The Five Pillars of GEO Skepticism: A Critical Analysis

Few terms in B2B online marketing have taken off as quickly in 2025 and 2026 as GEO (Generative Engine Optimization), the optimization of content for AI search systems such as ChatGPT Search, Perplexity, Google AI Overviews and Gemini. Alongside the euphoria, a serious countermovement has emerged. Experienced SEO professionals, data analysts and agency critics accuse GEO of being old wine in new bottles, of resting on thin evidence and, above all, of selling one thing: uncertainty. This article organizes the criticism along five pillars, checks it soberly against the verifiable facts and derives a maturity model for Austrian B2B companies from it. If you are responsible for GEO budgets, you need both: the counterargument and the counter-counterargument.

Why GEO skepticism matters (and must not be ignored)

GEO skepticism is not a knee-jerk reaction from people stuck in the past. It comes from the center of the discipline and names real weaknesses: missing measurability, questionable studies, a channel that is still tiny in economic terms, and a consulting market that lives off the confusion. For decision-makers in the DACH region, the skepticism is doubly valuable. First, it protects against costly overreaction that is out of proportion to the channel's current business contribution. Second, it sharpens your view of which GEO tactics actually hold up and which are pure cargo cult marketing. The honest position rarely lies at the extremes. Anyone who sells GEO as a pure revolution or as a pure scam is oversimplifying. The following analysis separates verifiable criticism from gut feeling.

Pillar 1: "GEO is just relabeled SEO"

The most common argument goes like this: GEO is not a craft of its own, just SEO under a new name. This thesis has prominent supporters. In an industry analysis by Digiday, practitioners argue that GEO should be understood as an extension of SEO, not as a completely separate strategy, and that most GEO tactics rest on the same foundations as SEO because LLMs often pull their information from high-ranking, authoritative web content. In the same analysis, uSERP CEO Jeremy Moser boils it down to the formula that 80 percent of GEO is good, fundamental SEO.

The reduction is not wrong, but it falls short. If you dissolve GEO entirely into SEO, you overlook the remaining 20 percent in which the two disciplines differ: AI systems do not necessarily cite the position-1 page, but the passage that answers a specific question most precisely and earliest in the text (front-loading). They prefer extractable, self-contained answer blocks (answer islands) and weigh brand mentions differently than classic backlinks. The sober middle ground is therefore: GEO is a layer on top of SEO, not a replacement. GEO tactics cannot repair a foundation that is weak in technical and content terms. A strong foundation, on the other hand, becomes incrementally more accessible to AI answers through them.

Pillar 2: Criticism of the Princeton study's evidence base

The GEO movement likes to cite the Princeton GEO study, which reports visibility gains of up to 40 percent for individual methods. The paper was published at KDD 2024, according to arXiv, so it went through peer review. The most substantial criticism targets the methodology. SandboxSEO's analysis GEO Targeted: Critiquing the Generative Engine Optimization Research still referred to the preprint version in January 2024 and named the core problem like this: the author believes the methodology outlined in the paper contains serious biases, flaws and question marks. The main experiments did not run in a live search engine. The researchers fed each query together with its top five pages into GPT-3.5 via a template prompt, which means they simulated a generative engine. An additional test did run on Perplexity.ai, but without its web search: the researchers uploaded the source texts as files. It covered only 200 queries and, depending on the metric, showed up to 37 percent more visibility from added statistics and 22 percent from added quotations. According to SandboxSEO, this test used the same procedure and therefore carried the same biases.

In practice, this means: the often-cited percentage gains come from a lab environment with an older model and artificial prompts, and the values from the additional test on Perplexity.ai come from a small sample with uploaded source texts. They point to possible mechanisms, but they are not a sound basis for business decisions. A DACH B2B company that justifies GEO budgets solely with the Princeton figure is building on sand. If you cite the study, label it for what it is: a peer-reviewed lab experiment whose additional test on Perplexity.ai was limited to 200 queries.

Pillar 3: The volume vs. conversion paradox

The skeptics' core economic argument: AI traffic may convert well, but its revenue is still tiny. Both halves of that sentence are true. A GA4 analysis by Visibility Labs, reported by Search Engine Land, shows that ChatGPT traffic converted at 1.81 percent compared with 1.39 percent for non-branded organic search, which is 31 percent higher. The higher conversion rate is real. But the same evaluation of 94 e-commerce brands puts it into perspective: it compared 9.46 million non-branded organic sessions with 135,000 ChatGPT referral sessions, and ChatGPT generated 474,000 US dollars in revenue compared with 32.1 million US dollars from non-branded organic search, which is 1.48 percent of organic revenue.

The honest conclusion for Austria today: a channel with a revenue share of one to two percent rarely justifies a large project of its own. In B2B, where buying cycles are long and target audiences are small, the absolute volume is likely to be even lower than in the Visibility Labs e-commerce sample. The economically sensible path is not GEO as a separate budget, but GEO as a byproduct of an already strong SEO and content foundation. High conversion is an argument for attention, not for a blank check.

Pillar 4: The measurement and attribution problem

GEO has a hard measurement problem, and this is the most convincing pillar of the skepticism. Reliable, standardized metrics are missing, LLM ranking is a black box, and results cannot be guaranteed. On top of that comes an attribution defect that can be quantified. The Ghost Citations Study by Semrush and Kevin Indig examined 3,981 domain appearances across 115 prompts, 14 countries and four AI engines and found that 61.7 percent were ghost citations, meaning the domain was cited without the brand being named. Put differently: 62 percent of AI citations lead to no brand mention.

This gap is the Achilles' heel of any attempt to measure GEO success. When a source feeds an AI answer without the user seeing the brand name, there is neither a click nor conscious brand awareness, and the effect does not show up in any classic analytics report. Tools for AI visibility exist, but they measure mentions and citations, not guaranteed business value. For controllers in the DACH region, this means GEO reporting has to be understood as an approximation, not as reliable revenue attribution. Anyone who promises hard success guarantees misjudges the black-box nature of these systems.

Pillar 5: The "GEO grift" and the penalty risk

The fifth pillar is the most uncomfortable one: part of the GEO market mainly monetizes uncertainty. Agencies sell GEO packages whose effect they cannot prove, llms.txt is marketed as a mandatory file although no relevant AI search engine has confirmed that it takes the file into account, and some recommended tactics are worse than ineffective: they are risky. Self-congratulatory listicles in which a brand repeatedly ranks itself as the top pick can be classified as manipulation by search engines. Thin AI content produced at scale carries the same risk of being downgraded.

This pillar overlaps with pillars 2 and 4: where effectiveness cannot be measured and the research base stays thin, an ideal market emerges for promises without proof. The sober advice is to hold GEO tactics to the same standard as any other marketing investment. Tactics that are already good SEO and good content without the AI argument are safe. Tactics justified exclusively by an unverifiable AI promise belong under scrutiny. A technical file that nobody demonstrably reads does not replace substance.

What GEO skeptics overlook: the empirical click collapse

As accurate as the five pillars are, they leave out the real pressure to act. User behavior in classic search is changing measurably, and at the expense of clicks. A study by the Pew Research Center of 68,879 searches by 900 US adults shows that users who saw an AI summary clicked on a classic search result in 8 percent of visits, compared with 15 percent without an AI summary. The link in the summary itself was clicked in only 1 percent of visits. The analysis by Ahrefs of 300,000 keywords is even clearer: the presence of an AI Overview now correlates with a 58 percent lower average click-through rate for the top-ranking page, up from 34.5 percent in April 2025.

That is the decisive counterargument. Even if AI traffic brings in little revenue today, the classic organic click is shrinking as a source. If you only look at the status quo of referral volume, you optimize for a past that is currently eroding. The rational response is not panic but preparation: build content so that it ranks in classic search and can also be cited cleanly by AI systems.

Best practices: verifiable tactics instead of cargo cult

A disciplined practice follows from the criticism. Only invest in tactics that create value even without the AI argument.

• Foundation first: Technical SEO, topical authority and content depth remain the basis because, according to Digiday, LLMs often cite high-ranking, authoritative content.
• Front-loading: Answer the core question of a section in the first sentences before you elaborate. Extractable answers increase the likelihood of being cited.
• Answer islands: Structure content into self-contained blocks with clear headings that an AI can extract without losing context.
• Cultivate brand mentions: Since, according to Semrush, 62 percent of AI citations happen without a brand mention, it pays to build your brand directly into quotable core statements instead of only listing it in the legal notice.
• Schema for machine readability: Structured data helps extraction, but it is hygiene, not a miracle cure. It does not replace substance.

Common mistakes in dealing with GEO

• Selling the Princeton figure as proof: The 40 percent gain comes from a simulated engine with GPT-3.5, not from a live search engine, as SandboxSEO documents. That is not a sound basis for business decisions.
• GEO as a separate budget instead of a layer: At 1.48 percent of organic revenue in the Visibility Labs e-commerce sample, the channel rarely justifies an isolated large project.
• Promising guarantees: The black-box ranking of LLMs and 61.7 percent ghost citations make hard success guarantees dubious.
• Treating llms.txt as mandatory: No relevant AI search engine confirms that it takes the file into account. The file costs little but does not replace substance.
• Self-congratulatory listicles: Repeated self-praise placements risk downgrades instead of visibility.

Metrics and measurement: approximation instead of certainty

GEO success can only be measured approximately, and reporting has to make exactly that transparent. Three levels make sense. First, AI visibility: how often and with what share of voice the domain appears in answers from ChatGPT, Perplexity, Google AI Overviews, Google AI Mode and Gemini. Second, the ghost citation rate: what share of citations leads to an actual brand mention.

With 38.3 percent of appearances including a brand mention, the Semrush study provides a benchmark for the gap between being cited and being noticed. Third, the classic referral and conversion path from GA4, keeping in mind that AI referrals, according to Visibility Labs, convert at an above-average 1.81 percent but remain small in absolute numbers. None of these metrics delivers guaranteed revenue attribution. If you communicate that openly, you protect your own reporting from false promises.

Going further: a maturity model for Austrian B2B companies

The synthesis of skepticism and counterargument is not an either-or, but a question of timing. Three stages help you place your company.

• GEO is a waste if the SEO and content foundation is weak. In that case, every GEO euro goes into a layer with nothing underneath it. Foundation first, then AI optimization.
• GEO is optional if the foundation is in place and the target audience is still hesitant to use AI search. For Austria, there is an indicator for this: according to Statistics Austria, 73 percent of 16- to 74-year-olds in the population rate their AI knowledge as low or non-existent, and only 4 percent use generative AI tools daily or almost daily. This is a population figure, not direct evidence of how B2B decision-makers behave. At this stage, it is enough to write in a quotable way and monitor developments, without a separate budget.
• GEO is mandatory if the target audience actively uses AI systems for research and shortlisting and the classic click noticeably erodes. The Ahrefs finding of a 58 percent drop in CTR at position 1 shows that this point is getting closer for information-driven topics. If you start too late here, you lose visibility in both worlds.

The low AI self-confidence of the Austrian population suggests that the point at which GEO becomes mandatory will come later for many Austrian B2B companies. That does not cancel it. The smart approach is to build the foundation so that GEO comes along as a byproduct without extra effort, and to scale the channel exactly when your own target audience uses it. GEO skepticism and GEO preparation are not a contradiction. They are two sides of the same rational strategy.