Perfect for Whom?

An SEO Manager, a Digital Campaign Manager, a Product Marketing Manager, a Product Manager, a Copywriter, a Comms Lead, an Account Executive, and a Social Media Specialist all walk into a pub.

They’re asked: What does the perfect landing page look like?

Unsurprisingly, everyone has a very different answer.

The Copywriter, who has been through enough of these cycles to know how it ends, continues drinking their Guinness.

In walks a newly minted GEO specialist.

Silence.

Then a brawl ensues.

There is a framework for the perfect landing page. It is sixteen steps long. It does not ask who the page is for.

Alfred Korzybski had a line that has aged well: the map is not the territory. A representation of a thing is not the thing itself. The more useful the map, the easier it is to forget that distinction.

A 16-step framework has been circulating on LinkedIn.^[5] The perfectlanding page, reduced to an ordered list. It is a well-drawn map. Clear structure, sensible sequence, a coherent philosophy. On the surface, it makes perfect sense.

Hook the reader with an entity-based headline. Give them a reason to act above the fold. Build credibility with a citable asset and authority signals. Frame the problem as an AI-style query, answer it directly, back it with social proof and case studies. Close with transparent pricing, FAQs written for natural language search, a final CTA with urgency, and a human support signal. Make it fast on mobile.

The underlying philosophy: write for a skeptical buyer and an AI summariser simultaneously. Specificity beats superlatives. One page, one goal.

Below is that framework applied faithfully to Claude. Toggle the annotations to follow the map.

Walk through the specimen section by section. The same absence appears in every one.

The headline

The framework says: entity-based, outcome-driven, AI-readable. Fine. But whose outcome? The SEO Manager wants a keyword. The Product Marketer wants a positioning statement. The AE wants an objection handled before the call.Claude: AI that turns your best thinking into done work is a reasonable sentence. It is also three different briefs collapsed into one, with the conflict edited out.

The capabilities section

Six cards, six different jobs-to-be-done, implicitly six different people. The framework says to group by user goal, not by product module — sensible advice — but never says to pick one user. The result is a section that addresses a developer, a salesperson, a researcher, and a manager simultaneously. Each one sees themselves for one card out of six.

The right version of this section is persona-driven. One audience, one problem statement, all messaging in service of that context. That specificity also compounds outside the page: the landing page becomes the natural destination for persona-led campaigns on Search and Social where you control who arrives and with what intent.

Worth critiquing: persona-driven pages are not always the right call. If the product has genuinely horizontal value — a developer, a marketer, and an ops lead all arriving from different channels to the same self-serve trial — broad messaging is correct. Segmentation before value is established creates friction, not conversion. Persona specificity earns its cost when the acquisition channel is controlled, the buyer’s role shapes both the problem and the objection, and the sales motion is long enough for context to matter. When all three are true, a generic capabilities grid is not inclusive — it is a missed opportunity.

The social proof section

This is where the contradiction is sharpest, because it is written into the framework’s own instructions: vary personas — include different roles, company sizes, and use cases. That is the problem presented as a feature. If your testimonials cover an Account Director, an Engineering Lead, and a Head of Legal Ops, none of them sees themselves primarily reflected. You have made everyone feel partially seen. And one page, one goal is right there in the same document.

The problem and solution sections

Framed as an AI query and answer. This is where the GEO specialist walks in. The section is no longer written for the person reading it. It is written for the model that might summarise it. The human buyer’s actual problem — the one that got them to this page in the first place — is now a secondary concern, restructured to satisfy a retrieval algorithm.

The pattern repeats. The pricing section assumes transparency is always the right strategy — it is not, and any enterprise AE will tell you why. The FAQs are written in natural language so an AI can index them; real buyer objections live in sales call transcripts, not search queries. The final CTA assumes anyone who scrolled to the bottom is high-intent. Some are. Others were sent the link. High-intent is not a property of position on a page.

The framework is internally coherent. That is exactly what makes it dangerous. It answers every question except the one that matters first: who is this page for?

We try to make a single page work for everyone. In doing so, it works for no one.

Beneath that question, there is another: whether AI citation delivers a human to the page at all.

Nobody actually knows what AI looks for

The framework presents GEO as a layer of established optimisation logic: write directly, use natural language, answer questions, add schema, signal entities. The implication is that we know what generative engines respond to, and that following these rules improves your chances of being cited.

The researchers who coined the term are not that confident. The original GEO paper, published at ACM SIGKDD in 2024 by a Princeton-led team, is explicit: there remains no principled understanding of the underlying preferences of generative engines.^[1]

A September 2025 large-scale study across multiple verticals found that AI search has a systematic and overwhelming bias towards earned media — third-party, authoritative sources — over brand-owned content.^[2] A landing page is brand-owned by definition. It starts at a structural disadvantage in AI retrieval regardless of how it is written.

There is also the retrieval question. Analysis of ChatGPT behaviour finds that roughly 60% of queries are answered from parametric knowledge — the model’s training data — without triggering a live web fetch at all.^[3] Optimising the structure of a page for a retrieval event that may never happen is a speculative act dressed as a strategy.

Even when retrieval does happen, the citation is not a reliable signal. Language models generate citations the same way they generate everything else: token by token, from learned probability distributions, with no ground-truth check. The result looks authoritative. It may not exist. Citation fabrication is one of the most routine expressions of AI hallucination — plausible author names, real-sounding journal titles, confident URLs that return 404. Optimising to be cited by a system that sometimes invents its sources is a peculiar goal.

When retrieval does occur, there is no verify step. Standard retrieval-augmented pipelines retrieve, augment, generate. They do not confirm that the retrieved source says what the model claims it says, or that it is the most authoritative available reference on the point. The citation arrives in the output with the same confidence as a sentence the model fabricated entirely. There is no structural difference from the outside.

And for that 60% answered from training data: any citation produced is frozen at the training cutoff. A market figure, a published study, a product claim — if the model did not retrieve it live, it is citing a document it last encountered before the world moved on. The framework calls this channel AI citation. A more accurate label is a reference to a past state of knowledge presented as current evidence.

None of this means GEO thinking is useless. It means the framework presents evolving, contested, partially-evidenced practice as received wisdom — and that the people most likely to be harmed by that framing are the ones who follow it most faithfully.

Traditional SEO and landing pages are often in conflict

The framework positions itself as building on traditional SEO. That framing obscures a structural problem that predates AI entirely.

SEO optimises a page to be found. A landing page is built to convert someone who has already found you. These are different jobs. The buyer who arrives from a paid search ad, a sales email, a LinkedIn post, or a referral link has already passed the informational stage. They are not searching. They have arrived. The content written to rank — keyword-dense, question-answering, structured for crawlers — is written for a person who is one step earlier in the journey than the person now reading the page.

The documented consequence is intent mismatch: research on SEO landing page performance consistently finds that pages optimised for search terms rather than buyer stage show significantly higher bounce rates and lower conversion, because the content answers a question the visitor stopped asking before they clicked.^[4]

Writing FAQs exactly as a user would ask an AI assistantis a version of this problem. The user is on the page. They are past the query stage. They do not need the page to answer the question that brought them here — they need it to answer the question that determines whether they act.

None of these sections are wrong in isolation. Most of them are good practice, for the right product, sold to the right buyer, arriving from the right channel.

That qualifier — for the right— is doing all the work.

The framework omits it entirely. It treats the landing page as a genre with fixed rules, the way a sonnet has fourteen lines. But a landing page is not a form. It is a conversation with a specific person who arrived from a specific place with a specific thing they want to resolve. The rules change every time.

Back to the pub. The brawl does not start because the GEO specialist is wrong. It starts because every person in that room has been optimising for their own metric without anyone first agreeing what the page is supposed to do — and for whom. The framework gave them sixteen things to argue about instead of one question to answer together.

The map is not the territory. The framework is a map drawn before anyone decided where they were going.

The 16 steps are not the problem. The order of operations is. Agree on the audience, the acquisition channel, and the one thing you need them to believe before they act. Then open the framework. Most of the sections will sort themselves out.

References

1. Aggarwal, P., Murahari, V., Rajpurohit, T., Kalyan, A., Narasimhan, K., & Deshpande, A. (2024). GEO: Generative engine optimization. In Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining. ACM. https://arxiv.org/abs/2311.09735
2. Chen, M., Wang, X., Chen, K., & Koudas, N. (2025). Generative engine optimization: How to dominate AI search. arXiv preprint arXiv:2509.08919. https://arxiv.org/abs/2509.08919
3. Digital Bloom. (2025). 2025 AI citation & LLM visibility report. https://thedigitalbloom.com/learn/2025-ai-citation-llm-visibility-report/
4. Hashmeta. (n.d.). Why most SEO landing pages fail to convert: The hidden performance gaps. https://hashmeta.com/blog/why-most-seo-landing-pages-fail-to-convert-the-hidden-performance-gaps/
5. York, A. (n.d.). Your landing page ranks on Google, but it still doesn’t convert [LinkedIn post]. LinkedIn. https://www.linkedin.com/posts/anna-york-seo_your-landing-page-ranks-on-google-but-it-activity-7442544592496799744-UIKn