In the past 18 months, I have driven over 1,100 Google AI Overview citations, 114 ChatGPT citations, and 29 Perplexity citations for a single client in the competitive health vertical. These are not estimates or projections — they are verified counts from live tracking tools. And the methodology behind them is what I want to break down in this article, because most of what you read online about GEO and AI optimization is either outdated or flat-out wrong.
GEO — Generative Engine Optimization — is not a separate discipline from SEO. It is the next evolution of the same game we have been playing for two decades. The search engines that matter (Google, Bing, and now ChatGPT, Perplexity, and Gemini) all rely on the same underlying signals to determine which sources to cite. If you understand what those signals are, you can engineer citation visibility the same way you engineer ranking visibility.
What Google AI Overviews actually pull from
Google’s AI Overviews do not generate information from thin air. They synthesize answers from pages that already rank well for the query, then cite the sources they pulled from. This means the foundation of AI Overview visibility is still traditional SEO — if you do not rank on page 1 for a query, you are extremely unlikely to be cited in the AI Overview for that query.
But ranking on page 1 is necessary, not sufficient. I have tracked hundreds of queries where a client ranks in positions 1-3 but is not cited in the AI Overview, and other queries where a position 5-7 page gets cited instead. The differentiator, consistently, is how parsable the content is for extraction.
Google’s AI needs to be able to extract a clean, concise answer from your page. This means your content structure matters enormously: clear H2/H3 headers that match query intent, concise paragraphs that directly answer specific questions, and structured data that explicitly defines the entities and relationships on the page.
The structured data layer that most SEOs miss
JSON-LD schema is the single highest-leverage tactic for AI citation visibility. Specifically, three schema types have consistently correlated with AI Overview citations in my testing: Article schema with proper author and datePublished markup, FAQPage schema for question-answer content, and Speakable schema that explicitly tells Google which sections of your page are suitable for voice and AI extraction.
Speakable schema is particularly underutilized. Most SEOs have never implemented it because it was originally designed for Google Assistant voice responses. But in practice, it signals to Google which content blocks on your page are optimized for extraction — and that signal applies to AI Overviews too. For the telehealth client where I achieved 1,100+ AI citations, every key content page had Speakable schema pointing to the primary answer paragraphs.
The entity-based content architecture that underpins this approach is what makes it scalable. When your structured data accurately represents the relationships between your content entities, AI systems can map your content to their knowledge graphs more effectively.
ChatGPT and Perplexity: different engines, same principles
ChatGPT with browsing enabled and Perplexity both pull from live web results, but they weight different signals. ChatGPT tends to favor sources with strong domain authority and clear, authoritative content. Perplexity pulls more aggressively from recent, well-structured content and tends to favor pages with explicit citations and data.
For the telehealth client, achieving 114 ChatGPT citations and 29 Perplexity citations required a slightly different content approach than the AI Overview strategy. ChatGPT citations came primarily from long-form, authoritative content that positioned the client as the definitive source on specific health topics. Perplexity citations came from content that included specific data points, statistics, and cited research — the kind of content that Perplexity’s retrieval system identifies as high-credibility.
The common thread is entity clarity. Every page that gets cited across multiple AI platforms has one thing in common: it is unambiguous about what entity it represents, what question it answers, and what evidence supports its claims. This is E-E-A-T translated into machine-readable signals.
The YMYL factor
The telehealth client operates in a YMYL (Your Money or Your Life) vertical — health content where wrong information could harm users. Google applies significantly higher quality standards to YMYL content, and this extends to AI Overviews. Achieving AI citations for health queries required not just technical optimization but genuine expertise signals: author credentials, citations to peer-reviewed research, and content reviewed by medical professionals.
This is the part that cannot be faked. Google’s quality rater guidelines explicitly evaluate whether YMYL content is created by people with relevant expertise and first-hand experience. The structured data signals, the content quality, and the author credentials all need to be genuine. If they are, the AI citation visibility follows naturally from the same E-E-A-T foundations that drive traditional ranking.
Practical implementation
If you want to start engineering AI citation visibility for your brand, the priority order is clear. First, ensure you rank on page 1 for your target queries — AI Overviews cite ranking pages. Second, implement comprehensive JSON-LD schema including Article, Organization, and Speakable markup. Third, restructure your content so that each page clearly answers specific questions with concise, extractable paragraphs under descriptive H2 headers. Fourth, build entity authority through GEO-specific optimization that makes your content the most parsable, most authoritative source for your topic.
The AI search landscape is evolving rapidly. But the fundamental principle has not changed since the earliest days of SEO: be the most useful, most authoritative, most trustworthy source of information for your topic, and make sure the machines can understand that you are. The technology changes. The strategy does not.