What Not To Automate With AI: The SEO Deskilling Trap

When AI arrived in the marketing mainstream, it was accompanied by a persistent and scary narrative: The machines are coming for our jobs. On the surface, those fears might appear well-founded. According to the Content Marketing Institute, 43% of surveyed marketers said their organization had laid off marketing employees within the last year – a staggering 30% increase from 2024. For organizations with 1,000 or more employees, this number rises to 62%.

But a single stat can never give us the full picture. One thing we have in abundance right now is research papers, reports, and surveys trying to understand AI’s impact on business, on consumers, on creativity, on cybercrime, and, of course, on the workplace – from how we work to if we work. Taken together, these studies reveal a far more complicated story.

Anthropic recently published its report on the Labor Market Impacts of AI (March 2026), which found “no systematic increase in unemployment for highly exposed workers since late 2022.” And the World Economic Forum predicts that, while AI and information process technologies will displace about 9 million jobs by 2030, it will also create about 11 million new jobs. It seems AI might eventually drive a net gain in jobs.

Of course, stats like these are no reassurance to anyone who finds themselves “displaced” by AI in the interim. Anthropic’s report also ranked the 10 occupations with the greatest potential exposure to AI. Computer programmers (74%) top the list, while marketing specialists (64.8%) come fifth, which should leave us in no doubt that SEO as a profession is extremely exposed to AI disruption.

So, should we be worried or not?

The question isn’t how much you can automate or safely delegate to AI, or how small a team you can get away with. As it turns out, some of the most mundane or repetitive jobs, many of which might seem ripe for automation, may be far more valuable retained as manual, human-led tasks. Just because it’s easy or even cheaper to automate something doesn’t necessarily mean you should.

Augmented Versus Autonomous AI

Anthropic also publishes a quarterly Economic Index report, analyzing Claude usage data to track how people are working with AI in professional settings.

At the time of writing, the most recent report, Learning Curves, came out in March and draws on data from February 2026. It found that more than half (53%) of all interactions on Claude.ai are now “augmented” – human-in-the-loop interactions where the user learns, collaborates, and iterates on a task with Claude. Automated use – defined as interactions where the user delegates tasks entirely to Claude with little back-and-forth – has fallen to 44%.

So, is this more efficient?

The January edition, Economic Primitives, delves deeper into questions of task complexity, completion speed, and success rates – and this is where things get complicated.

It turns out that more complex tasks benefit from greater time savings. Working with AI can help users to complete tasks that would typically require a high-school education 9x faster, 12x faster for tasks requiring a college degree.

But these huge time savings come with a trade-off – and it’s a biggie. The same report found that basic queries or tasks, such as answering straightforward questions about products, currently achieve a 70% success rate. For more complex tasks, the success rate falls to just 66% for college-level work.

While that’s only a 4% difference, I’d argue neither result is particularly encouraging. To put it another way, the outputs from Claude aren’t up to snuff approximately one-third of the time.

One area where this low success rate has the potential to create issues is in code generation, which currently makes up 35% of all Claude usage.

Research from code review platform CodeRabbit found that AI-generated code produces roughly 1.7 times more issues than human-written code, including logic errors, readability problems, and, perhaps most concerning, security vulnerabilities.

If you’re an experienced developer, you’re more likely to spot the errors and improve on what AI has given you, treating Claude’s output as rough prototype rather than a finished product. But what if you’re not?

This is the dilemma: AI is not a replacement for genuine expertise. On the contrary, it appears a level of expertise is essential to using AI effectively.

Ironically, the people who would once have carried out a lot of these routine tasks – juniors or entry-level hires – lack the experience to assess what AI gives them.

That’s why I strongly believe no one should delegate a task to AI that they couldn’t do themselves. Once someone has learned a task, and developed a deep understanding of the concepts involved, then AI becomes a tool to speed up the process.

The Deskilling Trap

If expertise is vital to working with AI effectively, then it follows that businesses should focus on hiring people with the necessary skills and experience.

And multiple studies suggest this is exactly what’s happening.

At the same time, organizations appear to be increasing, rather than decreasing, their overall hiring of marketing talent by a significant margin.

This suggests organizations aren’t laying off staff or cutting back on junior hires to shrink their teams, but to reshape them. They’re hiring more senior, skilled, and experienced marketing talent who, as the CMI report puts it, “can direct, oversee, and – when necessary – rebut AI rather than compete with it.”

But is that conclusion supported by the data?

Both the Anthropic Labor Impacts report and the Revelio Labs research attempted to answer this question by comparing entry-level hiring patterns in industries and occupations with differing levels of exposure to AI disruption. The Anthropic findings, based on tracking the monthly job-start rate for younger workers (aged 22-25), were suggestive but not conclusive.

However, the Revelio Labs data focused on advertised entry-level job openings across four categories, finding that AI exposure has had a clear impact on entry-level demand:

Taking all the evidence together, the picture we’re left with is of a skills market in crisis. Most of the demand is now concentrated at the top, while the bottom of the pipeline thins out.

There’s a crunch coming.

The Qanat Problem

These days, talk of AI “transforming the landscape” has become an overused cliche. But around 2,500 years ago in ancient Persia, qanats were an equally revolutionary technology that quite literally transformed the landscape.

Qanats are precisely engineered underground channels, each one dug by hand by skilled workers called muqannis, using gravity alone to carry water over great distances from the mountains to the deserts.

Farms flourished. Cities grew. Persia bloomed.

Like AI, the benefits were huge, but the infrastructure was largely invisible. People became accustomed to drinking and bathing and irrigating their gardens with little regard for how the water got there.

Well-maintained, there is absolutely no reason why a qanat could not continue bringing water for hundreds, even thousands of years. In fact, some ancient qanats are still active even today, with 11 of these systems collectively designated as a UNESCO World Heritage Site.

Even if a qanat fell into disrepair through neglect – shafts left uncleared, tunnel walls allowed to crumble, silt left to accumulate – or if other, deeper wells extracted too much groundwater, lowering the water table below the level of the qanat, the consequences weren’t always immediate. Water would continue to flow for a while, but it would gradually decrease over time, until the flow became a trickle, then a dribble, and eventually … nothing.

Right now, businesses are happily drawing as much metaphorical water as they can from AI. However, the consequences of overuse and poor planning – such as applying AI to the wrong tasks – might not become apparent for some time. For now, the water still flows – but that doesn’t mean there’s no damage.

Many of today’s entry-level hires will go on to become the mid-level and senior talent of tomorrow. But without a constant flow of new blood entering the industry and gradually learning the craft, that skilled talent pool will soon shrink. And with demand for senior marketing expertise on the increase, you can expect the cost of hiring that talent to go up.

By then, it’ll already be too late to start hiring and training the next generation of marketers.

→ Read More: Ask An SEO: Should I Hire Candidates Who Can Use AI Tools Or Have Traditional Skills?

What Not To Automate

The default approach to AI adoption seems to be to identify any tasks that are repetitive, time-consuming, or mechanical, and automate them – or at least as much of the process as possible.

This isn’t necessarily wrong, and there are plenty of such tasks that can easily be delegated to AI without stealing valuable experience from someone, like downloading files, formatting documents, or aggregating data from multiple sources. There’s little to no value to be gained from expending human effort on these.

However, some repetitive tasks do generate value, even if, on paper, manually completing the task looks like cost and inefficiency. These are the tasks, which, over time, imbue an understanding of why something works. The value is in the investment you are making in your team’s development.

You might already have some form of staff development program or provide support to employees wanting to take up training courses. But this isn’t about sending your devs on a two-week course in JavaScript. This is about mastering the everyday stuff no course or textbook can teach you.

Keyword research is a good example of a task where SEO theory turns into practical understanding. Yes, AI can produce a keyword list faster than anyone, clustered by intent, filtered by difficulty, and mapped to the funnel. You could generate the complete report in the time it takes a junior to open a spreadsheet.

But by conducting keyword research for a wide variety of clients in different verticals and targeting different customers, a fledgling SEO will gradually acquire and hone their commercial instincts. Why are certain keywords more valuable than others? How do factors such as intent, geographic location, or even the time of year impact the results? Which keywords represent the strongest opportunities for a client?

It’s one thing to present a client with a neatly formatted document setting out a long list of viable keyword options, but it’s quite another to be able to answer the client’s questions, absorb feedback, and make further recommendations.

Don’t approach AI in terms of roles or job titles. The key is to audit your tasks and workflows to identify which activities don’t increase understanding and, more importantly, which ones do – and assign value accordingly.

This allows you to be deliberate and strategic about which activities to preserve as training infrastructure.

Practice Makes Perfect

The key to mastering any form of expertise is repetition – and there are no shortcuts.

Want to play the clarinet? Perhaps, you dream of fronting a jazz band one day, creating music that clings to the soul.

AI can play music. AI can even create music. But AI cannot make someone into a musician. It cannot replace the repetitive, tedious practice required for someone to develop genuine expertise.

In SEO and marketing, all those routine, repetitive tasks aren’t inefficiencies to be automated away. They are the scales. They are the learning to read sheet music.

You can’t magically imbue fresh-faced graduates with five years’ experience overnight. You need to give them five years working on the job, developing their skills, deepening their knowledge, and honing their instincts.

That’s why it is vital for businesses to keep hiring and developing new talent. After all, it’s far cheaper to hire, nurture, and develop internal talent than it is to compete for senior expertise in a shrinking pool – with salary expectations to match.

No one notices when a music student stops practicing, even more so if they never had the opportunity to start. But if too many budding musicians never master their instruments, there will be no one left to play in those jazz clubs and concert halls until, one day, the music stops.

More Resources:

Featured Image: Andrey_Popov/Shutterstock