GEO by Engine: What’s Still Missing from 90% of Playbooks

The GEO Playbook Everyone Copies — and the Part It Misses

A client calls me on a Tuesday morning. He invested $8,000 in content designed for AI. ChatGPT cites him 3 times out of 47 checks. Conversions stayed flat.

The per-engine mechanics were the issue, not the content.

I review 15 sites a week. They all have the same blind spot. They treat GEO as an extension of 2019 SEO: keywords, a few FAQs, a bit of schema markup. Result: the pages exist, but Claude doesn’t cite them. Perplexity ignores them. Gemini drowns them in a generic answer.

Neil Patel published an analysis that moves the needle. According to him, GEO is not a single channel: ChatGPT, Claude, Gemini, and Perplexity apply different decision rules. What makes a page surface on ChatGPT can make it invisible on Claude. And vice versa.

What playbooks are missing is an engine-by-engine reading. One more layer. You keep your current tools. You keep your current content. You just understand where each engine looks for sources, then structure your pages accordingly.

Most teams stop at citations. They think a strong domain score is enough. No. A site can dominate Google and remain invisible on Perplexity. I see this gap every week.

In his article, Neil Patel gives no magic recipe. He breaks down the mechanics. That’s where it becomes useful. A list of generic best practices doesn’t help much. Knowing that ChatGPT favors certain content structures and Claude looks for others — that’s concrete.

Simply put: a page that performs on Google can remain invisible on ChatGPT. And a page that surfaces on Perplexity can be ignored by Gemini. Winning teams target all four engines, each with its own structure.

When a playbook stays descriptive, it gives the impression of action. In reality, it hides the blind spots. The only way to see them is to dig into the details: which engine cites which page, for which query, and how consistently.

How many of your pages does Claude actually cite today?

ChatGPT, Claude, Gemini, Perplexity: Four Engines, Four Citation Logics

Neil Patel’s analysis lands at the right time. Instead of treating GEO as one block, he details what each engine looks at before citing or summarizing a source.

ChatGPT leans heavily on page structure and named entity density. Claude looks first for clarity of argument and extractable passages. Gemini pulls data from the Google ecosystem, especially the Knowledge Graph. Perplexity aggregates citations: the more different sources and conversational mentions it finds, the better.

A table beats a long speech:

What 90% of playbooks miss: segmentation of optimizations. The same page cannot please ChatGPT and Claude in the same way. You need structural variants. Not duplicate content. Different semantic layers depending on the engine’s reading angle.

For example: a product page built for ChatGPT benefits from exposing attribute lists and precise entities. For Claude, the same page benefits from adding a summary box with an authority sentence. For Perplexity, you need an internal citation mesh that points to other pages on the site.

This granularity is missing. Most teams stop at the global level: « you need to be cited. » But being cited by whom, for what, and with which page structure — that’s still the playbook’s job.

The DOSE Gap: Why Closing a Blind Spot Triggers More Action

I learned the DOSE framework with Guillaume Attias at BMO Academy. The completeness effect is still underused. When you identify a specific gap, the brain wants to close it. A generic list has no such effect.

Classic GEO playbooks list actions: « optimize your FAQs, » « use schema markup, » « refine your entities. » These lists are true. But they create no tension. They don’t point to what’s missing.

The DOSE gap flips the logic. Instead of « here’s what to do, » you say « here’s what you’re missing on Claude, and how to close it. » Completeness kicks in. Action follows.

Neil Patel got this. His article doesn’t just define GEO: it shows the per-engine blind spots. That’s what pushes people to act. You see the missing piece.

In the field, I see the same effect. When I show a client that ChatGPT cites them 17 times out of 100 queries while Claude cites them only 4 times, the decision to restructure doesn’t need debate. The gap is visible. The brain wants to close it.

A live audit beats a 60-page report. We look at the gap together, decide, act.

DOSE works as a reading grid: it turns a technical finding into a decision. And in GEO, closing per-engine gaps pays more than producing more content.

A Real-World Case: 47 Queries, 945 Pages, and the Architecture Shift

A B2B client called me 14 months ago. 945 pages. 47 high-volume queries. Organic traffic steady at around 4,000 sessions per month.

The content held up. The pages answered the questions. But Claude citations remained rare. On Perplexity, the site appeared in 6 out of 100 answers. On ChatGPT, 11 out of 100.

The problem was architecture. The 945 pages lived in isolation. Semantic clusters were absent. Links between pages were absent. Each page had to convince alone. An engine like Claude doesn’t like that. It looks for internal coherence. A source that relies on other sources.

We stopped content production. Three months. All in on restructuring. Pages grouped into 47 clusters, one per strategic query. Summary boxes rewritten. Contextual links to source pages, not conversion pages. Then separate structures: ChatGPT versions, Claude versions, Perplexity versions, with a common base.

We also reinforced entity signals: attribute lists, technical synonyms, exact product names. For Gemini, I added fresh data and recognized external sources. For Perplexity, cross-mentions with partners and niche directories.

14 months later: 37,000 organic sessions per month. +290% conversions, ad spend unchanged. Claude citation rate rose from 4% to 22%. On Perplexity, from 6% to 31%.

The per-engine mechanics made the difference. Each page knows why it exists. For whom. And how it will be cited.

Voici la mécanique que j’applique, dans l’ordre. Cinq étapes suffisent pour transformer un discours générique en structure que les moteurs citent réellement.

Three Blind Spots to Fix Before the End of the Quarter

Three zones I see neglected, week after week.

First zone: the per-engine citation layer. Optimizing for ChatGPT and Claude with the same page means losing the completeness effect. Create structural variants: a « summary » box for Claude, an « entities » block for ChatGPT, a « fresh data » block for Gemini.

Second zone: citation-oriented internal linking. Answer engines favor sites that link to themselves logically. A visitor doesn’t need those links. An engine does. Build contextual cross-links between pages in the same cluster. Make every link establish a hierarchy of evidence.

Third zone: the excerpt before the citation. People often forget to optimize the first 40 words of each section. Yet that’s what Claude extracts for its answers. If your first 40 words are descriptive, you get summarized. If they’re assertive and quantified, you get cited.

These three zones form a system. Not three isolated actions. The combination shifts visibility.

According to Neil Patel, Perplexity favors a multiplicity of sources; a site that is never co-cited starts with a handicap. A page that checks these three boxes doesn’t look like a classic blog post. It looks like a structured source. That’s what AI engines are looking for.

Deux sites, deux moteurs, deux mesures avant / après. Le verdict est net : dès qu’on optimise pour un moteur précis, les citations décollent — sans budget pub ni volume de contenu en plus.

What I See in Clients Who Finally Shift to Engine-Specific GEO

Of the 15 sites I audit each week, 4 have shifted to an engine-by-engine logic. On Google, little change. On answer engines, a lot.

One local site, 22 target queries on Perplexity. Before the shift: 3 citations out of 100. After 5 weeks: 17 citations out of 100. Paid links unchanged. Content volume unchanged. Just restructured excerpts and internal linking.

Another, an e-commerce player, saw its Claude answers rise from 6% to 19% on 240 product queries, by adding « decision summary » blocks and comparison tables at the top of the page.

What I take from these observations: AI engines reward clarity, not density. A short page structured for citation beats a long page that piles up keywords.

The per-engine gap matters more than volume or backlinks. Structure decides.

These four clients have one thing in common: they first identified the engines that cited them least. Then they structured. Most teams do the opposite: they produce content and hope it surfaces.

Generic playbooks don’t offer this precision. They stick with « be visible in AI. » True. But useless. Visibility is won in 3-to-5-point gaps per engine. That’s where the game is decided.

Before Publishing Your Next Page, Ask One Question

The next page you write has a function. Which one?

If the answer is « cover a keyword, » the page has already lost on ChatGPT and Claude. If the answer is « serve as a structured source for a specific engine, on a specific query, with a specific excerpt, » it can rise where others stall.

A descriptive playbook tells you to publish. An engine-by-engine playbook tells you for whom, for which reading, and with which citation structure.

The prep work is simple:

• Identify the 20 queries for which you want to be cited.
• Check on ChatGPT, Claude, Gemini, and Perplexity whether you’re already there.
• Note the type of content each engine surfaces for those queries.
• Design a structure that gives each engine what it expects: entities for ChatGPT, summary for Claude, freshness for Gemini, co-citations for Perplexity.
• Link this page to other pages in the cluster, not to conversion pages.

Same workload, more precision. And visibility moves faster than with 20 generic articles.

Nothing to sell. Just the pages.

Which page are you going to restructure next for Claude, ChatGPT, Gemini, or Perplexity?

Stéphane Jambu

SEO & AI Engineer

I build growth systems / AI / Neuroscience | 650+ clients · 80 LinkedIn testimonials · 30 years of expertise · 15 years of systems running without me.

Follow on LinkedIn