More output, flatter results: model collapse, distillation, and a hollowed junior pipeline. Why this is a leadership problem, not a switch-off-AI debate, plus three checks you can run this week.
Published on 13 March 2026
You are producing more marketing than ever, but it has never mattered less. The calendar is full, the tools are faster, and the output looks polished. Engagement is flat, leads have gone quiet, and everything you publish sounds fine. Just fine.
No, you are not imagining it.
Mark Ritson flagged this in Adweek: Anthropic's research suggests roughly 65% of marketing tasks are now AI-exposable, with marketing ranking fifth out of 800 occupations for that exposure. This is not a distant future problem. It is already reshaping the function I have worked in for more than twenty years from the inside.
The real danger is not only job displacement. It is what is happening to marketing output globally when the same narrow patterns get copied, distilled, and scaled.
Most commentary either cheers adoption or warns about redundancy. Less often discussed is the effectiveness crisis: the function disappearing into AI-generated sameness unless someone senior protects positioning, voice, and editorial judgement while still using the tools. This is not a lecture on switching AI off. It is a leadership design problem: who in your organisation keeps the 20% that differentiates from dissolving into the 80% that now costs almost nothing to produce.
This article names the photocopy problem, ties it to research on recursive training, distillation, and a thinning junior pipeline, and spells out what that means for founder-led teams of roughly twenty to fifty people. It ends with three actions that cost nothing but attention. If your differentiation lives in how you sound and what you dare to say, it pairs naturally with work on why value propositions feel right but do not convert and branding and positioning that actually sells.
Imagine a photocopier making a copy of a copy, then a copy of that copy. By the third pass, the text is blurry. The contrast has faded, and the detail is gone.
That is what is happening to a lot of AI-assisted marketing right now.
As AI-generated content spreads, large language models are increasingly trained on corpora that already contain synthetic text. Researchers call this model collapse: each cycle can narrow the distribution of what models produce. The range of plausible outputs shrinks, tails of the original distribution can thin, and the average becomes more average. Shumailov et al. formalised this in "The Curse of Recursion" (arXiv, 2023). A plain-language overview of causes and prevention sits on WitnessAI, still anchored in that paper. Follow-on work at NeurIPS 2025 traces a shift from generalisation toward memorisation as synthetic rounds pile up (Shi et al., arXiv:2509.16499).
Distillation compounds the effect. Large "teacher" models are compressed into smaller "student" models that inherit narrowed probability mass; Quanta Magazine explains the mechanism. In the market, cheaper variants and bulk distillation-style mimicry of frontier systems mean many tools sample a similar band. Less careful actors also bulk-query frontier systems to mimic behaviour. Your "unique" AI draft may be drawing from the same statistical neighbourhood as everyone else's.
Once you see it, you see it everywhere. Sign up for ten B2B SaaS trials and try to tell the onboarding sequences apart. Welcome, quick win, social proof, urgency, we noticed you have not logged in: same arc, tone, and slightly-too-keen energy. AI did not invent that playbook, but it industrialises it. You can generate a twelve-email sequence in twenty minutes that sounds like everybody else's twelve-email sequence.
It is not only email. Social captions, blogs, case studies, web copy, anything high-volume and templated drifts toward the same patterns. If you can describe the format in a prompt, the model nails it first time, which means everyone's version is a cousin of the post you scrolled past twenty minutes ago. Spend five minutes on LinkedIn and see if you disagree.
Marketing only works when it stands out and connects with buyers. That is the whole job. Average marketing was already easy to ignore before AI, simply from volume. Now there is more average content, produced faster, across every channel.
A survey reported via WebFX (syndicated through the Rio Grande Guardian, 2025) found that 90% of marketers say AI helps them produce content faster, while 78% saw a drop in organic reach within three months. However you slice the methodology, the headline tension matches what many teams feel: more output, weaker signal.
When effective volume doubles while attention stays fixed, perceived signal strength falls. Your newsletter competes with dozens that read the same way. Your LinkedIn post sits in a feed of interchangeable "thought leadership." Trade commentary increasingly names AI saturation as a risk: the more generic assistive copy a category produces, the harder it becomes for buyers to recognise a credible, specific voice.
The tier that used to be good enough, competent posts, acceptable case studies, solid but unremarkable copy, has become wallpaper. If you were already average, you are now aggressively invisible. That is why marketing strategy has to protect the slice of the work that is genuinely yours, not just accelerate production.
Here is where it becomes structural.
Automation of routine work has led many businesses to cut entry-level marketing roles. Constantine von Hoffman's reporting for MarTech (2025) notes hiring for AI-exposed junior roles down roughly 14% since 2022, with under-employment among recent marketing graduates above 41%, and ties those cuts to a future pipeline risk.
Junior marketers are the learning-by-doing engine: taste formed by getting it wrong and hearing why. Remove that pipeline and you lose tomorrow's strategists, not just today's capacity. Jennifer Spire, partner and chief executive of Preston Spire, quoted in the same piece, describes realising that skipping junior hires would leave a hole in the staffing plan by 2029 and 2030: who becomes the mid-level talent next? That is not abstract: it is the bench that would otherwise grow into the editors who stop your brand sliding into generic AI defaults.
The MarTech article also cites wider labour-market context: the New York Fed on weak outcomes for recent graduates, Pave data on sharply lower representation of 21- to 25-year-olds at large tech employers alongside an ageing workforce, and AWS chief executive Matt Garman's warning that replacing entry-level roles with AI to save money is short-sighted because juniors are often the cheapest and most AI-literate people in the building. The through-line is the same: if you delete the rungs, you cannot complain when nobody knows how to steer the machine five years out.
Senior marketers who remain often have judgement without prompt-native fluency. There is a gap between we have the tools and we know how to steer them.
So junior roles shrink, seniors are not all comfortable directing the machine, and the middle hollows out. That is a recipe for more generic output unless you design against it. It is also why treating AI as a hire you onboard and manage matters more than treating it as a magic typewriter.
If you lead a professional services firm, proposals may already sound the same: identical structure, reassuring filler, we pride ourselves on. Competitors read similarly because the same tools shape the same sentences. Differentiation used to sit in how you expressed expertise. That layer is easy to flatten.
If you are in retrofit or green energy, you were already fighting scepticism. Spec-sheet language instead of outcomes, now produced faster, makes skimming even likelier. Category noise rises.
If you run SaaS, onboarding, feature mail, and case studies blend into the same hum as every other B2B product. Churn is not always product failure. Sometimes nobody can say why your tool matters because your messaging sounds like everyone else's.
You may not have a £120k CMO. Your marketing coordinator may live in ChatGPT. Quality can degrade slowly, like a fifth-generation photocopy, unless you guard the edges.
The answer is not to stop using AI. Founders who use it well will outpace those who pretend it does not exist.
The answer is to know where AI breaks and protect what matters.
AI is strong on the bulk of the work: research, structuring data, first drafts, speed. It is weak on taste, discernment, and lived experience. It struggles to say this is not right when the grammar is perfect. It does not edit for judgement or empathy. It does not know what to leave out.
Teams that keep cutting through will use AI for velocity while protecting the minority of the work that differentiates: positioning, voice, strategic bets, the willingness to say something that halves the room. That is the same instinct behind getting clear on signal versus noise before you scale production.
The other half of the puzzle is making what you already know easy to retrieve and compose, so prompts stay short and outputs stay grounded in your language rather than the generic web. For a founder-level setup, read what if your business answers are already in your files.
Three moves this week, no new software required:
This is not only a technology problem. It is a marketing strategy problem.
The more marketing is generated recursively, the more future models train on AI-heavy text. Outputs can drift flatter and more interchangeable unless someone insists on originals, edits, and risk. The photocopier keeps running; someone has to check the tray.
At some point more organisations will realise they need leaders who can pair AI speed with taste: separate insight from polished slop, refuse volume-for-volume's sake, and build programmes that stand out rather than cheapen noise. The scarce skill is not prompting. It is direction: knowing what must never be delegated to the average of the internet. The question is whether you wait for that wake-up call or get ahead of it.
If you want clarity on what is actually driving growth versus what is activity, a Discovery Diagnostic is the right container: we map where knowledge and messaging live, where the gaps are, and the twenty percent move that buys most of the momentum. The deeper question is often whether you have a marketing problem at all, or a strategic clarity gap that more output won't close.
Explore more insights and strategies to elevate your marketing approach.
Federation beats one-tool thinking. Semantic search surfaced a coaching pattern in 30 seconds. Briefing tax, knowledge bridges, and a 15-minute island audit for UK founders.
Your value proposition feels right, but conversions aren't happening. Discover why most value props miss the mark despite perfect canvases, and learn a framework to build value propositions that actually convert.
Harvard tested 6 AI models on strategy. All gave the same trendy advice regardless of context. Learn why LLM strategy bias matters for founders and what to do instead.