SEO Copywriting and Generative AI: Regaining Control Over Editorial Content Creation
Generative AI tools accelerate content production. However, search engines and generative engines reward what no LLM can produce on its own: human effort.
SEO & GEO Summit x Acquisition Summit, at Disneyland Paris, October 15–16, 2026.
In May 2018, at WebCamp Day in Angers, I was already advocating a simple idea: a writer isn’t a typewriter—they’re a source of trust.
Eight years later, with E-E-A-T, search engines have proven us right. But not in the way we had anticipated.
The Top 10: That Average We've Learned to Emulate
For years, the SEO writer’s job has often been to imitate competitors. We’d analyze the top ten search results, compile a list of keywords from them, and then optimize each keyword one by one until we reached a certain score. Semantic optimization tools like Yourtext.Guru make this work measurable. That’s their strength. But when the pressure to meet KPIs takes over, you end up writing for a score rather than to convey a message. The result: content that looks just like the competitors’. But Yourtext.Guru has the major advantage of encouraging relevance and quality by providing new insights. Anyone who knows how to use this tool wisely already knows this.
Today, LLM-based tools quickly give the impression that things are simpler—even magical. It’s as if, right from your screen, you have something special that makes all the difference. There’s a certain technical aspect that’s becoming part of the profession. However, our use of technical methods for content creation is nothing new. If we had to name just one technique that could fall into the “hacks” category, have you ever tried extracting common phrases from a corpus—for example, Google Books summaries on photovoltaic panels? Conducting the right research and finding the right information also requires creativity in order to
The Delphic Cost: What We Delegate to Machines
Why did people consult the Pythia rather than search for answers themselves? Because she saved them the effort. Today, the LLM plays this role: it handles the searching, sorting, and thinking. The user no longer bears this cost; they simply ask a question and wait for an answer. The image is from an article on Yourtext.Guru titled “Google, the Pythia, and Me.” The LLM takes on the mental load of searching, thinking, comparing, and so on. It’s like looking for a vacation destination, but instead of scanning a world map for all the countries we could visit, the LLM immediately tells us the best destination to choose based on our profile. Convenient, isn’t it?
This convenience has a downside. Generative models encourage us to delegate our critical thinking. If we let an AI decide for us what information is best to share, we quickly fall back into excessive and meaningless optimization. Yesterday’s “top 10” becomes today’s run-of-the-mill answer.
The first "E" in E-E-A-T—the only one you can't copy
An LLM has a great deal of knowledge, but no experience. Expertise, authority, and reliability can be mimicked—at least on the surface. Real-world experience, however, cannot be generated.
At BrightonSEO in April 2024, I put it this way to the audience: it’s not just about knowledge. Knowledge is a skill, yes. But what matters is how I see the world, my perspective, what I feel. Good news for writers: ranking well in SEO already qualifies you for an LLM. The fundamentals still hold true. What’s changing is how we put them into practice.
Review Your Content Before Revolutionizing Your GEO Strategy
Before reinventing everything, take stock of what already exists. When faced with a question, a generative engine breaks it down into subqueries: this is called query fan-out. The way it’s handled isn’t all that different from inbound marketing strategies. The subqueries reflect several intentions: exploration, clarification, in-depth analysis, comparison, and entities.
The goal: to anticipate the LLM’s reasoning by leveraging in-depth knowledge of your product and customers. Product and category pages are a good example of this. Chris Walsh (IDHL Performance) demonstrated this at BrightonSEO in October 2025: an LLM extracts specific criteria (size, level, budget, local inventory) from a vague query. A product listing that meets these criteria is highly likely to be selected.
Optimization
Outdated message
Insufficient documentation
Significant research volume
Competitive Potential
Not in the top 3
Creation
Significant research volume
Competitive Potential
Query fan-out not yet processed
Internal reference resource
Formats that appeal to readers, search engines, and LLMs alike
Each content format serves as a communication tool between a brand, its audience, a search engine, and an LLM. The most useful formats include: FAQs, industry glossaries, case studies with statistics, objective comparisons, guides, annual industry overviews, training programs and certifications, and media interviews.
Regardless of the format, credibility must be evident. It's a true KPI.
What sets you apart is what is uniquely yours: your own data, fieldwork, and firsthand experience, as well as interviews with experts and recent, up-to-date, and annotated sources.
| Signal | Index |
|---|---|
| Human effort | The real, tangible work that's visible on the page |
| Author identified | Author Profile, Links to Their Profiles |
| Substantial revisions | A real update, not just a date change |
| Thematic Consistency | Semantic universe related to the same topic |
| Original images | Photographs, Infographics, Illustrations |
| Reinsurance | Legal Notices, User Reviews |
Our approach: building on what the company already knows
We view GEO as an opportunity to assess a company’s internal resources. The process consists of six steps:
Conduct an assessment of internal resources.
Gathering information: interviews, workshops, internal documentation, and any company-specific information.
Match this material with the fan-out queries.
Repeat the previous step.
Measure.
Start over.
The advantage of this approach is that the content produced inherently reflects the maximum human effort possible.
For 2027, two principles summarize the strategy for establishing authority with generative models. First, establish yourself as a primary source through the authors’ expertise and original information. Second, cite your sources, especially when providing specific information.
We explored this topic in greater depth in an episode of the Euphonie podcast dedicated to the authority and reliability of sources, featuring Eva Belgherbi, who holds a Ph.D. in art history, and Julie Guesdon, a journalist.
The pace of content production will not be enough
LLM-based tools speed up editorial production. But generative engines, like traditional search engines, base their criteria on human effort. The paradox: the LLM mass-produces content that it places little value on when searching for a source (watermarking).
Our real challenge in terms of visibility and acquisition will be to keep up with the volume without sacrificing quality, in the face of increasingly demanding LLMs.
