SEO solves “Google finds you.” GEO solves “ChatGPT recommends you.” These are two different problems.
Why GEO Matters
Have you noticed how search behavior is shifting?
Instead of typing keywords into Google, more and more people are going straight to ChatGPT, Perplexity, or Gemini: “What are the best gold investing platforms?”
AI search engines give direct answers — with citation sources. If your site gets cited, traffic flows in. If it doesn’t, you won’t even know what you missed.
That’s the problem GEO (Generative Engine Optimization) aims to solve: how to get AI to prioritize your content when generating answers.
Traditional SEO optimizes for Google’s ranking algorithm. GEO optimizes for LLM content preferences. Different rules, different playbook.
What the Research Says
Princeton KDD’24: Where GEO Began
The Princeton team published the first systematic GEO study at KDD 2024 and open-sourced GEO-BENCH, a benchmark dataset.
They tested 9 content optimization strategies. Key findings:
| Strategy | Impact | Notes |
|---|---|---|
| Add statistics | Most effective | Embed specific numbers, percentages, research data |
| Cite authoritative sources | Very effective | Reference known institutions, papers, official data |
| Use domain-specific terms | Effective | Precise terminology beats plain language for LLM preference |
| Confident tone | Effective | Avoid hedging (“maybe”, “perhaps”) — use assertive language |
| Structured content | Effective | Clear heading hierarchy, lists, tables |
| Simplify readability | Limited impact | Lowering reading difficulty doesn’t help GEO much |
| Improve fluency | Limited impact | Writing quality isn’t a major factor in LLM citation selection |
Key takeaway: Adding statistics + citing authoritative sources can boost AI citation probability by ~40%.
This is counterintuitive — many assume that writing in plain, accessible language matters most. But LLMs actually prioritize information density and credibility.
CMU ICLR’26: AutoGEO — Automatically Discovering AI Engine Preferences
The CMU team took it further. Their AutoGEO paper doesn’t hand-craft strategies — it lets AI automatically learn each search engine’s preference rules.
The approach:
- Feed an LLM a batch of content
- Observe which pieces get cited and which don’t
- Automatically extract a rule set
- Rewrite content using those rules
- Validate citation rate improvement
Result: 51% improvement over manual strategies, and it transfers across models — rules discovered on ChatGPT also work on Perplexity and Gemini.
Top rules AutoGEO discovered:
- Answer directly: Lead with the conclusion, skip the preamble
- Clear structure: H2/H3 + lists + tables — easy for LLMs to extract
- Specific data: Prices, percentages, ratios — the more precise, the better
- Stay focused: One idea per paragraph
- Take a clear stance: Don’t hedge both sides — commit to a position
AutoGEO has open-source code and supports API mode — no GPU needed. Feed it a rule set + source text, and the LLM rewrites automatically.
Practical Tools: How to Know If AI Is Recommending You
Academic papers tell you how to optimize content. But there’s a prerequisite question: how do you know if AI search engines are currently recommending you?
Traditional SEO has Google Search Console. GEO? No official tool yet. But there are open-source options:
gego — AI Search Engine Brand Monitoring
gego does something simple: it periodically sends preset prompts to ChatGPT, Perplexity, Gemini, and Claude, then checks if the responses mention your brand or URL.
What it tells you:
- Does ChatGPT mention you when asked “best gold investing platforms”?
- Does Perplexity cite your URL as a source?
- How much do responses differ across AI engines?
- Trends over time
Deployment cost: Docker one-click setup, API costs <$1/day.
This is real GEO detection — based on actual queries, not guesswork.
SerpBear — Google Rank Tracking
SerpBear is a self-hosted Google keyword rank tracker with 3k+ stars. While it tracks traditional search rankings, GEO and SEO aren’t mutually exclusive — content that ranks well on Google tends to get cited more by AI too (since AI training data and retrieval sources heavily draw from top-ranking pages).
Features:
- Track keyword ranking changes on Google
- Google Search Console integration
- Free scraping tier covers 100 keywords/day
- Self-hosted, runs on a $5/month VPS
SEOnaut — SEO Baseline Audit
SEOnaut is a Go-based SEO audit crawler that checks dead links, redirect chains, and heading structure.
No frills, but reliable. Limited scope though — doesn’t check schema markup, Core Web Vitals, or image optimization. Good for an occasional full-site scan, not something you need running 24/7.
GEO Optimization Checklist: What You Can Do Today
Based on both papers + practical tool feedback, here’s an actionable GEO checklist:
Content Level
1. Lead every article with a direct answer
Don’t:
Gold investing is a complex topic that requires considering multiple factors…
Do:
The two most compelling gold investment vehicles in 2026 are physical gold ETFs (e.g., GLD) at 0.4% annual fees and on-chain tokenized gold (e.g., PAXG) with 24/7 trading and self-custody support.
2. Embed specific data
Don’t: “Gold prices have risen significantly recently”
Do: “Gold broke through $3,100/oz in Q1 2026, up 18.7% year-over-year, hitting an all-time high”
3. Cite authoritative sources
Don’t: “According to reports…”
Do: “According to the World Gold Council’s Q1 2026 report…”
4. Use structured formats
LLMs extract information from source content when generating answers. Tables, lists, and comparison formats are easier to extract and cite than plain paragraphs.
5. Use domain-specific terminology
LLMs perform semantic matching. When a user asks about “RSI overbought signals,” content containing “RSI,” “overbought,” and “technical indicator” is more likely to get matched.
Monitoring Level
6. Deploy gego to track AI citations
Set up 5-10 core prompts covering your target keywords. Check trends weekly.
7. Track Google rankings in parallel
Use SerpBear to monitor 50-100 target keywords. Google rankings and AI citations are positively correlated — ranking improvements usually mean higher AI citation probability too.
Process Level
8. Run through a GEO checklist before publishing
- Does the opening contain a direct conclusion?
- Are there 3+ specific data points?
- Are authoritative sources cited?
- Are tables or lists used?
- Is domain terminology accurate?
9. Consider AutoGEO API for automated optimization
If you publish at volume (e.g., daily), integrate AutoGEO’s rule set into your pipeline to auto-optimize key paragraphs before publishing.
SEO vs GEO: Not Replacement — Compounding
A common misconception: GEO will replace SEO. It won’t.
AI search engines currently account for less than 10% of total search traffic. Google is still the dominant force. But the trend is clear — AI search share is growing fast, especially for information-dense queries (“what’s the best X,” “how does A compare to B”) where users are already migrating.
The right strategy: SEO as the foundation, GEO as the multiplier.
The good news is that many GEO tactics (adding data, citing sources, structuring content) also benefit SEO. This isn’t either/or — doing GEO well naturally improves your SEO too.
Final Thoughts
GEO is still in its very early stages. The first academic paper (Princeton KDD’24) was published less than two years ago, and practical tools are just starting to emerge.
But that’s exactly the opportunity.
Most content creators are still operating with a 2020 SEO mindset — stuffing keywords, writing long-form, chasing Google rankings. AI search engines have a completely different selection logic: they don’t care about your domain authority or backlink count. They care whether your content directly answers the user’s question, whether the information is credible, and whether the data is specific enough.
Starting GEO now means claiming your position before most people realize the game has changed.
References:
- Princeton KDD’24: GEO: Generative Engine Optimization
- CMU ICLR’26: AutoGEO — Automatic Generative Engine Optimization
- Monitoring tools: gego | SerpBear | SEOnaut
