The GEO pricing trap: why ‘Mount Stupid’ is costing you customers
Summarize this article with AI
Le décalage entre la promesse et les résultats est saisissant. Voici comment se comparent les chiffres de trafic pour ce client.
Promesse contre réalité : le gouffre GEO à 9 500 €
Le vendeur promettait +35% de trafic ; le client a obtenu 47 visites en 6 mois.
A Tuesday morning, €9,500 gone
A client calls me one Tuesday morning.
He runs a nutritional supplements store, 1,200 SKUs, 4,000 organic sessions per month.
He invested €9,500 in a turnkey GEO program.
He was promised:
« guaranteed citations on AI engines »
« +35% traffic in 90 days »
« schema markup optimized for retrieval »
Six months later: 47 visits.
Not a single order from this channel.
Regular organic traffic didn’t budge.
47 visits for €9,500. That’s €202 per visit.
Not a typo.
I look at the contract.
Big word: « certainty ».
The vendor committed to KPIs that nobody controls.
He was betting on a gradient that doesn’t exist: the louder you talk, the more the client pays.
My client lived through what Pedro Dias describes in a recent Search Engine Journal article.
He calls it « Mount Stupid with a pricing page ».
GEO, sold by generalists who don’t understand the model, is a product of broken promises.
And the client pays the bill.
Mount Stupid, breeding ground for GEO
Dunning and Kruger showed in 1999: people who know the least about a subject are often the most confident.
GEO amplifies this effect at industrial scale.
Why?
Because the technology is opaque.
Generative AI models are black boxes.
Even their creators don’t know exactly why one answer comes out instead of another.
The further you are from the model, the more certainty you can sell.
The closer you are to it, the more cautious you become.
The gradient runs backwards from common sense.
Dunning-Kruger bias applied to pricing:
At the foot of Mount Stupid, the vendor thinks he knows everything. He builds packaged offers with numbered promises.
At the summit, he talks about « 13% citation lift », « 2.8x on conversions ».
In the valley of humility, the researcher says: « we don’t fully understand what the model does ».
The GEO market has settled comfortably on the ascending slope of this curve.
Vendors of certainty are proliferating.
Their pricing pages display tiers: €2,500/month for basic visibility, €6,000 for « premium ».
And yet those who build the models speak in radically opposite terms.
The AI builders tell you: we don’t control everything
Pedro Dias, in his analysis for Search Engine Journal, compiles the most cautious statements from industry heavyweights.
« There is now abundant evidence that AI systems are unpredictable and difficult to control. »
— Dario Amodei (Anthropic), January 2026
Remember: Amodei himself sells these systems.
He’s not a detractor.
Anthropic, in its research on interpretability in May 2024, writes in plain terms:
« We mostly treat AI models as black boxes: something goes in and an answer comes out, and we don’t really know why the model gave this answer rather than another. »
Neel Nanda, head of interpretability at Google DeepMind, declared in September 2025 that the most ambitious version of mechanistic interpretability is probably dead.
He doesn’t see a realistic world where this discipline would deliver « the kind of robust guarantees some expect ».
And Ilya Sutskever, co-founder of Safe Superintelligence, used the NeurIPS platform to drop a bomb:
« The more it reasons, the more unpredictable it becomes. »
These are the voices of people who touch the metal.
None of them will sell you a ticket to the podium of guaranteed citations.
The business of certainty: what you’re being sold
Meanwhile, on your LinkedIn feed, the script runs like clockwork:
- « Schema markup ensures AI engines analyze your content. »
- « The first sentence of each section must be the answer. »
- « Optimize for chunk-level retrieval. »
- « A 13% citation bonus is available if you do X. »
These are mechanical recipes.
They make you believe GEO is a lever as manageable as a Google Ads campaign with bids and exact keywords.
My client who lost €9,500 bought exactly this narrative.
He was sold a certainty plugin.
Nobody explained to him that models are slippery surfaces, that a version change can relegate your site to oblivion without notice, and that the « 13% citation lift » depends on a test corpus that isn’t your market.
The Dunning-Kruger gradient has a very concrete expression in pricing:
The further the provider is from real complexity, the more they structure their offer with sharp price lines and supposedly guaranteed KPIs.
The closer they get to the code, the more they charge for time and architecture, with no promise of numbered results.
And buyers, badly advised, choose the shiniest ticket.
Forging an architecture that runs without magic promises
I don’t sell GEO.
I build semantic architectures.
With the same client, once we cancelled the GEO contract, we did exactly the opposite of what the LinkedIn script was selling.
We stopped chasing citations.
We mapped his domain: 47 structuring queries.
We built thematic clusters, page by page, silo by silo.
We put order back into a 1,200-SKU catalog that was competing with itself.
No visibility promises on ChatGPT or Perplexity.
No magical conversion percentages.
Result: +290% organic sessions in 10 months.
And AI engine traffic followed, without us targeting it directly.
Because structured, well-linked content answers questions… without needing to mimic a parser.
The DOSE framework I apply, forged by Guillaume Attias at BMO Academy, rests on one principle: architecture, not content.
Each page must live in an ecosystem of linked pages, with clear intent.
Your clusters are already natural chunks; no need to add chunk-level retrieval.
I charge by deliverable, not by promise.
And the system runs without me.
The counterintuitive: the less certainty you promise, the more value you create
Here’s what I observe with my clients year after year:
Transparency about uncertainty is a magnet for trust.
When you tell a prospect « this percentage of lift isn’t guaranteed, here’s why », you don’t lose the deal.
You earn the right to be heard.
The Mount Stupid vendors seduce fast.
But the wake-up is brutal.
And churn is high.
Their clients understand, after 6 months and 47 visits, that the pricing page displayed numbers resting on no verifiable mechanism.
There is no reproducible « 13% citation lift ».
But when you charge for semantic restructuring, an architecture audit, clusters delivered with a rollout plan, you’re not selling a miracle.
You’re selling a process.
And this process, you can demonstrate live.
That’s why my first client call is never a quote.
It’s a live audit.
I don’t tell you what I’ll do.
I show you what isn’t working.
A 3-week engagement, one deliverable, a measurable result: that’s the logical next step.
No GEO subscription.
Fewer promises, more movement.
And the client who lost €9,500?
Still with me.
Three years now.
Is your GEO on shaky ground?
Live audit of your ecosystem. I show you in 45 minutes what works, what limps, and what 3 weeks of architecture can change.
Book a strategic call — 45 minFrequently Asked Questions
What is « Mount Stupid » applied to GEO?
It’s the Dunning-Kruger effect: the less a provider understands the reality of generative AI, the more numbered certainties they sell. This bias drives untenable promises and inflated pricing pages.
Why are AI creators so cautious?
Models are black boxes. Dario Amodei (Anthropic), Neel Nanda (DeepMind), and Ilya Sutskever repeat it: AI becomes hard to control and increasingly unpredictable.
How do you spot an unrealistic GEO promise?
Watch for precise percentages (+13%, 2.8x) with no reference to the test corpus, guarantees of citations on engines that change constantly, and packaged offers without prior architecture audit.
What replaces a typical GEO engagement?
I map structuring queries, build content clusters, implement tight internal linking. AI traffic arrives if the architecture is solid.
Why do you charge by deliverable rather than subscription?
I forge a system that runs without me. A deliverable (cluster, architecture plan) gives tangible results. A typical GEO subscription promises maintenance without a real lever.
