GEO: the new name for SEO? Why GEO tool promises deserve skepticism
Summarize this article with AI
A client calls me Tuesday morning. He invested $8,000 in a « GEO solution. » Six months later, 3,400 sessions.
The voice on the phone is tired. Six months ago, this e-commerce owner read articles about the urgency of positioning in generative AI responses. He bought a « turnkey » GEO platform.
$8,000.
Six months.
His site has 1,200 product pages. Technical datasheets, buying guides, an active blog. A solid foundation.
Before the tool: 3,200 organic sessions per month.
Today: 3,400.
200 sessions gained.
For $8,000, the ROI borders on ridiculous.
I see this scénario once a week in my audits. Executives who bet on the latest acronym. Who believed that an injection of markup or an « AI score » report would trigger a stream of qualified visitors.
The problem isn’t the budget.
The problem isn’t the need.
The problem is the promise.
They sold him classic SEO in fresh packaging. And because the packaging looked nice, he paid.
The GEO playbook: warmed-over SEO
In May 2026, Semrush publishes an infographic. Four pillars to dominate AI optimization.
Four pillars that every SEO specialist could have signed off on five years ago.
Pillar 1: quality content, search intent.
Pillar 2: authority, backlinks.
Pillar 3: SEO technique – speed, clean architecture.
Pillar 4: « Technical GEO » – schema, structured data.
3 out of 4 pillars are pure SEO, with zero generative specificity.
The fourth, the infamous Technical GEO, is the one that kills.
Pedro Dias, independent consultant, writes in Search Engine Journal: « The architecture of large language models is, by design, the opposite of certainty. »
schema.org has a precise role. It feeds rich results, it helps disambiguate entities in the Knowledge Graph, it serves voice assistants. But no LLM reads microdata markup to understand a sentence.
LLMs read words.
Nothing but words.
Yet GEO vendors swear that schema « guarantees that AI engines can analyze and connect your content. »
A phrase that sounds good in a webinar.
That slides into a PowerPoint.
That gets shared on LinkedIn.
But it’s false.
I observe in my clients that GEO scores from these tools predict nothing. Pages rated « perfect » by these platforms get zero citations in AI Overviews. Pages with zero markup appear naturally, because the text answers a specific question.
The model doesn’t cheat. It reads what you wrote. If you paint a technical layer over a poorly built wall, you don’t make the house more solid.
LLMs read text. Not tags.
To understand why technical GEO is a dead end, you need to look at how a language model processes information.
A transformer – the architecture behind ChatGPT, Gemini, Claude – receives a sequence of tokens. There’s no XML parser, no schema interpreter. Just word vectors and attention mechanisms.
The model never looks for an <FAQPage> tag. It reads the question and answer like a human reads a paragraph.
« There’s no analyzer in the model searching fortags. The model reads the words. That’s the mechanism. »
Pedro Dias, Search Engine Journal, May 2026.
LLM training data is the public web. A messy web, full of forums, typos, machine translation, snippets of code. The mess isn’t a bug: it’s the models’ strength. They learn language in all its roughness.
That’s why the idea of a « guaranteed structure » for AI is an illusion.
The GEO vendor tells you: « If you add these tags, the AI will understand better. »
Reality: the AI doesn’t care about your tags, it wants sentences that answer an intent.
I recently audited a B2B services site displaying a GEO score of 94/100 from the trendy platform. The tool recommended adding structured data for Organization, WebSite, and Article. We did the opposite: we removed that superficial markup and rewrote the page intros to contain a direct answer, readable, in natural language.
Result: in 8 weeks, the site appeared in 7 different AI Overviews.
No GEO tool.
No score.
I applied the DOSE framework. Result: +820% sessions in 14 months.
Back to the client from Tuesday morning.
After that call, I set one condition: stop all actions dictated by the GEO platform. No automated markup. No score to watch. We started from scratch.
I applied the framework I’ve been forging since 2016, DOSE, learned from Guillaume Attias (BMO Academy).
Four pillars that owe nothing to trends:
- Demand: map the real intentions behind each query.
- Offer: build pages that respond exhaustively, without bloat.
- System: connect each page to its thematic cocoon, with surgical internal linking.
- Expansion: roll out progressively, measure, adjust.
The client’s site: 1,200 pages, but thousand-layer architecture. Isolated pages, no link to parent, no cluster.
We restructured into 45 cocoons. Each around a main intent, then satellite pages to cover variants.
We stopped generic content production.
We restructured.
We invested the next $8,000 in the right place: information architecture, internal linking, title tag and H1 optimization based on real intent, and some targeted external linking.
14 months later:
4,000 organic sessions per month before my work.
37,000 sessions last month.
+820%.
No GEO tool. No AI score.
Just SEO that understands how any engine processes information: by meaning, not by tags.
The academic research GEO vendors cite… points the opposite way
GEO tools love citing research papers to give themselves credibility. One of the most cited is a Google DeepMind publication describing how LLMs exploit the implicit structure of text.
Problem: the paper doesn’t say what vendors claim.
It demonstrates that models draw their power from the apparent disorder of natural language. The author insists that the web, with its errors, redundancies, spontaneous microstructures, constitutes the best training possible. The paper’s title? The Whole Point Was The Mess.
Disorder is the fuel, not the enemy.
GEO vendors flip this idea: « Since AI needs to understand structure, give it more explicit structure. »
Except the paper concludes the opposite.
It’s the intrinsic disorder of language that lets transformers generalize. Not metadata.
I’m not saying you should write carelessly. I’m saying the race for perfect markup is a distraction.
When I observe sites performing in AI Overviews without complex schema, the commonality isn’t technical cleanliness. It’s message clarity, logical information hierarchy, content depth.
Exactly what Google has rewarded for 10 years.
Stop chasing acronyms. Return to fundamentals.
I’ve seen AEO, VSO, GEO pass by, each with the same story: « SEO is dead, here’s what replaces it. »
The playbook is polished.
A new acronym.
A fear.
A monthly subscription platform.
Those who fall for it are the same ones who bought voice positioning tools three years ago. The same ones who panicked at every algorithm update.
I’m not saying generative AI has no impact. It changes the search surface, it shifts user journeys. But the foundation stays the same: relevant content, trustworthy, well-organized, intelligently connected.
One of my clients, a health SME, never used a GEO solution. His site is built around information cocoons, every patient question has its answer, every answer points to a verifiable source. It appears in 12 AI Overviews for sensitive medical queries. Not because he marked up his articles, but because his architecture answers exactly what users ask.
SEO and GEO blur together. And if you master the first, you have no need for the second.
I’m not selling you the method. I’m showing you the pages.
GEO tools surf on your uncertainty. They make you believe a new discipline was born, that you need different expertise, more dashboards.
The truth, straight:
Optimizing for generative AI is optimizing for humans.
Humans who ask a question and expect a clear, structured, reliable answer.
Not humans who’ll check if you added an FAQ tag.
I’ve built more than 1,300 semantic cocoons since 2016. Every time I drop a magic solution and bring the client back to text, architecture, linking, results come.
Not in 3 weeks. But in 6 to 14 months. With solidity that survives updates.
SEO isn’t dead. It’s evolving.
Those selling its death to resell you the same thing under a new name are abusing your trust.
I’m not a miracle-solution salesman.
I build systems that run without me.
And you – how much did you spend on a GEO tool that never delivered a single qualified session?
Your site deserves architecture, not magic potion
In a live audit, I show you exactly what blocks your organic traffic. We talk real numbers, not promises.
Book a strategic call — 45 minFrequently Asked Questions
What exactly is GEO?
GEO (Generative Engine Optimization) refers to practices intended to boost visibility in generative AI engine responses (ChatGPT, Google AI Overviews, etc.). In practice, 3 out of 4 measures sold under this banner recycle traditional SEO: content, authority, technique.
Are GEO tools a scam?
Not a legal scam, but intellectual deception. They sell old wine in new bottles. Their « Technical GEO » (schema, structured data) doesn’t help LLMs understand text — these models read natural language, not tags.
Should I ignore AI optimization completely?
No. But optimizing for AI comes through fundamentals: answering real questions clearly, organizing info into thematic cocoons, building authority. Technical gadgets add nothing.
How do I measure impact in AI Overviews?
KPIs are still fuzzy. My clients track appearances manually via monitoring tools and correlate with organic traffic. But the best measure remains the same as classic SEO: relevant content attracts humans and, incidentally, AI.
Stéphane, how do you work with clients?
I don’t sell talk. First call is a live audit. I examine the site directly, show architecture knots, broken silos, uncovered intent. The client leaves with a clear roadmap, not a tool score.
