Felicity Carson, the CMO of onsemi, wrote a piece in Fast Company last month that I keep coming back to. Her argument is deceptively simple: AI hasn’t reinvented marketing. It’s just made it painfully obvious which marketing was built on substance and which was built on noise.

She’s right. And I think she’s being too polite about it.

Carson frames this as a discovery shift — LLMs reward structure, provenance and depth where search engines rewarded volume and keywords. Her "Keep, Drop, Scale" framework is a useful distillation for CMOs trying to work out what to stop doing. Keep what signals authority. Drop what exists to feed the algorithm. Scale explainability and context. It’s practical, it’s clear, and if you’re running a marketing team right now, you should probably read it.

But here’s what I want to push past. Carson writes for CMOs. Her framing assumes the reader already has a marketing strategy and needs to adapt it. Most of the business leaders I talk to aren’t in that position. They’ve got a website, maybe a blog nobody updates, a LinkedIn presence that’s somewhere between sporadic and abandoned, and a vague sense that they should probably be doing something about AI. They don’t need a framework for reallocating their content marketing budget. They need to understand why the rules changed and what that means for a business that never had a content team in the first place.

The correction, not the crisis

There’s a familiar anxiety doing the rounds: AI is going to destroy marketing, nobody will visit websites any more, search is dead, we’re all doomed. I’m bored of it, honestly.

What’s actually happening is more interesting and less dramatic. AI hasn’t broken anything. It’s corrected a market distortion that’s been running for about fifteen years.

The distortion was this: search engines made it possible to generate traffic through volume. Write enough blog posts stuffed with the right keywords, build enough backlinks, publish often enough, and you could rank for terms you had no real expertise in. It worked. An entire industry — content marketing, as practised by most agencies — was built on the principle that more content equals more visibility equals more leads. The quality bar was "good enough to rank." Not good enough to teach anyone anything. Not good enough to change how someone thinks about a problem. Just good enough to satisfy an algorithm that was fundamentally measuring the wrong things.

LLMs don’t work that way. They don’t count your blog posts. They don’t care about your publication cadence. They synthesise information from across the web and surface what’s most useful, most clearly structured, and most credibly attributed. Content that was built to game a ranking algorithm performs terribly in this environment. It was never built to inform anyone. It was built to be found, and there’s a difference.

And fair enough, that’s what search incentivised. The marketers who played that game weren’t stupid. They were responding rationally to the system they were operating in. But the system has changed, and a lot of that content is now invisible to the discovery mechanisms that matter.

Call it a correction rather than a crisis. The content that was always built to be genuinely useful — structured, clearly authored, actually informative — is doing better than ever. The rest is quietly disappearing from the discovery layer, and honestly, it’s hard to argue it didn’t deserve to.

What business leaders need to understand

If you’re running a business and you don’t have a marketing team tracking how AI search engines represent your brand — what Jellyfish calls "Share of Model," which is basically your visibility inside AI-generated answers — here’s the version that matters.

The way people find information is changing fast. Google announced in March that AI Overviews now reach more than two billion monthly users across 200 countries. ChatGPT still dominates AI-driven search referrals — around 80% market share according to Statcounter, though Gemini and Perplexity are chipping away at that. The point isn’t the exact numbers, which shift month to month. It’s what happens when someone asks an AI a question about your industry, your product, your expertise. What comes back isn’t a list of websites to browse. It’s a synthesised answer, drawn from whatever sources the model considers most authoritative.

Your website might not be one of them. And if it isn’t, you’re not just losing a ranking position. You’re absent from the conversation entirely. Knowing whether you’re in those answers at all is a separate discipline, one I’ve since taken apart in is your brand invisible in the answer engine?

I wrote recently about the agent-first brand — the argument that your next customer might not have eyes at all, because autonomous purchasing agents are starting to mediate buying decisions. That piece focused on structured data and schema markup for the AI agent economy. This is the precursor to that shift. Before agents start buying on your customers’ behalf, the humans using AI to research purchases need to be able to find you. And the mechanism for being found has changed underneath everyone while they were still optimising for Google’s old algorithm.

The unfashionable work that suddenly matters

The marketers who are best positioned right now aren’t the ones who’ve jumped on some new AI marketing tool. They’re the ones who were doing boring, unglamorous work that the rest of the industry considered old-fashioned.

Structured content with proper headings and logical hierarchy. Clear authorship with real expertise behind it. Genuine depth on topics they know about, rather than shallow coverage of everything vaguely adjacent. Content updated regularly, because they cared about accuracy rather than just publication volume.

This stuff wasn’t fashionable. The conference circuit was talking about brand storytelling, viral campaigns, growth hacking. The people carefully building structured, authoritative content libraries were the SEO nerds in the corner that the creative team didn’t invite to lunch.

Turns out they were building for where things were heading all along. Previsible’s research into nearly two million LLM sessions found that half of all content cited in AI-generated responses is less than thirteen weeks old. Content updated quarterly is three times more likely to earn a citation than stale material. I’m not sure all of those numbers will hold as the landscape matures — the methodology for measuring AI citations is still young — but the direction is consistent across every study I’ve seen. LLMs reward clarity, structure and demonstrated expertise. Not novelty. Not cleverness.

None of this required AI tools to build. It required discipline, subject knowledge, and the unfashionable belief that content should teach people something.

The three-question audit

Carson’s Keep/Drop/Scale framework is aimed at marketing teams with budgets and strategies to reallocate. Here’s a simpler version for everyone else.

Look at your content — your website, your blog, your LinkedIn posts, whatever you’re putting out into the world — and ask three questions about each piece.

Does it teach? Not sell, not promote, not vaguely gesture at your expertise. Does it help someone understand something they didn’t understand before? Content that teaches gets cited. Content that promotes gets ignored by AI systems, because it’s not useful to the person who asked the question.

Then ask about attribution. Is there a real person behind it with real experience? AI models increasingly weight authorship and provenance. A blog post signed by "the team" with no named author and no evidence of expertise is invisible to this new discovery layer. Put your name on things. Explain why you’re qualified to say what you’re saying — not as credentials-waving, as evidence.

And the honest one: does it game, or does it earn? Was this content created because you had something worth saying, or because someone told you that you need to publish twice a week to maintain your ranking? If the answer is the latter, it’s the stuff Carson would tell you to drop. She’s right.

Most businesses I work with fail on all three counts for at least half their content. That sounds harsh, but it’s not a criticism of the people who made it. They were building for a system that rewarded volume. That system is being replaced by one that rewards substance. The adjustment isn’t complicated, but it does require honesty about what you’ve been producing and why. And with AI tools now making it possible for non-specialists to build genuinely useful content without a full marketing team, the excuse that "we don’t have the resources" is wearing thinner by the month.

Why this isn’t a marketing problem

The reason I’m writing this for business leaders rather than CMOs is that the correction Carson describes doesn’t just affect marketing teams. It affects how your entire business gets discovered, evaluated, and chosen. The same reckoning is happening in software — this is the compiler moment for AI-assisted work — and it’s showing up anywhere the old volume-based playbook is being exposed.

A potential customer asks an AI for recommendations in your space. The AI synthesises its answer from the best available sources. If your competitors have clear, structured, expert content and you’ve got a brochure website with stock photography and a mission statement nobody reads — you’re not in the conversation. The customer doesn’t know you exist. They can’t choose to ignore you, because you were never presented as an option.

This used to matter less because Google would still surface your website if your SEO was decent. The game has changed. Being findable now requires being useful, and useful is harder to fake than it used to be.

I’m genuinely not sure this is settled yet. The discovery landscape is shifting fast, and I don’t think anyone — Google included — knows exactly where it ends up. But the direction is clear enough to act on. The businesses building genuine, structured, expert content are becoming more visible in AI-mediated discovery. The ones relying on volume, keywords, and algorithmic tricks are becoming less visible. That gap is widening every quarter.

The correction Carson identified isn’t coming. It’s here. And the good news, if you’re a business that actually knows something worth sharing, is that the playing field just tilted in your favour. The noise is getting quieter. The question is whether you’ve got signal to replace it.