You've almost certainly seen this infographic. Nearly everyone talked about it.
Few people looked at the opportunities behind this shift, which is a tipping point in how we create SEO content (I covered the strategic context in my current thinking on AI SEO).
The problem with humans is that when you hand them something powerful, they often end up misusing it.
Since the start of the year, AI-generated content is said to have overtaken content written by hand. What looks like a problem is in fact the opportunity of the decade.
Let me explain:
If AI-generated content is fed only on AI-generated content (and no longer on content written by humans), a loop forms.
In the end, no original content gets produced anymore.
The only way out would be for AIs to generate new content themselves (which is possible, but complex).
The opportunity for SEO consultants:
If 100% of the content produced on the web is no longer the result of human thinking, general-interest informational content becomes a commodity. But a whole segment is left to exploit: micro-content.
Performance compared: AI content vs human content
The most significant study to date is the one run by Neil Patel. His team analysed 744 articles across 68 websites over five months.
Result: content written by humans generated 5.44 times more traffic than AI-generated content.
Another analysis, run by Ahrefs on 600,000 pages, finds that:
- 86.5% of the content ranking in Google's top 20 contains at least some AI-generated parts
- But purely AI-generated pages make up only 4.6% of the results
- The percentage is even across the whole top 20
I was against AI writing for a long time
Then I followed Google's recommendations and started using AI to fill "content gaps": adding new information that has never been used on the SERP.
I go much further today, and AI lets me:
- Identify a different editorial angle
- Add dissenting views to my own analysis
- Add context to my arguments (through specialised bots such as Notebook LM)
- Improve my semantics
What I think about all this:
One point to watch: some studies show that AI-generated content converts less well than content written by humans.
But that needs to be put in context:
- At the start, AI content was poorly made, because it was written with no thinking behind it
- Most consultants were doing full-AI writing: structure, text, FAQ, intro… everything
- And they used open, non-specialised LLMs (instead of closed bots like Notebook LM)
- Another problem: many wrote on subjects they had no real expertise in
Result: the content didn't rank, or ranked badly (except on non-competitive SERPs).
It's worth remembering that 80% of the web before AI was already mediocre, often word-for-word copies of the top 3 or translations of English-language articles.
Things are changing today:
- AI tools have got better at improving the base of a hand-written piece (see the prompts I use to produce expert content)
- Specialisation is now possible thanks to AI bots dedicated to a client's topic
⚠️ The real danger is losing control:
Like a developer who can no longer fix their own code, you end up not understanding why a text doesn't rank, or how to find complex search intents (primary / secondary).
You become the assistant's assistant, instead of making AI your assistant.
Does that mean giving up on informational content? No, as long as you have something new to add. Some page templates, such as "guide" pages, may only exist to feed your semantics and educate your reader, and no longer to rank.
Micro-content = micro-intent + content gap
The opportunity of the decade lies in micro-content, for three main reasons:
- Google and LLMs will need specialists, not generalists, to counter the wave of full-AI content
- AIs will absorb informational content, which will turn into conversations in the chat
- 80% of content will become a commodity. But the remaining 20% will make all the difference (and perhaps only 5% tomorrow). Those are the pieces LLMs will cite
How does it work?
You have to go ultra-niche and become an expert in micro-intents around a main keyword. That is exactly the logic of the semantic grid: targeting the pockets that generalists don't cover.
LLMs know how to break complex problems down into micro-intents:
- They will cite general sources for basic queries
- And expert sites for specialised queries
Your goal: exploit the under-covered areas of a topic.
Tip: scanning subreddits, LinkedIn and forums can help you spot content gaps. There are prompts dedicated to this; this one is in my top 5 (to send to Perplexity in deep search).
"Recency bias": write without wasting your time!
One data point to take into account:
What I told you above about writing expert content and targeting micro-intents is not enough.
100% of the content in the top LLM queries was written or updated in 2025.
Question: does updating a piece of content increase its chances of appearing in AI answers?
Yes. And a study proves it.
This phenomenon is called Recency Bias, and it will inevitably affect our SEO strategies.
Do LLMs prefer recent content?
Yes, systematically.
What the studies show
7 models tested: GPT-3.5-turbo, GPT-4, GPT-4o, LLaMA3-8B/70B, Qwen2.5-7B/72B.
All of them favour recent or updated passages, across the whole SERP.
Another study (Seer Interactive) confirms it:
- 65% of AI interactions involve content published in 2025
- 79% content from the last two years (2024–2025)
- 89% over three years (2023–2025)
- 94% over five years (2021–2025)
How to build this into your SEO strategy
- Avoid obsolescence: check figures, data, links and screenshots every year
- Add "news" context: laws, trends, product releases, dated benchmarks
- Build time-based internal linking: connect your evergreen pages to time-stamped content (events, laws, news)
- Add an update section: e.g. "What changes in 2025"
For the past few weeks, I've been building AI news bots by topic.
Goal: to automatically feed a new URL with fresh, relevant content, supporting (through internal links) the evergreen pages.
It's currently in the testing phase on my side.