You're Helping Your Marketing AI. The Research Says You're Making It Worse.

A meta-analysis published in Nature Human Behaviour analysed 106 studies covering 370 separate experiments. The question: does combining human judgment with AI produce better results than either one alone?

The answer was supposed to be obvious. Humans bring creativity and context. AI brings speed and data processing. Together, they should be unstoppable.

They're not. Human-AI teams performed significantly worse than the best of either humans or AI working alone. When AI outperformed humans on a task, adding human oversight actively degraded the result.

Eric Topol, one of the most cited medical researchers alive, watched this play out in clinical trials. His response: "That isn't the way it was supposed to work. It was supposed to be that the combined hybrid performance was going to be the best."

The same dynamic is playing out in your marketing right now. Every time you override Google's Smart Bidding because the CPC "feels too high." Every time you restrict Meta's Advantage+ targeting because you "know your audience better." Every time you rewrite AI-generated ad copy because it doesn't sound like what you'd write.

You're not helping. You're probably making it worse.

Two Psychological Traps (You're in One of Them)

Daniel Kahneman's research on cognitive biases explains why human-AI collaboration fails so consistently. Two opposing traps catch almost everyone.

Here's what makes this devastating: you're falling into one trap or the other, and you probably don't know which one.

The business owner who sets up Smart Bidding and never looks at the account again? Automation bias. The business owner who manually adjusts bids every day because last Tuesday's CPC spiked? Algorithm aversion. Both are making their campaigns worse.

Where This Hits Your Campaigns

The performance data on AI marketing tools is now overwhelming, and it mostly says the same thing: when advertisers let the AI run, results improve. When they fight it, results get worse.

Google's data shows that broad match paired with Smart Bidding delivers 20-30% more conversions at similar cost-per-acquisition compared to exact match with manual bidding. Their AI Max product has become their "fastest-growing AI search product," and from September 2026, they're auto-upgrading campaigns into it. Eighty-six percent of Google Ads campaigns already run automated bidding. Google has picked a side.

Meta's numbers tell the same story. Advantage+ shopping campaigns achieve 32% lower CPA than manually structured campaigns. Since March 2025, Meta removed detailed targeting exclusions entirely. They're not giving you the option to fight the algorithm anymore.

But here's where it gets interesting.

The BCG-Harvard "Jagged Frontier" study tested 758 consultants on real business tasks. On tasks within AI's capability frontier, quality improved 40% and speed improved 25%. On tasks just outside that frontier (tasks that looked similar but required different capabilities), consultants using AI performed 19 percentage points worse than those working without it.

The frontier is jagged. Some tasks that look straightforward for AI are actually terrible for it. Some tasks that look too complex are where it excels. You can't tell the difference by looking. And this is precisely why the problem was never your ads: the mechanical parts of campaign management sit inside AI's frontier. The strategic parts don't.

What AI Gets Right (Better Than You)

Let's be direct. There are entire categories of marketing decisions where your instincts are actively worse than the algorithm's data.

Byron Sharp's research across 130+ brands and 13 product categories reveals a consistent pattern: marketers systematically overestimate how differentiated their brand is, how loyal their customers are, and how well they understand their audience. The data shows most brands in a category have near-identical buyer profiles. Most loyalty is a function of market share, not brand love. Most "targeting insights" are confirmation bias with a dashboard.

AI doesn't carry these blind spots. When Smart Bidding adjusts your cost-per-click at 2am because conversion patterns shifted, it's responding to signals you'll never see. When Meta expands your audience to people you'd never have targeted, it's finding buyers you didn't know existed. When Google's AI Max generates headline variations you wouldn't have written, it's testing language that reflects how people actually search.

A Deloitte study found that 60% of executives use AI in decision-making, but only 5% manage it well. The remaining 55% are stuck in the worst of both worlds: AI has some authority, but humans keep pulling it back. The algorithm never accumulates enough data to learn, and the human never fully commits to either approach.

The Wharton algorithm aversion research uncovered a useful detail here. People were considerably more likely to trust an algorithm when they could make even small modifications to its output. The modification didn't need to improve the result. It just needed to exist. Google and Meta have clearly read this research. Every "manual adjustment" they offer (seasonality bids, audience signals, asset pinning) exists partly to make you feel in control. Whether those adjustments actually help is a different question.

What AI Gets Wrong (And Why You Still Matter)

Everything above might suggest the answer is simple: just let the AI run everything. That's automation bias talking, and it's just as dangerous.

Rory Sutherland has spent decades proving that human behaviour runs on psychology, not logic. AI, by design, runs on logic. It finds patterns in data and optimises toward them. It cannot generate the insight that renaming "Patagonian Toothfish" to "Chilean Sea Bass" transforms an unsellable product into a restaurant hit. It cannot invent the fly etching in urinals that reduced spillage by 80%. It defaults to what the data says should work, which is exactly what makes your ads sound like everyone else's.

The things AI can't do in marketing are precisely the things that create breakthroughs:

The Override Framework

So when do you trust the machine, and when do you step in?

Sam Tomlinson captured it precisely: "Belief can't be outsourced to AI." The brands that outperform in the coming decade will let AI handle the mechanics while humans handle the meaning.

What This Means for Your Business

The businesses getting the best results from AI marketing aren't the ones using the most tools or automating the most tasks. They're the ones who've figured out where the jagged frontier runs through their specific marketing.

The research is clear. Your marketing AI doesn't need your help with the parts you keep helping with. And the parts where it genuinely needs you? Those are the parts most businesses ignore entirely.

Further Reading

• Vaccaro et al. - Human-AI combinations perform worse than the best of either alone (Nature Human Behaviour, 2024) - The meta-analysis of 106 studies that started this conversation
• Dietvorst, Simmons & Massey - Overcoming Algorithm Aversion (Management Science, 2018) - Foundational research on why we fight algorithms even when they outperform us
• Harvard/BCG - Navigating the Jagged Technological Frontier - 758 consultants tested on when AI helps and when it hurts
• Binet & Davis - Go Big or Go Home (IPA, 2025) - Why budget level matters 8x more than ROI optimisation
• Avinash Kaushik - Bye, Bye Human-Powered Marketing Analytics - Why it's time to hand analytics to AI