Jef Van Gool

Generative Engine Optimization (GEO) became my department over the weekend. As a forensic SEO (shout-out to all my fellow SEOs in the chat, make sure you leave a comment for the people needing SEOs after reading this 😆), the first thing I ask in my new role as a GEO expert is: "What facts are we working from?" I go to the academic paperwork. Here is day 1, 2 minutes into the job. "Has anyone actually looked at the original November 2023 academic paper (arXiv:2311.09735, later formally published and presented in August 2024 at the ACM SIGKDD 2024 Conference on Knowledge Discovery and Data Mining / KDD ’24, DOI: 10.1145/3637528.3671900) that coined and defined the acronym GEO - titled "GEO: Generative Engine Optimization" - authored by Pranjal Aggarwal et al, in which the authors explicitly take credit for inventing the term ("To address this, we introduce GENERATIVE ENGINE OPTIMIZATION (GEO), the first novel paradigm to aid content creators in improving their content visibility in generative engine responses...") while claiming in Section 2.2 that "traditional SEO methods are not directly applicable to Generative Engines" because traditional search engines are supposedly "limited to keyword matching"?" I see a lot of "GEO" agency and SaaS tools in Google repeating the same line from the KDD ’24 paper (P Aggarwal · 2024 · Cited by 401!!): "Adding citations and statistics increases AI visibility by over 40%!" I ask the team (myself): "Did nobody actually read Table 4 of the paper to see HOW they got that number? They didn't do real research. They told gpt-3.5-turbo to rewrite pages and add stats/citations "without adding any substantial new information". So GPT-3.5 - AI - HALLUCINATED a fake chocolate institute and invented a 70% robotics stat out of thin air - and the downstream GPT-3.5 summarizer cited the page 132.4% more often purely because it fell for an authoritative-sounding bracket: (According to a survey conducted by The International Chocolate Consumption Research Group).😄 Even better? Look at Appendix C.1 (Page 12): their "in the wild" Perplexity test didn't even use live web search. They manually uploaded 5 text files into a closed chat window. Even their methods without SEO... are the definition of low-quality scaled content abuse! It's not GEO they are selling - it's SPAM. Strip away the traditional SEO that got those 5 pages into Google's Top 5 in the first place, and the entire "GEO" industry is built on one LLM lying about a fake Willy Wonka chocolate factory institute (literally!) to trick another LLM in a 5-file sandbox. 🍫🤡 And it gets worse....! Anyways, I quit my job in GEO after 2 minutes, and I went back to my day job in SEO. The term Generative Engine Optimization (GEO) was coined and introduced in a 2023 academic paper titled "GEO: Generative Engine Optimization" (presented at KDD 2024) by: - Pranjal Aggarwal (IIT Delhi & Allen Institute for AI). - Vishvak Murahari (Princeton University). - Tanmay Rajpurohit (Georgia Institute of Technology). - Ashwin Kalyan (Allen Institute for AI). - Karthik. Narasimhan (Princeton University) - Ameet Deshpande (Princeton University & Allen Institute for AI). In that "foundational" paper, Aggarwal and his colleagues explicitly took credit for the name, writing: "To address this, we introduce GENERATIVE ENGINE OPTIMIZATION (GEO), the first novel paradigm to aid content creators in improving their content visibility in generative engine responses..." While the Princeton paper proved that LLMs can be tricked by sycophantic, authoritative-sounding fluff, calling this "optimization" IN ANY WAY is a dangerous ILLUSION (at best!!). In the real world, injecting fake data into a webpage is just adversarial spam, and search engines are heavily engineered to destroy it. Some balls there, boys; I will give you that. When you strip away the academic jargon and the mathematical symbols, what Pranjal Aggarwal and his co-authors actually stood on stage at KDD '24 and presented was a closed-loop prompt-injection experiment where gpt-3.5-turbo Graded Its Own Homework. This peer review group demonstrably has zero knowledge of how search engines work and absolutely no knowledge about real SEO other than they need it for GEO lol. They walked out of a university lab with a "GENERATIVE ENGINE OPTIMIZATION (GEO), the first novel paradigm," when they have no clue how the current working paradigms actually work! When any folk came after you published your paper and tried to write about GEO with any historical depth or original concepts that weren't traditional SEO (AI SEO) and perform similar academic parasite seo, they had to steal my work in AI SEO (Inference Optimisation and a Fortress of Facts strategy - AKA IP Armour!) to have any grounding in fucking reality in modern marketing. But for this: "However, traditional SEO methods are not directly applicable to Generative Engines. This is because, unlike traditional search engines, the generative model in generative engines is not limited to keyword matching, and the use of language models in ingesting source documents and response generation results in a more nuanced understanding of text documents and user query."" ...ignoring over a decade of Google’s semantic and neural search architecture (Knowledge Graph in 2012, Hummingbird in 2013, RankBrain in 2015, Neural Matching in 2018, BERT in 2019, and MUM in 2021). etc etc. Here's my peer review of your paper: 100% of the legitimate optimization in the paper is SEO and AEO (0% is "Pure GEO"). 88.9% (8 out of 9) of their tested "GEO methods" are textbook On-Page SEO, and the remaining 11.1% (1 out of 9) is 1998 black-hat keyword stuffing. The only ~15% of the paper that isn't directly lifted from SEO is a passive word-count thermometer (Imp_pwc) paired with automated GPT-3.5 hallucination spam inside a closed 5-file sandbox. To say this paper introduces "the first novel paradigm" is frankly an embarrassment to everyone involved. ... Oh, you will be getting full, and I mean FULL, credit tomorrow on the Hobo **SEO** blog.😄 Tomorrow: 1. A Line-by-Line Forensic Audit of the KDD ’24 Paper: Deconstructing every single section, equation (G_qr -> SE -> S), and all 9 methods of Aggarwal et al. side-by-side against 25 years of SEO history (BM25, Panda, Hummingbird, RankBrain, BERT, Passage Indexing, and the 2024 Google Content Warehouse Leak). 2. A Mathematical Takedown of Imp_wc and Imp_pwc: Proving that their formulas are just post-hoc word-count scorecards (calibrated using Brian Dean's Backlinko SEO blog), exposing the 5 frozen variables required to hide LLM non-determinism, and proving the zero-sum Prisoner's Dilemma where Rank #1 loses 30.3% of its visibility. 3. Connecting Table 4 + Appendix C.1 + Section 9 into the "Spam" Verdict: Proving that when you strip out the traditional SEO that got the 5 pages into Google's Top 5, the only thing left in the KDD ’24 paper is automated GPT-3.5 hallucination spam (the fake chocolate factory and fake 70% robotics stat) tested inside a closed 5-file Perplexity upload sandbox. 4. Exposing a 2nd case of Academic Parasite SEO in Real Time (The 2026 WJARR Paper): Catching a 2026 "GEO-First Framework" paper red-handed scraping my August 2025 Hobo Strategic AiSEO ebook and Inference Optimisation while hallucinating that headless AI bots care about human screen-paint metrics (INP). 5. The 7-Source "Pure GEO" Isolation Test: Running Wikipedia, Coursera, Search Engine Land, Aspectus Group, Reply, and Semrush - those who currently rank on Google for GEO - through a strict isolation filter and proving that 0% of legitimate optimization tactics survive outside of SEO/AEO.