Why Your AI Marketing Keeps Falling Flat (It's a Psychology Problem)
A Harvard and BCG study gave 758 consultants a set of realistic business tasks, including writing marketing slogans, segmenting markets, and designing creative campaigns. Consultants using GPT-4 completed tasks 25% faster and scored 40% higher on quality ratings than those working without AI.
That sounds like the case is closed. AI makes marketing better. But the same study revealed something the headlines mostly ignored.
Wharton professor Ethan Mollick, one of the study's authors, described the results as a "jagged frontier" of AI capability. On some tasks, AI was extraordinary. On others, consultants who used AI performed 19% worse than those who worked alone. The AI didn't just fail to help on those tasks. It actively dragged performance backwards, because people trusted output that sounded confident but missed the point entirely.
The tasks AI struggled with shared a common thread: they required reading between the lines. Sensing what wasn't being said. Understanding the emotional and contextual layers beneath the surface logic. In other words, exactly the kind of thinking that makes marketing actually persuade someone.
The Most Expensive Gap in Marketing
If AI defaults to logical, feature-driven output, does that actually matter? The Institute of Practitioners in Advertising (IPA) has been tracking what makes advertising effective since the 1980s. Their database now covers over 1,200 campaigns across every category. Les Binet and Peter Field's analysis of that data produced one of the most robust findings in marketing science.
Emotional campaigns are nearly twice as profitable as rational ones. 46% of emotionally-driven campaigns reported very large business effects, compared with just 28% of campaigns built on rational persuasion. Over longer timeframes, the gap widens further. Rational campaigns produce short-term spikes that decay. Emotional campaigns build cumulative brand effects that compound.Their recommended budget split, now widely adopted across the industry, is 60% brand (emotion-led, broad reach) and 40% activation (direct response, performance). Most businesses using AI for marketing are unknowingly producing 100% activation content. Feature lists. Logical arguments. Rational calls to action. The kind of marketing that Binet and Field's data says delivers roughly half the business impact it could.
This isn't a small optimisation opportunity. It's the difference between marketing that works and marketing that looks like it should work but quietly underperforms for years.
Why AI Defaults to the Wrong Mode
Rory Sutherland, Vice Chairman of Ogilvy, has a line worth remembering: "The human mind does not run on logic any more than a horse runs on petrol."
His central argument is that most business problems people try to solve with engineering are actually problems of perception, feeling, and meaning. His Eurostar example is the clearest illustration. Engineers spent £6 billion building new tracks to cut 40 minutes off the London-to-Paris journey. Sutherland pointed out they could have spent a fraction of that making the journey enjoyable, WiFi, better food, comfortable seating, and passengers would have been happier with a longer trip.
AI has the same blind spot as those engineers. It solves for the explicit, measurable, logical version of the problem. Ask it to write a Google Ad for a scaffolding company and you'll get: "Reliable scaffolding services. 20+ years experience. Free quotes. Licensed and insured." Every word is true. Every word is identical to what every competitor is also saying. And every word speaks to the rational part of the brain that, for most buying decisions, isn't actually in charge.
The scaffolding buyer isn't sitting down to compare five options on a spreadsheet. They're calling someone they remember or trust. The decision was already made before the conscious evaluation began, driven by what Sutherland calls "psycho-logic": the irrational but entirely human reasoning that governs real choices.
AI doesn't generate psycho-logic. It generates logic. And logic, as Binet and Field's data proves, is the less profitable half of the equation.
The Trust Problem Beneath the Quality Problem
The gap goes deeper than just tone. Recent benchmark data shows that when consumers identify content as AI-generated, purchase intent drops 14% and premium brand perception falls 17%. A Raptive study of 3,000 adults found that suspected AI content reduces reader trust by nearly 50%.
This creates a compounding problem. AI-generated marketing defaults to rational, feature-driven copy. Rational copy already underperforms emotional copy by roughly half, according to IPA data. And when that rational copy also reads as AI-generated, the trust penalty compounds the performance gap further.
Online mentions of "AI slop" grew over 200% in 2025. Merriam-Webster named "slop" its word of the year. Consumers are developing pattern recognition for generic AI output faster than most marketers realise. The businesses still pasting briefs into ChatGPT and publishing whatever comes back are training their audiences to tune them out.
What the Jagged Frontier Actually Looks Like in Marketing
Mollick's jagged frontier concept is useful because it's specific. AI doesn't uniformly succeed or fail. It excels at some marketing tasks and actively harms others. Here's where the frontier falls for a typical small business:
| AI excels here | AI struggles here |
|---|---|
| Generating headline variations at volume | Choosing which variation will resonate emotionally |
| Structuring landing page layouts | Writing copy that makes someone feel something |
| Keyword research and clustering | Understanding what a buyer fears in the moment they search |
| Producing technically correct ad copy | Creating distinctive brand voice that doesn't sound like everyone else |
| A/B test variant generation | Identifying the psycho-logic behind a purchase decision |
| Data analysis and reporting | Translating data insights into persuasive creative |
The pattern: AI handles structure, logic, and volume brilliantly. It struggles with emotion, distinctiveness, and the kind of buyer empathy that separates growing businesses from busy ones. The businesses getting the most from AI aren't using it for everything. They're using it on the right side of the frontier and bringing human psychology to the left.
How to Fix It: Three Shifts
1. Brief for emotion, not features. Instead of "write a Google Ad for scaffolding hire," try: "Our buyer just discovered their scaffolding hasn't arrived for a job starting tomorrow. They're stressed. They want to feel certain this won't happen again. Write an ad that immediately signals reliability and takes away the anxiety of a missed deadline." That shift in framing moves AI output from generic feature lists to something that connects with what Binet and Field's research says actually works. 2. Encode psycho-logic into your process. Sutherland's framework asks: what does the buyer fear? What would make them feel foolish? What status does this purchase represent? Build a one-page buyer psychology brief that answers these questions with specifics from real customer conversations, and make it a standard input for every AI content task. This is the single highest-leverage change you can make to your AI marketing workflow. The brief takes a few hours to build once. The improvement compounds across every piece of content you produce. 3. Use AI for the right half of the frontier. Let AI do what it does well: generate variations, structure content, analyse data, handle production volume. But keep humans in charge of the strategic and emotional layer. What Mollick's research calls the "Centaur" approach, a clear division where each does what they're best at, consistently outperforms both pure AI and pure human work. Don't use AI to replace the psychology. Use it to amplify the psychology you've already defined.What This Means for Your Business
64% of marketers now use AI in their content workflows, up from 35% in 2024. Most of them are producing more content, faster. Almost none of them have solved the psychology problem. They're generating rational marketing at unprecedented scale, which Binet and Field's data says is roughly half as effective as it could be.That's not a threat. It's an opportunity. While competitors flood channels with AI-optimised content targeting the 5% of buyers actively searching, the businesses that encode genuine buyer psychology into their AI workflows will produce marketing that actually moves the other 95%.
The leverage isn't in the tool. It's in what you feed it. Give AI logic, and you get logical marketing. Give it the emotional truth of why your customers actually buy, and you get something competitors using AI at "Level 1" simply cannot replicate.
AI amplifies whatever strategic insight you give it. Right now, most businesses are amplifying nothing.
Further Reading
- Navigating the Jagged Technological Frontier - The Harvard/BCG study on AI's uneven impact on knowledge worker performance, by Dell'Acqua, Mollick, et al.
- The Long and the Short of It - Binet and Field's landmark analysis of 1,200+ IPA campaigns proving the effectiveness gap between emotional and rational advertising
- Alchemy by Rory Sutherland - The case for psycho-logic over conventional logic in marketing
- AI Ad Creative Benchmarks 2026 - Current performance data comparing AI-generated vs human creative across platforms
- The Authenticity Premium - Why consumers are increasingly rejecting AI-generated content and what it means for brands
Dream Outcome is an Australian digital marketing agency helping SMEs grow through Google Ads, Facebook Ads, and Email Marketing.