AI search is here. Tools like Bing Chat, Perplexity.ai, and ChatGPT are changing the way people find information and in this new-world order, visibility isn’t about being listed. I’s about being cited. A groundbreaking paper from KDD 2024 titled “GEO: Generative Engine Optimization marks the first serious attempt to help content creators adapt to this shift.
This deep dive explores the paper’s insights, focusing on what’s new, what’s measurable, and what works when optimizing content for generative engines.
- Why GEO Matters Now
- Formalizing Generative Engines (GE's)
- Redefining Visibility: The Metrics
- GEO-Bench: A New Standard for Testing
- Nine Optimization Strategies: What Works (and What Doesn’t)
- Real-World Results: Testing on Perplexity.ai
- Domain-Specific Insights
- Limitations and Future Directions
- What This Means for Creators and Marketers
- Final Thoughts
Why GEO Matters Now
Traditional SEO is built around search engine results pages (SERPs). But generative engines (GE’s) don’t return lists of links. They generate synthesized, cited answers using large language models. As GE’s become the default way people find information, creators risk losing traffic, attribution, and control over their content.
The GEO framework offers a response. It proposes new ways to structure, style, and enhance content so that it is more likely to be included and cited in these AI-generated responses. This shift is more than cosmetic. It is foundational.
Formalizing Generative Engines (GE’s)
The paper defines GE’s as a hybrid of search and generation: they retrieve relevant sources and then synthesize them into a single natural language response. Each citation is embedded directly into the answer, making it harder for creators to understand or influence their presence.
To tackle this, the authors introduce GEO: a black-box optimization framework. Rather than reverse-engineering the engine, GEO focuses on improving the content itself through stylistic and structural enhancements aimed at boosting visibility in GE responses.
Redefining Visibility: The Metrics
One of the paper’s major contributions is its proposed visibility metrics, replacing traditional “rank” with more nuanced measurements:
1. Position-Adjusted Word Count: This measures how many words from a given source appear in the GE’s response, weighted by their position. Sentences cited earlier in the response count more, reflecting real user attention patterns.
2. Subjective Impression Metrics: These use a framework called G-Eval to evaluate –
- Relevance to the query
- Influence on the response
- Uniqueness of content
- Likelihood of user engagement
Together, these metrics allow creators to quantify and benchmark their visibility, something that’s been almost impossible with GEs until now.
GEO-Bench: A New Standard for Testing
To ensure broad applicability, the researchers built a dataset of 10,000 queries across 25+ domains ranging from law and government to history, science, and business. This so called GEO-bench is the first benchmark explicitly designed to test content performance in generative engines.
This benchmark allows optimization strategies to be tested in a controlled way, simulating how content appears across a wide range of GE responses.
Nine Optimization Strategies: What Works (and What Doesn’t)
The team applied nine different optimization methods to source content and tested their performance using GEO-bench. Here’s what they found:
| Method | Performance Lift |
|---|---|
| Quotation Addition | +38% |
| Statistics Addition | +41% |
| Cite Sources | +34% |
| Fluency Optimization | +22% |
| Technical Terms | +21% |
| Authoritative Style | +18% |
| Easy-to-Understand | +16% |
| Unique Words | +14% |
| Keyword Stuffing | –2% |
Key Takeaways:
- Quotations and statistics significantly improve citation rates
- Keyword stuffing, a common SEO tactic, actually reduced visibility
- Fluent, clear writing helps, as does using domain-appropriate technical language
Real-World Results: Testing on Perplexity.ai
The researchers tested their methods on Perplexity.ai, a live generative engine. The best-performing strategies showed up to 37% improvement in subjective visibility metrics. These tests confirm that GEO isn’t just theory. It works in production settings too.
Domain-Specific Insights
Optimization isn’t a one-size-fits-all affair either. The paper finds certain strategies work better in specific domains:
- Quotations are most effective in history and people & society queries
- Statistics perform best in law, government, and science
- Authoritative tone helps with debate and opinion-based content
Content teams can use these insights to tailor strategies to their niche, boosting the effectiveness of GEO across different verticals.
Limitations and Future Directions
The paper openly discusses GEO’s limitations:
- GE’s are black boxes and as such strategies may need constant testing and recalibration.
- GEO doesn’t yet handle multimodal content (e.g., images, video).
- Ethical concerns remain: poorly designed optimization could amplify misleading or biased content.
Still, the framework gives us a real starting point. With tools like GEO-bench and clear metrics, optimization is no longer guesswork but an evolving science.
What This Means for Creators and Marketers
If your brand relies on being seen, whether through organic search or thought leadership, GEO is a game-changer. Traditional SEO won’t disappear, but it’s no longer enough. Understanding how generative engines work, and how to shape content for them, is essential.
Instead of just ranking, the new challenge is citation visibility. The GEO paper offers:
- A new way to measure success in AI-driven search.
- Tested methods to improve your content’s inclusion in responses.
- Insight into how domain-specific tactics can amplify results.
Final Thoughts
The GEO paper isn’t just a research exercise. It’s a wake-up call. Generative engines are already changing how people discover and consume content. Without adaptation, creators risk becoming invisible in a world where visibility is defined not by position on a page, but by presence in a paragraph.
For marketers, editors, and digital strategists, GEO offers the first real roadmap into this new terrain. Whether you’re preparing content for a specific LLM or exploring AI-powered search tools, this paper is essential reading.
Want to optimize your content for the future of search? Start by aligning your editorial strategy with the methods proven to work in GEO and be ready to test, adapt, and evolve as generative engines continue to shape the digital landscape.
Need help making your content GEO-ready? Let’s talk.