GEO vendors misread the paper they keep citing

What happened

Per Search Engine Journal, GEO vendors pitching enterprise marketers on "Generative Engine Optimization" are repackaging conventional SEO tactics under a new acronym, and the academic paper they cite as foundation actually argues the opposite of what their decks claim. Pedro Dias, writing for SEJ, points to the 2023 Princeton-led paper "GEO: Generative Engine Optimization" by Aggarwal et al. as the document being misread across the vendor category.

The paper's finding most often quoted by vendors: certain content tweaks lifted citation visibility in generative engines by up to 40%. The finding vendors quietly skip: the lifts came from tactics like adding quotations, statistics, and citing sources, not from keyword density, backlinks, or the structural SEO levers their products optimise for. The mess in the title is the point. Generative retrieval is messy, probabilistic, and resistant to the deterministic playbooks the SEO industry spent twenty years building.

So the category selling itself to CMOs as "SEO for AI" is, by the evidence of its own founding citation, mostly selling SEO. The thing that actually moves LLM citations looks different.

Why it matters for your brand

If you run marketing at a bank, an industrial group, a UN agency, or a foundation, you are being pitched GEO tooling right now. The pitch usually arrives bundled: a dashboard showing your "share of voice" in ChatGPT, a crawler that audits your site for "AI readability," and a retainer to optimise pages against a proprietary score. The price points sit between 60K and 250K a year. The underlying work, in most cases, is schema markup, internal linking, and content refreshes. That is SEO. It is useful. It is not what wins citations in generative answers.

What actually wins citations, per the research the vendors cite, is content that LLMs find quotable: specific statistics, named expert sources, direct quotations, and clear authorial stance. This is closer to journalism and analyst publishing than to SEO. For a financial services brand, it means your economist's quarterly note matters more than your product page H1s. For a multilateral, it means a named country statistic in a report matters more than the meta description on the report's landing page. For an industrial group, it means a quoted engineer with a number attached beats a thought leadership PDF without one.

The implication for content strategy is sharper than most CMOs have absorbed. If you are buying GEO tooling to tell you which keywords ChatGPT prefers, you are buying a map of the wrong territory. The model is not ranking keywords. It is selecting passages it can defensibly quote. Your job is to manufacture quotable passages and distribute them where models retrieve from: Wikipedia, Reddit, established trade outlets, and your own site when it carries genuine authority signals.

For multilaterals and policy institutions specifically, this is good news that has been mispackaged as bad news. Your output already looks like what LLMs want to cite: named authors, specific numbers, methodology footnotes, source attributions. UNDRR, the World Bank, the IMF, and the OECD already publish in the format the models reward. The work is not to "do GEO." The work is to make sure those documents are crawlable, that the quotable lines surface in HTML rather than buried in PDFs, and that the named author signals reach the entities models recognise. That is a six week project, not a 200K retainer.

For industrial and financial brands without a research function, the implication is harder. You cannot fake a chief economist. You can, however, commission specific original data, name the analyst who produced it, and place it where models retrieve from. A single proprietary statistic, properly attributed and widely quoted across third party outlets, will out-cite a year of optimised blog posts. The budget reallocation is from content volume to content singularity.

The vendor risk is real. CMOs who sign multi year GEO contracts in 2025 will be defending those line items in 2027 when the dashboards still cannot prove citation lift, because the lever the dashboards pull is not the lever the model uses. Procurement teams at large enterprises and UN agencies should be asking GEO vendors a single question: show me the controlled experiment, on my content, where your recommendations moved citations in a named LLM. Most cannot.

The signal in context

The GEO category is following the trajectory of every previous marketing tooling wave: a real research finding gets compressed into a sales narrative, the sales narrative loses the nuance, and the tooling optimises for whatever is measurable rather than whatever matters. The same pattern produced the "social media ROI" dashboards of 2012 and the "content marketing platforms" of 2016. Both categories had real underlying truth. Both categories sold tooling that mostly automated the wrong work.

What is different this time is the speed of the feedback loop. LLM citations are visible. A CMO can ask ChatGPT, Perplexity, and Claude the twenty queries that matter for her category and see whether her brand appears. That visibility will, within twelve to eighteen months, separate the GEO vendors who can demonstrably move citations from those selling SEO with new vocabulary. The brands best positioned in that shake-out are the ones who treated 2025 as a year for original research and named expertise, not as a year for buying dashboards.