“Which employer branding service provider is most recommended?”
Not you, unless you know what language models actually do when they compile an answer. Most organisations don't know that. Researchers from Princeton University did, and their findings are now the scientific foundation for every serious GEO strategy.
| DEFINITION Generative Engine Optimization (GEO) is the strategic practice of optimising content and brand authority for visibility in AI-generated answers from language models such as ChatGPT, Perplexity, Gemini, and Google AI Overviews. GEO focuses not on search result rankings, but on direct citation within AI-generated responses, the new frontier of information retrieval. |
Princeton, Georgia Tech and the Allen Institute published the first rigorous GEO study
In 2023, researchers from Princeton University, Georgia Tech, and the Allen Institute for AI published the first peer-reviewed research on Generative Engine Optimisation, later presented at ACM SIGKDD 2024. The main finding is unambiguous: brands that optimise their content for generative search engines increase their visibility in AI answers by up to 40%.
The 40% is not a marketing claim. It is a measured, replicated result, now cited in hundreds of follow-up studies and validated by the introduction of GEO functionality at Semrush, Ahrefs, and HubSpot.
A supplementary research paper by Chen et al. (arXiv, September 2025) refined the Princeton work: AI search engines differ significantly in domain diversity, data freshness, cross-lingual stability, and sensitivity to phrasing. The conclusion is strategically far-reaching: a single universal GEO strategy will not work. Platform-specific optimisation is a prerequisite for results.
The Princeton study identified nine demonstrably effective GEO tactics.
The original research isolated nine core tactics with a measurable impact on AI citation opportunities. These tactics are not theory; they are measured effects on controlled datasets.
| Tactics | Visibility impact | Mechanism |
|---|---|---|
| Cite external sources | High | Increases perceived reliability |
| Add statistics and data | High | Language models prefer quantifiable claims |
| Use direct expert quotes | High | Transfer of authority through attribution |
| Make content scannable | Medium-High | Headers and tables increase the chance of extraction. |
| Answer questions in the first sentence | High | Answer structure = citable |
| Define concepts clearly | Average | Definition pages are frequently cited |
| Build internal consistency | Average | Consistent perspective increases model confidence |
| Mention geographical context | Average | Location relevance increases citations for local queries |
| Optimise per platform | High | Each language model weighs signals differently |
The common denominator of all nine tactics: write for the judgment of a language model, not for the gaze of a human reader.
Each AI platform uses different citation logic, so one GEO strategy is not sufficient
ChatGPT, Perplexity, Gemini and Google AI Overviews use fundamentally different ranking mechanisms to select citations. The research by Chen et al. (September 2025) systematically documented these differences:
- ChatGPT prefers content with high external link authority Consistent entity recognition across platforms.
- Perplexity weighs current events more heavily: Vers content and recently published sources score significantly better than older pages with higher authority.
- Gemini is meer gevoelig voor Google's E-E-A-T signals en rewards structured data and author authority more visibly than other platforms.
- Google AI Overviews combines traditional SEO ranking with a separate citation matrix that does not correlate one-to-one with classic search ranking.
“AI Search services differ significantly from each other in their domain diversity, freshness, cross-language stability, and sensitivity to phrasing.”
Chen et al., arXiv, September 2025
Platform-agnostic GEO is an illusion. Effective GEO strategy requires platform-specific optimisation.
Language models show a structural preference for large brands, but niche players can bypass that bias
The study identifies an uncomfortable truth: language models exhibit a structural big brand bias. Large, well-known brands are cited more often than smaller players, even when the content of those smaller players is of higher quality. This bias can be overcome, provided you apply the four counter-strategies formulated in the research.
1. Build domain-specific authority
Niche domination triumphs over generic competition with big brands. Become the undisputed expert in one specific subject. Language models cite niche experts when the question is specific enough, as the broader authority of big brands is less relevant on narrow topics.
2. Earn media in trusted sources
Earned media from trusted sources builds authority that owned content cannot deliver. Appear in publications that language models recognise as authoritative: trade journals, research reports, industry associations, and certified data sources. Every mention in a trusted source strengthens your entity profile.
3. Formulate quotable sound bites
Quoteable sound bites are sentences, definitions, and statistics formulated so sharply that language models pick them up almost automatically. Characteristics: short (under 25 words), precise, unique, and declarative. A quoteable sound bite is a complete statement that can stand on its own.
4. Be consistent across all platforms
Consistency generates entity confidence, and entity confidence predicts citation likelihood. When your description on LinkedIn, your website, Crunchbase, and Trustpilot is inconsistent, the language model doesn't know which version is reliable. The result: fewer citations, despite strong individual signals.
GEO will have shifted from an experimental niche to enterprise mainstream by 2025.
Three market signals confirm that GEO has transitioned from an emerging discipline to a mainstream standard:
- Ahrefs launched a Brand Radar in March 2025. The language model visibility tracks over ChatGPT, Google AI Overviews, Gemini, Perplexity and Copilot, powered by a database of over 100 million real prompts.
- Semrush added an AI Visibility Score to her Enterprise platform, as a standard metric alongside classic SEO rankings.
- Analysts predict a 50%% drop in traditional search volume before 2028, with language model traffic surpassing Google traffic before the end of 2027.
Organisations starting with GEO now are building a lead that will be exponentially harder to catch up on as AI search volume increases.
AI Rebels meet GEO-visibility across five proprietary dimensions
AI Rebels developed the EAR model to make GEO-visibility fully measurable for Dutch organisations. The model measures five independent dimensions, each scored on a scale of 0 to 100. The sum forms the EAR Score, the most complete benchmark for AI-visibility in the Dutch market.
| Dimension | What we measure |
|---|---|
| Braking power | External authority and citation in sources that trust language models |
| Narrative Coherence | Consistency of your brand message across all external platforms |
| Signal strength | Technical visibility signals: structured data, schema, LLMs.txt |
| Competitive position | How you score against direct competitors in AI replies |
| Responsiveness | Speed at which you respond to changes in AI citation patterns |
The 90-day GEO sprint based on scientific research
This sprint translates the Princeton research and the Chen et al. refinement paper into an actionable 90-day implementation plan.
Weeks 1 and 2 – Baseline and audit
- Perform a platform-specific visibility test: 10 industry-relevant prompts per platform (ChatGPT, Perplexity, Gemini, Google AI Overviews).
- Document how your organisation is currently described and whether it is described at all.
Month 1 – Laying the Foundations
- Implement the five highest-impact Princeton tactics on your core pages: direct answers, statistics, expert quotes, scannable structure, definition pages.
- Correct entity inconsistencies between LinkedIn, website, Crunchbase and trade directories.
Month 2 - Building Authority
- Publish at least two pieces in external professional publications that recognise language models as authoritative.
- Launch an original data point or piece of research that you can claim as your own source.
Month 3 – Platform-specific optimisation
- Refine per platform: which adjustments demonstrably work on ChatGPT versus Perplexity?
- Set up structural monitoring with a fixed prompt set of 20 to 50 questions, tested monthly.
Request an EAR Intelligence Audit to know your GEO starting score
AI Rebels is the Dutch GEO-agency that conducts audits across all five EAR dimensions. No theory, no buzzwords. A measured score, a priority-based plan, and a team that executes it for you.
Begin uw EAR Intelligence Audit via ai-rebels.nl
Sources & References
[1] Aggarwal S., Meerza S.I., Guo S., Zhong V. et al., ‘GEO: Generative Engine Optimisation’. Princeton University / Georgia Tech / Allen Institute for AI. Published: ACM SIGKDD 2024. arxiv.org/abs/2311.09735
[2] Chen M. et al., ‘Generative Engine Optimisation: How to Dominate AI Search’. arXiv, 10 September 2025. arxiv.org/abs/2509.08919
[3] GitHub / amplifying-ai, ‘Awesome Generative Engine Optimisation’. February 2026.
[4] Semrush, ‘Generative Engine Optimization (GEO): How to Win AI Mentions’. Search Engine Land, February 2026.