The New Frontier of Generative Engine Optimization
The landscape of search engine optimization is undergoing its most significant transformation since the advent of mobile-first indexing. As OpenAI’s ChatGPT continues to evolve from a simple chatbot into a sophisticated research tool, the focus for digital marketers has shifted toward “Generative Engine Optimization” (GEO). For years, the goal was simply to appear on the first page of Google. Now, the goal is to be cited as a primary source by the world’s leading AI.
A comprehensive study by AirOps, which analyzed 16,851 unique queries and over 50,000 generated responses, has shed light on exactly what it takes to earn a citation in ChatGPT. The findings challenge many long-held SEO beliefs, particularly the notion that “longer is always better.” Instead, the study reveals that ChatGPT prioritizes retrieval rank, heading precision, and content focus over the sheer volume of information.
The Power of Retrieval: Why Traditional SEO Still Matters
One of the most striking revelations of the AirOps study is that traditional search engine rankings remain the single most important factor for earning an AI citation. ChatGPT does not exist in a vacuum; it uses a process called Retrieval-Augmented Generation (RAG) to browse the live web, find relevant information, and synthesize an answer. If your content is not visible to the retrieval mechanism, it will never be cited.
According to the data, the page in the top search position was cited 58.4% of the time. This percentage drops significantly as you move down the search results. A page in position 10 has only a 14.2% chance of being cited. This suggests that while ChatGPT is “intelligent,” its initial selection of sources is heavily dependent on existing search engine algorithms. To win the AI citation game, you must first win the traditional SEO game.
The Retrieval Gap
The drop-off from position one to position ten highlights a “retrieval gap.” ChatGPT tends to favor the most authoritative and highly-ranked sources provided by its underlying search engine (primarily Bing). For brands, this means that the core pillars of SEO—backlinks, technical performance, and domain authority—are still the foundation upon which AI visibility is built. You cannot optimize for ChatGPT if you haven’t first optimized for the search engines that feed it.
Precision Over Breadth: The Decline of the “Ultimate Guide”
For the last decade, the “Ultimate Guide” has been the gold standard of content marketing. SEOs believed that by creating a 10,000-word skyscraper article that covered every possible facet of a topic, they could capture more keywords and provide more value. However, the AirOps study suggests that for ChatGPT, this approach may actually be counterproductive.
The data shows that focused pages—those that answer a specific query narrowly and directly—consistently outperformed broader, more comprehensive guides. When a user asks a specific question, ChatGPT looks for the most direct answer. A page that meanders through twenty different sub-topics before reaching the core answer creates “noise” that can interfere with the AI’s ability to extract the relevant data.
The Danger of Keyword Dilution
When a page attempts to be everything to everyone, its topical relevance becomes diluted. ChatGPT’s citation mechanism rewards precision. If a page is laser-focused on a single intent, it is much easier for the AI to verify that the content is a perfect match for the user’s request. This shift marks a move away from “comprehensive content” toward “specific content.”
Heading Relevance: The Strongest On-Page Signal
While retrieval rank is the strongest external signal, heading relevance is the most critical on-page factor identified in the study. Pages that used headings that closely mirrored the user’s query were cited 41.0% of the time. In contrast, pages with weaker or more creative heading matches saw citation rates hover around 30%.
This suggests that ChatGPT’s “browsing” behavior relies heavily on the document’s structure to navigate and understand its contents. If a user asks “How to calibrate a gaming monitor,” a page with an H2 titled “How to Calibrate a Gaming Monitor” is far more likely to be cited than a page that uses a more stylistic heading like “Getting the Most Out of Your Display’s Colors.”
Best Practices for AI-Ready Headings
To maximize your chances of being cited, your subheadings should be functional and descriptive rather than clever or evocative. They should serve as clear signposts for the AI. Use natural language that reflects the way users phrase questions. If you can anticipate the specific questions a user might ask, and use those questions as your H2 or H3 tags, you significantly increase your “citation-readiness.”
The Goldilocks Zone of Content Length
One of the most surprising findings of the AirOps report is the impact of word count on citations. There is a “Goldilocks zone” for content length: not too short, but certainly not too long. Pages between 500 and 2,000 words performed best in terms of earning citations.
Surprisingly, pages longer than 5,000 words were cited less often than pages with fewer than 500 words. This confirms that ChatGPT values efficiency. Long-form content often contains “fluff” or tangential information that increases the token count for the AI without adding proportional value. In the world of RAG, more tokens often mean more processing and a higher likelihood of the AI missing the most relevant nugget of information buried deep in the text.
Why 5,000+ Words Can Be a Liability
When ChatGPT crawls a page, it has a “context window”—a limit to how much information it can process at once. Very long pages may be truncated or summarized in a way that loses the specific details needed to answer a query. Furthermore, longer pages are more likely to cover multiple topics, which, as established, reduces the precision that ChatGPT rewards. If you have a topic that requires 5,000 words, it may be more effective to break it into three or four separate, highly-focused articles linked together, rather than one massive guide.
The Timing of Freshness: The 30 to 90-Day Window
Content freshness has always been a ranking factor for Google, but its role in ChatGPT citations is more nuanced. The study found that pages published between 30 and 89 days ago performed the best. Interestingly, content newer than 30 days actually performed worse.
This “lag time” suggests that new content needs a few weeks to build the necessary retrieval signals—such as being properly indexed by search engines, gaining initial traffic, or earning early social signals—before it becomes a “trusted” source for ChatGPT. It appears that being “first” is less important than being “established.”
The Longevity of Content
On the other end of the spectrum, content older than two years saw a decline in citation rates. This highlights the importance of content maintenance. To stay relevant in the AI era, you don’t necessarily need to write something brand new every day, but you do need to refresh your existing high-ranking content every 6 to 12 months. Updating facts, refreshing headings, and ensuring the information remains accurate can keep a page in that high-performance citation window.
Technical Structure: JSON-LD and Subheadings
Beyond the words themselves, the underlying technical structure of a page plays a supporting role in earning citations. The AirOps study noted that pages utilizing JSON-LD (JavaScript Object Notation for Linked Data) markup had a 38.5% citation rate, compared to 32.0% for pages without it.
While a 6.5% difference might seem modest, in a competitive digital landscape, it represents a significant edge. JSON-LD helps search engines and AI models understand the context of the data—whether it’s a recipe, a product review, or a how-to guide. By providing this structured data, you make it easier for the AI to “ingest” your information accurately.
The Optimal Number of Subheadings
The study also found that articles with 4 to 10 subheadings performed the best. This range provides enough structure to organize the information logically without over-fragmenting the content. Too few subheadings make the page a “wall of text” that is difficult for the AI to parse; too many subheadings can make the content feel disjointed and thin.
A Deep Dive into the Study’s Methodology
To understand the weight of these findings, it is important to look at the scale of the AirOps research. The study, titled “The Fan-Out Effect: What Happens Between a Query and a Citation,” was not based on a small sample size. Researchers scraped the ChatGPT interface directly—rather than using the API—to mirror the actual user experience.
The dataset was massive:
- 16,851 unique queries across 10 different verticals.
- Each query was run three times to account for AI variability, totaling 50,553 responses.
- Over 353,799 unique pages were analyzed in the retrieval set.
- 1.5 million “fan-out” detail rows were tracked to see how the AI moved from a search result to a final citation.
This level of data provides a statistically significant look at the “black box” of ChatGPT’s decision-making process. It moves the conversation from theory and guesswork to data-driven strategy.
Actionable Strategies for Earning More AI Citations
Based on the findings of this study, here are the steps content creators and SEO professionals should take to optimize for the AI-driven future:
1. Prioritize Core SEO for Top 3 Rankings
Since 58% of citations go to the #1 search result, you cannot afford to ignore traditional ranking factors. Focus on building high-quality backlinks, improving page load speeds, and ensuring mobile responsiveness. If you aren’t in the top three for a query, your chances of being cited by ChatGPT are slim.
2. Audit and Tighten Content Focus
Review your existing “Ultimate Guides.” Are they too broad? Consider breaking large pieces of content into smaller, more targeted articles that answer one specific question exceptionally well. Aim for the 500-2,000 word range for your most important topics.
3. Align Headings with Intent
Analyze the queries that lead users to your site. Are your H2 and H3 headings phrased in a way that matches those queries? Transition from “creative” headings to “functional” headings. Use the exact phrasing of the questions your audience is asking.
4. Implement and Validate Structured Data
Ensure that every page on your site uses appropriate JSON-LD markup. This is no longer just for Google’s “rich snippets”; it is a vital signal for AI models trying to categorize your content.
5. Refresh Content Every 6 to 12 Months
Don’t let your high-ranking content rot. Because citations drop off significantly after two years, a regular update schedule is essential. Refresh the data, update the publication date, and ensure the headings are still optimized for current search trends.
Conclusion: The Future of Content Discovery
The AirOps study makes one thing clear: ChatGPT is not replacing the web, but it is changing how we navigate it. The era of “gaming the system” with high word counts and keyword stuffing is over. The era of precision, technical clarity, and high-quality retrieval has arrived.
By focusing on winning the retrieval race, keeping content focused and concise, and utilizing clear structural signals like headings and JSON-LD, brands can ensure they aren’t just found by search engines—they are cited by the AI. As the “Fan-Out Effect” shows, the path from a user’s question to your website’s citation is narrow, but for those who optimize correctly, the rewards in terms of authority and traffic are substantial.