What a Harvard Professor’s Rules of Writing Reveal About Why Most Marketing Content Gets Ignored

There is a particular kind of content that is technically correct but completely forgettable. The sentences are fine. The structure is there, and the ideas are in the right order. And yet, when you finish reading it, very little stays with you.

Steven Pinker, the Harvard professor and Author of Books like The Sense of Style and numerous other books, including Rationality, has spent his career studying language and cognition. Steve has an interesting explanation for why this happens. 

In a conversation with David Perell, he breaks down the rules of good writing in a way that felt surprisingly relevant to marketing content. It is not because he is talking about marketing. He is talking about how people understand, process, and remember what they read.

Two parts of the conversation stood out to me as a content strategist. Below, I am sharing things that hit me in particular and might resonate with you too, especially if you spend a good half of your day producing content or reviewing it. His podcast is especially timely in this era where content is a commodity, and we are all dealing with AI-generated content every day.

The AI Writing Problem Nobody Wants to Name

Pinker makes a notable observation about LLM output that is really precise.

He says AI writing tends to be structurally sound. The sentence structure is clean. The progression of ideas is orderly. There is an introductory sentence and a concluding one. In that narrow sense, it can often be clearer than what academics or corporate writers produce.

But then he says this: “You can almost recognise the output of a large language model. It’s so banal.” 

Banal. Unoriginal. Lacking creativity. Templatized. 

The problem is not necessarily that AI writing is wrong. Often, it is perfectly readable. The problem is that it can lack texture, specificity, or a point of view that feels like it came from someone who has actually spent time inside the problem. Pinker’s explanation for why this happens is one of the more interesting parts of the conversation. He compares LLM output to a composite photograph.

You see, if you take hundreds of faces and morph them together, the result can look technically attractive. Pleasant too. 

But it does not really belong to anyone. You have averaged out many of the characteristics that made the individual faces distinctive in the first place. There is something in that analogy that feels very familiar when you work with content regularly.

In our work, we use AI tools all the time. They are useful for research, structuring ideas, interrogating a topic, and sometimes getting an early draft onto the page. But that first draft is rarely where the work ends.

The reason is not always obvious grammatical problems. More often, it is a collection of smaller things.

You start noticing sentences that follow exactly the same rhythm. “Additionally” appears, followed a paragraph later by “Moreover.” One paragraph starts with “This,” the next with “That,” and then “This” appears again. Or every section follows the same neat pattern of setup, explanation, takeaway.

None of those things is terrible on its own. However, together, they create a sameness that becomes increasingly easy to spot once you have reviewed hundreds of pieces of content.

Good human writing usually has more variation. We change pace. We spend longer on one thought and move quickly through another. Sometimes we circle back because a later point changes how the earlier one should be understood.

AI tends to smooth those irregularities out. That matters when you consider how widely AI is already being used. Using its own detector, Ahrefs analysed 900,000 new English-language pages in April 2025. It found that 74.2% contained at least some AI-generated content.

The number itself is not the problem. But when a first draft receives little further work, all that polished writing can quickly begin to blend together. That is probably why some AI-assisted writing can feel perfectly competent and still leave very little behind after you finish reading it.

For marketing content, this matters. If ten companies publish broadly similar explanations of the same problem, the technically cleanest version is not automatically the one someone remembers, references, or recommends. 

Rather, the piece with a specific observation, an original example, or a perspective that could only have come from actually doing the work has a much better chance. That becomes even more important as discovery moves into tools like ChatGPT and Perplexity. Generic information is abundant. Giving someone, or an AI system, a reason to choose your explanation becomes harder when the page says little that could not have appeared somewhere else.

The Generalizations Plus Examples Problem

The second thing Pinker says that I keep thinking about is this: generalizations without examples are useless, and examples without generalizations are pointless.

You need both.

He demonstrates it in the conversation itself, which is probably why the idea sticks.

He uses this example: A bathroom isn’t necessarily a room with a bath. Breakfast isn’t necessarily breaking a fast. Christmas doesn’t necessarily refer to Christ’s mass. 

Pinker does the same thing when talking about concrete language. Instead of writing “the level of the stimulus was proportional to the intensity of the reaction,” he gives the much simpler version: “kids look longer at a bunny than a truck.”

The same underlying information has a very different reading experience.

This is something we run into constantly while building content. For instance, take a B2B SaaS company trying to understand why its marketing is not working. 

It could be for numerous reasons: 

• One company hired an agency and got traffic, but very little pipeline. 
• Another tried to build an internal team and struggled to get consistent output. 
• Another had a founder-led content engine that worked well for a while but is now seeing competitors appear more often in AI answers. 
• Another is doing well on traditional search but rarely appears when potential buyers ask ChatGPT or Perplexity for recommendations.

They are all experiencing a content marketing problem. But they do not have the same problem.

And that distinction matters when you are writing for them. In Edelman and LinkedIn’s 2025 B2B research, 85% of less-visible internal decision-makers valued a vendor who understood their specific challenges.

The content that makes someone continue reading usually names the situation precisely enough for them to recognise their version of it. You want the reader to move from “yes, that sounds broadly true” to “this is exactly what is happening to us.”

That is what the example does for the generalization. It gives it edges.

It matters for search in a similar way. Someone trying to solve a real problem does not always search for a neat category term. They might ask AI tools, “We hired a content agency and got traffic but no pipeline; what went wrong?”

That query contains context, history, and a very specific frustration. A page that actually addresses that situation has a stronger chance of being useful than one that talks broadly about why content marketing matters.

However, that’s where most writers miss the plot.  Pinker connects part of this problem to the curse of knowledge. The writer understands the category so deeply that they skip the part the reader needs. They jump straight into the framework, terminology, or recommendation without first establishing the situation that makes any of it relevant.

The reader has to do the missing work. Many simply will not.

One More Thing Worth Remembering

Pinker quotes two lines in the conversation that are short enough to overlook and useful enough not to.

Shakespeare – “Brevity is the soul of wit.”

Strunk and White – “Omit needless words.”

He points out that both lines demonstrate the principle they are describing. Neither needs a paragraph to explain the value of brevity. I like this because “write shorter” is easy advice to give and surprisingly difficult advice to follow.

You see, the goal is not to make every article short. There are some ideas that need space and brevity can be the enemy. The better question during editing is whether each sentence is earning the space it occupies.

We see this with both human and AI-assisted drafts. The first version often explains something, then explains it again slightly differently, then adds a sentence summarising what the previous two sentences already made clear.

Removing that third sentence does not make the writing less useful. Usually, it makes the argument easier to see. And the readability a ton better.

Takeaways From This Podcast 

What I liked about Pinker’s advice is that none of it requires another content framework. It’s in fact very actionable and can be summarized as follows: 

• Start with the specific situation, not the broad category. 
• If you make a general claim, show the reader what it looks like. 
• Use examples that make an abstract idea easier to picture. 
• When a draft feels polished but strangely forgettable, look beyond grammar and structure.

He urges writers to ask what is actually specific to this piece. To introspect on things like:

What did you observe that another writer would not automatically know? What example could only have come from doing the work? What detail makes a reader recognise their own situation?

While AI can help enormously with the writing, it still cannot do the job best. The specificity that makes a piece memorable usually comes from the research you did, the problem you have seen firsthand, the examples you chose to narrate, the customer conversation you remembered, or the judgement you brought to the edit.

That is the part worth protecting. There’s more in the podcast, and I’d definitely recommend it to every writer.