Key Takeaways
- LLMs retrieve information and present it — they’ve never built, executed, diagnosed, and iterated on a strategy. Not once.
- SEO is judgment calls, not tasks — and every campaign has moments where the standard playbook is dead wrong for that specific business.
- Context Rot is real — research shows every frontier LLM degrades as input increases, with performance dropping up to 85% even with perfect information retrieval. Full-campaign context is structurally impossible.
- AI agents work inside human-built systems — without strategy and guardrails, they just burn money faster.
- The market confirms it — SEO pros got pay raises while sites relying on AI alone lost 50–80% of their traffic after Google’s March 2026 core update.
The Blind Spot in “Fire Your SEO Team”
“Fire your SEO team. AI can handle it now.”
I see this take every week. And it’s not some fringe opinion — 85.56% of SEO professionals believe it’s moderately to highly likely that AI will assume SEO jobs in the future.
I get why it’s tempting.
I use AI in a ton of my SEO work. 86% of SEO professionals have integrated AI into their workflows. We’re not luddites clutching our spreadsheets. We see what these tools can do.
But there’s a blind spot in the “fire your SEO team” argument. It confuses doing SEO tasks with doing SEO.
Generating a keyword list is a task. Knowing which keywords to ignore because they’ll attract the wrong audience for this specific business — that’s SEO.
Running a technical audit is a task. Deciding which 3 of the 47 flagged issues actually matter for this site’s growth right now — that’s SEO.
The tasks are automatable. The job isn’t.
Retrieval Isn’t Strategy
Here’s what an LLM actually does. It retrieves information and presents it fluently.
That’s the whole trick.
It has never built a strategy, executed it, watched it underperform, diagnosed why, and iterated until it worked.
Not once.
And that cycle — build, execute, observe, diagnose, iterate — is the entire job. An SEO strategy isn’t a document you generate. It’s a living thing that changes every time you look at the data and realize something isn’t landing the way you expected.
LLMs can’t do that because they’re fundamentally designed for predicting next words. Research from INSEAD found that LLMs reproduce conventional, widely-accepted strategies — reducing novelty and uniqueness. They give you the textbook answer.
The textbook answer is frequently wrong.
Every campaign I’ve run has had moments where the standard playbook would’ve been the wrong call for that specific business, that specific market, that specific moment. LLMs can’t adapt to unseen scenarios because they’re pattern-matching against what already exists — not reasoning about what should happen next.
A Thousand Judgment Calls, None of Them Templated
SEO isn’t one big decision. It’s a thousand small ones stacked on top of each other. And almost none of them have a clean, universal answer.
Traffic drops. Is it technical debt? Algorithm shift? Seasonal noise? A competitor doing something smart? Or is it just the SERP volatility you cannot control?
Each diagnosis leads to a COMPLETELY different response. And the right diagnosis depends on knowing this specific business — its history, its competitive landscape, its technical setup, its content strategy, and a dozen other things that aren’t sitting in any single data source.
A B2B SaaS with 40 pages needs a completely different playbook than a local plumber with the same budget. Those are opposite ends of the spectrum. Now imagine two businesses that look almost identical on paper — and yet the correct approach for each might be vastly different.
AI can’t feel that difference.
The research backs this up. AI agents can identify keywords but can’t comprehend the true search intent behind them. They can’t determine if a keyword aligns with a brand’s voice, product, or business objectives.
Ronnel Viloria, Lead SEO Strategist at Thrive, put it well: “AI is great at executing commands, but it doesn’t know what questions to ask or what direction to take.”
That’s the whole problem. Knowing what questions to ask IS the job.
More Context Makes AI Worse, Not Better
The common comeback is: “Context windows are getting bigger. AI will eventually handle it.”
The research says the opposite.
Chroma’s Context Rot study tested 18 frontier models — GPT-4.1, Claude Opus 4, Gemini 2.5, all of them. Every single model degraded at every input length increment. No exceptions. More tokens in, worse output out.
That’s not a bug in one model. It’s universal.
A separate study published at EMNLP found that performance degrades 13.9%–85% as input length increases — even when the model can perfectly retrieve all relevant information. They controlled for everything. Replaced irrelevant tokens with whitespace. Forced models to attend only to relevant tokens. Placed all evidence right before the question.
Didn’t matter. The length itself causes the degradation.
Now think about what a real SEO campaign involves. Technical audit data. Keyword research across hundreds or thousands of terms. Analytics trends over months or years. Competitor backlink profiles. Content inventories. Search console data. Business context. Audience research.
For a small site, that’s already thousands of data points. For a multi-thousand-page site, multiply by orders of magnitude.
And the research says that every additional data point makes the AI’s output WORSE, not better.
This isn’t a “wait for bigger context windows” problem. It’s architectural. The more information you feed an LLM, the more it hallucinates, the more it loses track of what matters. SEO requires holding the full picture and making sharp calls against it. LLMs structurally cannot do that. And no amount of scaling fixes it.
Agents Need a System. Humans Build the System.
Here’s where the conversation gets honest.
AI agents execute tasks inside a system that already works. That’s real. I use them daily.
Content briefs, technical audits, rank tracking, reporting, data analysis — agents handle these well inside a framework a human designed. They’re fast. They don’t get tired. They catch things I’d miss at 2 AM.
But no system? The agent just burns your money faster than a shitty agency would.
No strategy means no guardrails. No priorities. No judgment about what matters for this specific business right now. The agent will execute whatever you point it at — but it can’t tell you what to point it at, and it can’t tell you when you’re pointing it at the wrong thing.
That’s why 58% of SEO professionals have evolved into what the industry calls “hybrid strategists.” Their value isn’t in executing tasks anymore. It’s in building the systems, setting the direction, and making the judgment calls that agents can’t.
Look at Bankrate.com — one of the most cited AI content success stories. They use AI to generate drafts. But every piece gets reviewed and modified by human specialists before publishing. The humans are the system. Remove them and the whole thing breaks.
The tool is only as good as the hand holding it.
The Market Already Picked a Side
If AI were about to replace SEO professionals, you’d expect the market to reflect it. Fewer jobs. Lower salaries. Less demand.
The opposite is happening.
64.5% of SEO professionals received pay raises in the past year. The market isn’t paying more for something it’s about to eliminate.
Meanwhile, the folks who went all-in on AI without human oversight got wrecked. Google’s March 2026 core update explicitly targeted scaled content abuse. Sites pumping out hundreds of AI-generated pages without editorial oversight saw 50–80% traffic drops. Niche sites with 500+ AI pages — high volume, thin depth, no author credentials, identical structure — lost 60–80% of their traffic.
That’s not a slap on the wrist. That’s a business-ending hit.
The companies winning right now aren’t replacing humans with AI. They’re giving their humans better AI tools. That’s not a semantic difference. It’s the entire difference.
You can’t automate judgment.
And SEO is almost entirely judgment.
Reach Out
Struggling to figure out what AI should and shouldn’t handle in your SEO? Reach out on LinkedIn or shoot me an email at tomislav@tomislavhorvat.com.
