Your AI Writes Perfect Marketing. That's Why Nobody Remembers It.

There is a word that barely existed in professional writing three years ago. The word is "delve." In 2023, it appeared in roughly 0.04% of biomedical research abstracts. By 2024, after ChatGPT became the default writing assistant for millions of professionals, its usage had increased 28-fold.

"Delve" is not the problem. The problem is what it represents. When millions of businesses feed their marketing through the same AI models, trained on the same data, optimising for the same metrics, something predictable happens. The output converges. Every headline starts sounding like every other headline. Every email opens with the same cadence. Every ad description hits the same notes of confident enthusiasm.

Your AI-generated marketing is grammatically perfect, strategically sound, and SEO-formatted. Your audience scrolls past it without a flicker of recognition. Not because it's bad. Because it's indistinguishable from the business next door.

The Most Expensive Assembly Line Ever Built

A 2025 study published in PNAS by researchers at Carnegie Mellon University measured the stylistic fingerprints of major language models against human writing. The finding was blunt: LLMs have a systematic bias toward information-dense, academic-style prose, and this bias persists regardless of the prompt, the genre, or the model size. You can ask for casual. You get formal with a casual hat on.

Separate research tracking 14 million PubMed abstracts identified a cluster of "excess vocabulary" words whose usage spiked the moment LLMs entered widespread use. Not technical terms. Style words. "Comprehensive." "Crucial." "Notably." "Insights." "Showcasing." Words that sound like they mean something but say almost nothing.

This matters because the same models writing those abstracts are writing your Google Ads descriptions, your email sequences, your website copy, and your social media posts. And they're doing the same thing for every one of your competitors.

The numbers tell the story. 74.2% of new web pages published in Q1 2026 contain some AI-generated text, according to content analysis by TheStacc. Nearly half of all new articles are now fully AI-written. The internet is being flooded with content that is competent, correct, and converging toward the same statistical middle.

Azeem Azhar, writing in his Exponential View newsletter, put it precisely: he ran natural language processing tools across hundreds of thousands of words of his own writing and discovered he uses 80% Germanic root words. Short. Direct. Anglo-Saxon. The average LLM output? Around 60% Latinate vocabulary. Longer. More abstract. More formal. "Utilise" instead of "use." "Demonstrate" instead of "show." "Commence" instead of "begin."

That's not a style preference. It's a measurable drift toward corporate committee-speak. And it's happening to every business that lets AI write their first draft without intervention.

Rory Sutherland Predicted This 10 Years Ago

Before AI wrote a single ad, Rory Sutherland identified the trap. His observation from Alchemy remains the sharpest diagnosis of what's happening right now:

AI is the most logical marketing tool ever built. It literally computes the statistically most probable next word. It optimises for patterns that have worked before. It benchmarks against existing content and converges toward what already exists.

Sutherland recently expanded this argument for 2026 specifically: "AI is a prediction engine that pulls everything back to the middle. Leaning on it too heavily puts you on the same field as every competitor." He calls benchmarking "a way of making you more similar to your immediate competitors." AI-generated content is benchmarking at scale, applied to every word you publish.

This connects directly to what we've written about before: the more systematically you follow the same playbook as everyone else, the more invisible you become. AI didn't create this problem. It industrialised it.

Distinctive Means Different, Not Better

Byron Sharp's research at the Ehrenberg-Bass Institute has been saying the same thing from a different angle for two decades. Brands grow by being easy to think of and easy to buy. Not by being objectively better. Not by having superior features. By being distinctive enough that they come to mind in the moment someone needs what you sell.

Sharp calls this mental availability. It's not the same as awareness. Awareness is binary: have you heard of us? Mental availability is whether your brand springs to mind across the full range of real-world situations that trigger a purchase. And recent research suggests that if your brand comes to mind in a buying situation, roughly 70 to 80 percent of the job is already done.

His colleague Jenni Romaniuk built the measurement framework. Her Distinctive Asset Grid plots brand elements on two axes: Fame (how many people associate the asset with your brand) and Uniqueness (how exclusively it's linked to you, not your competitors). An asset that's famous but not unique triggers thoughts of the whole category. An asset that's unique but not famous helps nobody.

Here's the table that should worry every business using AI for content:

Sharp's line at Cannes 2026, reported by Mi-3, was characteristically direct: great branding is simply a brand that looks like itself. Your colour. Your tone. Your rhythm. The thing people clock before they read a single word. AI doesn't know what "looking like yourself" means. It knows what "looking like everything" means.

The Numbers Behind the Trust Collapse

This isn't theoretical. Consumers are already reacting.

Online mentions of "slop," the term people use for low-effort AI filler, rose more than 200% in 2025. 82% of that conversation was negative.

And here's the kicker from Binet and Field's IPA effectiveness research. Their analysis of award-winning campaigns over three decades found that high-quality creative produces business effects up to 12 times more often than mediocre creative, for the same media budget. A 5-star campaign with $100K in media can outperform a 2-star campaign with $1M.

AI is producing 2-star creative at 5-star speed. We've explored the creative quality problem before in our piece on why everyone can make ads now but almost nobody knows which ones are good. The volume has exploded. The distinctiveness has collapsed.

The Fix Isn't Less AI. It's a Better System.

The answer is not to stop using AI. The answer is to stop using it as a replacement for thinking and start using it as an amplifier for what's already distinctive about your business.

Kieran Klaassen at Every coined a useful concept called compound engineering: the idea that each unit of work should make the next unit easier, not harder. Most AI marketing does the opposite. Each piece of content is generated from scratch, with no memory of what came before, no learning from what worked, no accumulation of distinctive voice.

A compound marketing system works differently. It treats every piece of content as an investment in a growing body of knowledge about what makes your business sound like your business, not like a committee.

What this looks like in practice:

What This Means for Your Business

The irony of AI in marketing is this: the tool that was supposed to give small businesses the output of big agencies has instead given every business the output of the same agency. A competent, forgettable, mid-range agency that writes in Latinate English and opens every paragraph with "in today's rapidly evolving landscape."

Your marketing doesn't need to be perfect. Binet and Field's data proves that polished but generic work gets crushed by distinctive work with a fraction of the budget. Sharp's research proves that brands don't grow by being better. They grow by being impossible to confuse with anyone else.

Use AI. Use it aggressively. Use it for research, for structure, for speed, for scale. But build a system that makes the output more distinctively yours over time, not less. The businesses that figure this out won't just survive the convergence. They'll profit from it, because every competitor drowning in AI sameness makes the distinctive voice stand out more.

Sutherland's rule remains the sharpest test: "If there were a logical answer, we would have found it." Your AI has found every logical answer. The valuable ones are hiding in the places it can't reach.

Further Reading

• Do LLMs write like humans? Variation in grammatical and rhetorical styles - PNAS study showing LLMs have systematic stylistic biases toward dense, academic prose regardless of prompt
• Delving into LLM-assisted writing through excess vocabulary - Research tracking the 28x spike in LLM-preferred words across 14 million abstracts
• The Slop Tax: AI Content and Brand Trust in 2026 - Data on the measurable trust penalty businesses pay for AI-generated content
• Marketing Creativity Crisis 2026: System1 & IPA Data - Binet and Field's evidence that creative quality drives 12x more business effects than budget
• Entering the Trillion-Agent Economy - Azeem Azhar's analysis of AI's vocabulary bias and the taste problem in LLM outputs