyer is in a high-elaboration state—actively researching a significant purchase, comparing alternatives, reading reviews—peripheral cues are not merely ineffective; they can be actively harmful.

Petty, Cacioppo, and Goldman (1981, n=145) demonstrated that when elaboration likelihood is high, strong arguments produce more persuasion than weak arguments, regardless of peripheral cues. But when elaboration likelihood is low, the strength of arguments has little effect, and peripheral cues dominate.² The implication is that AI-generated copy, rich in peripheral cues but poor in argument quality, will underperform with precisely the audience that matters most: informed, motivated buyers conducting deliberate evaluation.

This is particularly relevant for B2B marketing, where purchase decisions typically involve multiple stakeholders, extended evaluation periods, and high financial stakes—all conditions that promote high elaboration. A CTO evaluating a cloud infrastructure provider is in a high-elaboration state. The emotionally resonant, peripherally persuasive copy that Jasper generates is not just unhelpful for this audience; it signals a lack of substance that may trigger skepticism and distrust.

The Fluency Trap

One specific peripheral cue that AI content tools produce in abundance is processing fluency—the subjective ease with which information is processed. Research by Reber, Schwarz, and Winkielman (2004) established that high processing fluency produces positive affect, which is then attributed to the content of the message (a "how-do-I-feel-about-it" heuristic).³

AI-generated text is almost invariably fluent. Large language models are trained to produce coherent, grammatically correct, smooth-flowing text. This fluency creates a positive affective response that can be mistaken for substantive quality. A marketing team reviewing AI-generated copy may think "this is good" when what they are actually experiencing is "this is easy to read."

The danger is that fluency becomes a substitute for substance. If the copy reads well, the team approves it without scrutinizing the claims, the logic, or the specificity. The copy goes live, performs adequately on engagement metrics (because fluency drives engagement in low-elaboration contexts), and appears to validate the AI tool's value. But the engagement is shallow. Click-through rates may be healthy while conversion rates downstream are anemic, because the copy attracted attention (peripheral route) without building conviction (central route).

The Homogenization Problem

A related concern is content homogenization. Because AI tools are trained on the same or similar datasets and optimize for the same statistical patterns, they tend to produce copy that converges on a limited range of styles and structures. The "AI voice"—confident, slightly breathless, heavy on benefits, light on specifics—is becoming recognizable to experienced readers.

From an ELM perspective, homogenization undermines the distinctiveness that is necessary for both central and peripheral route persuasion. On the central route, distinctiveness signals original thinking and genuine product differentiation. On the peripheral route, distinctiveness captures attention and increases processing depth. When all competitors' copy sounds the same because it was generated by similar AI tools, neither route operates effectively.

There is also a longer-term risk. As consumers develop familiarity with AI-generated content—its cadences, its patterns, its characteristic vagueness dressed up as confidence—they may develop a general skepticism toward it. This would function as a persuasion knowledge structure (Friestad & Wright, 1994), allowing consumers to recognize and discount AI-generated persuasion attempts. The peripheral cues that currently work because they are processed automatically may become ineffective as consumers learn to identify them as markers of automated, low-effort content.

The Argument Quality Gap

Perhaps the most consequential limitation of current AI content tools is their inability to generate genuinely strong arguments. In the ELM framework, argument quality is the primary driver of persuasion via the central route. Strong arguments are those that are specific, evidence-based, logically structured, and difficult to counter-argue.

AI tools can produce arguments that look strong—they are grammatically polished and use confident language. But they systematically lack the specificity and verifiability that characterize genuinely strong arguments. "Our platform helps teams collaborate more effectively" is a weak argument dressed in confident language. "Teams using our platform resolved 34% more support tickets per sprint in a controlled comparison with three alternative platforms" is a strong argument. AI tools reliably produce the former and rarely produce the latter, because the latter requires access to proprietary data, genuine analysis, and the ability to make claims that are specifically true of one product and not others.

This gap matters most in competitive markets where consumers are evaluating multiple alternatives. When every competitor's AI-generated copy claims to be "the most powerful," "the most intuitive," and "trusted by industry leaders," these claims cancel each other out. The brand that breaks through will be the one that makes specific, verifiable, evidence-based claims—the kind that AI tools are not yet equipped to generate.

Where AI Content Tools Add Genuine Value

I want to be clear that this analysis is not a blanket dismissal. AI content tools add genuine value in specific contexts. They are excellent at generating high volumes of short-form copy where peripheral route persuasion is appropriate: social media posts, ad variations for testing, email subject lines, meta descriptions. They are useful for overcoming blank-page paralysis, generating structural ideas, and producing first drafts that a skilled writer can then enrich with substance.

The problem arises when they are used as a complete replacement for human thinking about what the message should substantively communicate—what arguments should be made, what evidence should be presented, what the audience genuinely needs to know to make a good decision. This substantive thinking is the domain of the central route, and it is precisely what current AI tools do not provide.