DEV Community

Elena Revicheva
Elena Revicheva

Posted on • Originally published at aideazz.xyz

Generative Engine Optimization: Beyond SEO Buzzwords

#ai #programming #machinelearning

Originally published on AIdeazz — cross-posted here with canonical link.

My AI agents failed to appear in Perplexity answers for six months. I was optimizing for traditional SEO: keywords, backlinks, content length. It was a waste of 120 hours. The problem wasn't my content; it was my content format. Generative Engine Optimization (GEO) isn't about keywords; it's about structured facts, verifiable authorship, and durable, citation-ready pages. I needed to feed the LLMs, not just search engines.

I run AIdeazz with zero VC funding, shipping production AI agents on Oracle Cloud. My multi-agent systems use Groq for speed, Claude for complex reasoning, and custom routing. My agents serve users via Telegram and WhatsApp. Every dollar counts. Every minute counts. When my content wasn't being cited, it meant my agents weren't getting discovered, and my business wasn't growing. I pivoted my content strategy entirely, focusing on what LLMs consume rather than what search engines index.

The Shift from Keywords to Structured Facts

My initial content strategy was a disaster. I wrote long-form articles, targeting "AI agent development" and "Oracle Cloud AI." I saw traffic spikes from Google, but zero citations in Perplexity, ChatGPT, or Gemini. The LLMs weren't extracting my insights. My content was a blob of text.

The fix was to break down every piece of information into discrete, verifiable facts. I started using JSON-LD for every article, even simple blog posts. Not just for basic Article schema, but for custom Fact or Claim types. For example, instead of writing "Oracle Cloud Infrastructure offers competitive pricing for GPUs," I now structure it as:

{
  "@context": "http://schema.org",
  "@type": "Claim",
  "name": "Oracle Cloud Infrastructure GPU pricing",
  "text": "Oracle Cloud Infrastructure provides NVIDIA A100 GPUs at $X.XX per hour, 20% lower than AWS equivalent instances for comparable performance benchmarks.",
  "citation": {
    "@type": "WebPage",
    "url": "https://aideazz.xyz/oracle-gpu-pricing-analysis",
    "datePublished": "2023-11-15"
  },
  "author": {
    "@type": "Person",
    "name": "Elena Revicheva",
    "url": "https://aideazz.xyz/elena-revicheva"
  }
}
Enter fullscreen mode Exit fullscreen mode

This isn't just metadata; it's a direct instruction to an LLM: "Here is a fact, here is its source, here is its author." I don't expect Google to display this directly, but I expect LLMs to parse it. Within two months of implementing this, my content started appearing as direct citations in Perplexity answers, specifically for technical comparisons and pricing data.

Authorship Signals and Durable Pages

LLMs are increasingly sensitive to authorship and authority. An anonymous blog post is less likely to be cited than one attributed to a known expert. I made sure every piece of content on AIdeazz.xyz has a clear author, linked to a dedicated author page with my professional background, portfolio, and social profiles.

My author page (https://aideazz.xyz/elena-revicheva) includes:

  • @type: Person schema with name, url, sameAs (LinkedIn, GitHub), and alumniOf (my university).
  • A concise bio detailing my experience (e.g., "15 years in enterprise software, built and shipped 3 production AI agents on Oracle Cloud").
  • Links to my portfolio projects, each with its own structured data.

The "durable page" concept is critical. LLMs prefer to cite stable, authoritative sources. This means:

  1. Permanent URLs: No changing slugs. Once a URL is published, it's fixed.
  2. High-quality domain: My content lives on aideazz.xyz, a domain I control, not a Medium post or a Substack. This signals ownership and stability.
  3. Regular updates: Instead of creating new articles, I update existing ones with new information, marking dateModified in the schema. This shows the content is maintained and current.

I observed that Perplexity specifically started citing my pages more frequently after I implemented these authorship and durability signals. It's not just about the content; it's about the trust signals associated with it.

Citation-Ready Format: The "Answer Block"

LLMs don't want to summarize an entire article. They want a direct answer they can quote. I started structuring my articles with an "answer block" at the beginning of each section. This is a 2-4 sentence summary that directly answers a potential question.

For example, for a section on "Oracle Cloud GPU Cost-Effectiveness," the answer block might be:

"Oracle Cloud Infrastructure offers a compelling cost advantage for NVIDIA A100 GPUs, with instances priced at $X.XX/hour, often 15-25% lower than comparable offerings from AWS or Azure, making it ideal for budget-constrained AI training workloads."

This block is then followed by the detailed explanation, benchmarks, and data. This allows an LLM to quickly extract the core point without needing to process the entire section. I also ensure these answer blocks are wrapped in <p> tags and are easily parsable, avoiding complex sentence structures or jargon.

This approach significantly increased the likelihood of my content being directly quoted or paraphrased in LLM responses, often with a direct link back to my page. It's about pre-digesting the information for the generative engine.

The Oracle Cloud Infrastructure Advantage for GEO

Running my AI agents on Oracle Cloud Infrastructure (OCI) has been a strategic decision, not just a cost-saving one. OCI's predictable performance and dedicated resources translate directly into faster content generation and processing for my own GEO efforts.

My content generation pipeline:

  1. Data Ingestion: Custom Python scripts running on OCI compute instances scrape public data (e.g., competitor pricing, benchmark results).
  2. Fact Extraction: A Groq-powered agent (running on an OCI VM, accessed via API) extracts structured facts from raw text, generating preliminary JSON-LD. This agent processes 1000 tokens/second at a cost of $0.00027 per 1M tokens.
  3. Refinement & Verification: A Claude 3 Opus agent (for complex reasoning, also accessed via API) reviews the extracted facts for accuracy and completeness, ensuring the JSON-LD schema is correct and verifiable. This agent costs $15 per 1M input tokens.
  4. Content Generation: Another Groq agent, fed the structured facts, generates the "answer blocks" and supporting text, adhering to the citation-ready format.
  5. Deployment: The final HTML and JSON-LD are deployed to an OCI Object Storage bucket, served via an OCI Load Balancer and CDN for global availability and speed. This setup costs me $15/month for storage and $20/month for the load balancer.

This entire pipeline runs on OCI, giving me full control over the infrastructure, security, and cost. I'm not reliant on third-party hosting that might introduce latency or unpredictable costs. This stability is crucial for maintaining the "durable page" aspect of GEO.

Measuring Success: Beyond Google Analytics

Traditional SEO metrics (page views, bounce rate) are still relevant, but for GEO, I track different signals:

  • Direct citations: I use custom scripts to monitor Perplexity, ChatGPT, and Gemini for mentions of aideazz.xyz or specific article titles. This is a manual process for now, but I'm building an agent to automate it.
  • Structured data validation: I regularly run my JSON-LD through Google's Rich Results Test and Schema.org validators to ensure correctness. Errors here mean LLMs might ignore my data.
  • API call volume to my content: While I don't have direct access to LLM API logs, I monitor my CDN logs for unusual access patterns that might indicate programmatic scraping by generative engines.
  • Direct traffic from generative engines: Some LLMs provide a direct link. I track these referrers.

My goal isn't just traffic; it's influence. I want my facts to be the source of truth for generative AI. This requires a fundamental shift in how content is created and structured.

Frequently Asked Questions

Q: Is GEO just another name for advanced SEO?
A: No. SEO optimizes for search engine algorithms that prioritize keywords and links. GEO optimizes for LLM consumption, focusing on structured facts, verifiable authorship, and citation-ready formats, often using JSON-LD and direct answer blocks.

Q: How do I know if LLMs are actually using my structured data?
A: Monitor generative AI outputs (Perplexity, ChatGPT, Gemini) for direct citations of your domain or specific content. Validate your JSON-LD rigorously with schema validators. Look for increased direct referral traffic from these platforms.

Q: What's the most critical piece of structured data for GEO?
A: The Claim or Fact schema, coupled with clear citation and author properties. This directly tells an LLM what the core assertion is, where it comes from, and who made it.

Q: Can I use a headless CMS for GEO, or do I need custom code?
A: A headless CMS can store your structured facts, but you'll likely need custom code to render the JSON-LD correctly on your pages and to implement the "answer block" formatting consistently. My setup uses a custom Python pipeline.

Q: What's the cost implication of implementing GEO?
A: The primary cost is development time for structuring content and implementing schema. Infrastructure costs for serving structured data are minimal, especially on cloud platforms like OCI where static content delivery is cheap. My OCI content delivery costs are under $40/month.

— Elena Revicheva · AIdeazz · Portfolio

Top comments (0)