How I think about the future of content marketing roles

The role that shrinks is producing volume. The one that grows still has no settled name on the org chart.

Disclosure: I am a Semrush Ambassador, and the study cited in this article was published by Semrush. The data is labeled with source and date checked, and the reading I make of it is my own.

Discussions about the future of roles tend to swing between two equally unproductive extremes: the announcement that the profession is over and the promise that nothing changes for people who are good. SEO has been declared dead so many times that it should have its own funeral home by now. The available data points to a third path, more boring and more useful, which is the redistribution of tasks within the same chain.

What the numbers support about return per piece

Before discussing return per piece, it is worth pinning down what has already been measured about click behavior when the answer shows up on screen.

What these numbers support is modest and useful: the chance of a piece being visited dropped, and the chance of it being cited without a visit went up. None of them says how much one more piece is worth in your portfolio, which is the budget question. To get there you need to measure in your own portfolio. The full reading of the Semrush study, with its coverage caveats, is in what 17 months of clickstream data say about ChatGPT traffic.

The four content functions I see splitting apart

• Entity curation. Someone needs to define and maintain the description of the brand, its products and its experts, consistently across every source that describes the company. Today this is scattered across branding, product and SEO, with no owner.
• Retrieval architecture. Structuring content to be retrieved in blocks, with local sufficiency, coherent internal linking and correct markup. It is technical and editorial work at the same time, which explains why few teams execute it well.
• Original proof. Producing the data that does not exist yet, be it an experiment, research, a benchmark or a case study with numbers. It is the only layer a model cannot generate from what has already been published.
• Distribution across surfaces. Taking the same asset to communities, video, events and third-party media, because citation depends on presence outside your own domain.

Which content tasks lose ground

Producing volume on topics already covered, generic briefs built from keyword exports, and surface-level review of generated text. These three tasks were the first to be absorbed by tools, and defending them on marginal quality has a short shelf life.

Does AI replace content writers?

The question comes up in every class and every team, and the answer changes depending on the task you look at.

The part of writing that consists of rearranging what has already been published is exactly the part a language model executes well, because it was trained on that material. Summaries of covered topics, support copy for long-tail keywords and format variations of the same piece fall into this band, and that is where price pressure concentrates.

The part that requires reporting stays with the writer: interviewing the people who did the work, running the test, checking the number in the primary source and signing the interpretation. It is also the part the numbers above show being rewarded, because the chosen source tends to be the one bringing what was not yet published. What leads a page to be chosen as a source is in how to become an AI reference source.

The skills I prioritize in people who write about SEO and AI

The list follows the four functions described above and works both for hiring and for planning your own development.

• Reporting. Getting to the primary data and logging where it came from, with a date.
• Reading data. Interpreting a Search Console series or an AI visibility tool series without confusing noise with trend.
• Writing for retrieval. Blocks that answer on their own, with the subject named and the source inside the passage.
• Enough technical grounding. Crawling, rendering and structured data at the level of holding a conversation with the dev team.
• Reviewing generated text. Checking claims, numbers and sources before publishing, with the same rigor applied to text written by hand.

For people who already work in content, the shortest path is choosing one of the four functions and producing public proof of it: a documented experiment, an entity curation done with the client’s permission, a piece with your own data. Public proof shows the method at work, which is what the job interview is trying to find out.

What I recommend for anyone building a content team now

Hire for the ability to investigate, with channel specialty in second place. A person who knows how to form a hypothesis, design a test and read the result adapts to any new surface. A person trained only in one tool’s workflow needs to be retrained at every market shift.