AImarketing-automationcontent-strategyopinion

Output Is Table Stakes

Steve Glen•

I got a cold email last week from a marketing automation vendor. It opened with “In today’s rapidly evolving digital world” and promised I could “generate 500 personalized emails in under two minutes.” I deleted it, obviously. But the irony stuck with me: a company selling AI-powered marketing had just demonstrated, with its own pitch, exactly why AI-powered marketing isn’t working.

The email was flawless in every way that doesn’t matter. Grammar, structure, a clear call to action, a tidy PS line. And it had the unmistakable sheen of content that was never touched by a person who understood what I actually do.

Gartner’s latest CMO survey says marketing leaders expect AI-driven automation to more than double in the next two years, from 16 per cent of marketing work today to 36 per cent by 2028. Kristina LaRocca-Cerrone, a VP analyst at Gartner, framed it as a widening gap between CMOs who are “still testing use cases” and those “confident enough to use AI to create real brand differentiation.” That framing assumes the problem is confidence. I think the problem is that most teams are confidently automating the wrong thing.

They’re scaling output. And output is now table stakes.

The Zero-Cost Commodity Trap

When the cost of producing something drops to nearly zero, the value of that something follows it down. An 800-word blog post used to cost a few hundred dollars and three days of agency turnaround. Now it costs a fraction of a cent in API tokens and takes four seconds. Every company on earth figured this out at roughly the same time, and responded by flooding every channel they could reach.

But here’s what didn’t scale with the output: human attention. Attention is still finite, still expensive, and now actively hostile to anything that pattern-matches as generated. A Klaviyo and Datalily survey of 8,000 consumers found that when people notice AI-generated content, they’re four times more likely to trust the brand less (31 per cent) than more (7 per cent). The audience isn’t just ignoring the flood. They’re punishing it.

And the people doing the flooding know something is off. GrowthLoop’s 2026 index found that only 20 per cent of marketers report high impact from their AI experiments, and 77 per cent say tests that win in isolation fail at scale at least sometimes. The volume works in a demo. It works in a case study. It stops working the moment real customers encounter it beside everyone else’s identical volume.

What Actually Differentiates

If generating content is a commodity, then the competitive advantage has to be somewhere else. It’s in knowing things about your customer that your competitor doesn’t, and using that knowledge to say fewer, better things at the right time.

Consider the difference between a marketing team that uses AI to produce five thousand emails per week and one that uses it to identify the one moment a specific customer is most likely to act, then sends a single message informed by purchase history, product usage, and context. The first team has a prompt. The second team has a data architecture. One of those is reproducible by anyone with an API key. The other isn’t.

This is especially visible in industries where trust is load-bearing. In marketing strategy, a generic drip sequence doesn’t just underperform — it actively damages the relationship. When someone is evaluating whether they trust you with a financial decision, a synthetic email that reads like it was written for everyone reads like it was written for no one. The customer moves on to whoever remembered what they actually asked about.

The Search Engine Version of This Same Lesson

The collapse of the volume play is showing up in search, too. As of February 2026, 48 per cent of tracked Google queries trigger AI Overviews, up from 31 per cent in early 2025. The overlap between top Google links and the sources AI engines actually cite has dropped from 70 per cent to below 20 per cent. Churning out 50 SEO articles a week used to be a growth strategy. Now the AI overview absorbs your words and sends the traffic to whoever had the original data or the practitioner authority the model decided to cite.

Content that survives is content a model can’t synthesize from the open web: original research, real operational detail, a perspective that came from sitting in the chair.

The Part I Find Interesting

It’s a taste problem. Anyone can acquire the tools. Fewer people can look at what the tools produce and tell the difference between something genuinely relevant and something that’s just filler.

That distinction used to be a nice-to-have. An editorial instinct, a gut call, the kind of thing a senior marketer would apply before hitting send. Now it’s the entire game, because the tools will happily produce an infinite quantity of it and never once flag that none of it is earning attention.

Sources